EQUITY RESEARCH | April 16, 2026 | 5:54PM EDT
The Goldman Sachs Group, Inc.
Goldman Sachs does and seeks to do business with companies covered in its research reports. As a result, investors should
be aware that the firm may have a conflict of interest that could affect the objectivity of this report. Investors should consider
this report as only a single factor in making their investment decision. For Reg AC certification and other important disclosures,
see the Disclosure Appendix, or go to Analysts employed by non-US affiliates are not
registered/qualified as research analysts with FINRA in the .
Allen Chang
+852 2978-2930
@
Goldman Sachs (Asia) .
Eric Sheridan
+1(917)343-8683
@
Goldman Sachs & Co. LLC
Daniela Costa
+44(20)7774-8354
@
Goldman Sachs International
Mark Delaney
+1 212 357-0535
@
Goldman Sachs & Co. LLC
Christian Frenes
+44(20)7051-8641
@
Goldman Sachs International
Global AVs
Analyzing the impact of AI on profit pools - Part II
A Transportation Case Study
Datacenter capex from leading public hyperscalers is now approaching ~$700 bn USD, roughly
10X the level in 2020. With spend continuing to grow, and consensus capex estimates being revised
higher yet again with the most recent set of earnings reports, we believe the return on that capex
remains a key focus for the financial markets and brings us to Part II in our series on analyzing the
impact of AI on profit pools. Recall this series seeks to determine how much profit AI enabled efforts
could help drive with different case studies, giving investors a tool to better contextualize investment
levels. Please see our June 2025 report on the advertising market for Part I of this report series. In
Part II - A Transportation Case Study - we examine autonomous mobility globally, including for light
and commercial vehicles.
See Inside for full list of Authors
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
Do Hyoung Kim
+82(2)3788-1376
@
Goldman Sachs (Asia) ., Seoul
Branch
Nick Zheng, CFA
+852-2978-1405
@
Goldman Sachs (Asia) .
Chandramouli Muthiah
+91(22)6616-9344
@
Goldman Sachs India SPL
Daniela Costa
+44(20)7774-8354
@
Goldman Sachs International
Eric Sheridan
+1(917)343-8683
@
Goldman Sachs & Co. LLC
Authors
Christian Frenes
+44(20)7051-8641
@
Goldman Sachs International
Ronald Keung, CFA
+852-2978-0856
@
Goldman Sachs (Asia) .
Kota Yuzawa
+81(3)4587-9863
@
Goldman Sachs Japan Co., Ltd.
Lincoln Kong, CFA
+852-2978-6603
@
Goldman Sachs (Asia) .
Verena Jeng
+852-2978-1681
@
Goldman Sachs (Asia) .
Tina Hou
+86(21)2401-8694
@
Goldman Sachs (China) Securities
Company Limited
Will Bryant
+1(212)934-4705
@
Goldman Sachs & Co. LLC
Mark Delaney, CFA
+1(212)357-0535
@
Goldman Sachs & Co. LLC
Allen Chang
+852-2978-2930
@
Goldman Sachs (Asia) .
Aman Gupta
+1(212)357-1549
@
Goldman Sachs & Co. LLC
Julia Fein-Ashley
+1(212)902-5070
-ashley@
Goldman Sachs & Co. LLC
Monika Mengting Liu, CFA
+44(20)7051-7601
@
Goldman Sachs International
Meihan Yang
+44(20)7051-6601
@
Goldman Sachs International
Xuan Zhang
+852-2978-1478
@
Goldman Sachs (Asia) .
Ayush Ghose
+1(212)902-7257
@
Goldman Sachs & Co. LLC
Emma Huang
+1(212)902-7229
@
Goldman Sachs & Co. LLC
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
The ecosystem of autonomy 5
PM summary 6
AI tech capabilities and capital - gauging costs and growth 10
Robotaxi deployments and market size accelerating, aided by safety and consumer adoption 13
Consumer AVs - partially and fully autonomous vehicles poised to grow 23
AV trucking market is currently small but now poised to ramp 26
What is at risk of disruption - human driven rideshare/trucking, and auto sales 30
Analyzing the Potential Impact of Robotaxis on Uber & Lyft 32
Spotlight on Waymo: Framing Operating Assumptions and the Path to $8 bn+ Gross Bookings by 2030E 38
Spotlight on Tesla 45
Stocks in Focus 48
Disclosure Appendix 57
16 April 2026 2
Goldman Sachs Global AVs
Table of Contents
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
The pace of autonomous technology commercialization has accelerated since our
June 2025 report, “Framing profit pools in rideshare, trucking, and insurance as AVs
move from concept to commercialization”. This is being driven by both a growing
number of deployments in the US and China, and also international expansions into new
geographies including EMEA. These deployments are enabled by both captive
technology development (. at Waymo, Tesla, Pony AI, etc) and a growing set of
merchant Physical AI tools including from companies such as Nvidia (. Alpamayo). In
fact, Nvidia discussed numerous Physical AI and AV partnerships at GTC 2026 including
with Mercedes, Uber, GM, BYD, Geely, and Hyundai. We expect there to ultimately be
multiple providers of autonomous technology.
With the now more global ramps of several companies, we update our AV profit pools
analysis to include key global regions for robotaxis and Class 8 trucks, and we also frame
the market potential from “robots that drive” more broadly including delivery bots. In
addition, we update our estimates for the mix of partially and fully autonomous
consumer vehicles (. L3-L5 classifications).
We now estimate that the AV robotaxi market in the US will reach $19 bn in 2030, up
from $7 bn prior, and we introduce our 2035 forecast of $48 bn. When including
global markets, we estimate that the robotaxi market in 2035 could be ~$415 bn.
We assume that the gross margins of a vertically integrated operator could be 30%-50%,
implying gross profit for the market of ~$150 bn in 2035 and cumulative gross profit of
~$440 bn over the next decade for robotaxis.
Class 8 trucking is also growing, and our forecast remains similar to our prior outlook
for the market by 2030 in the US (we assume growing to $16 bn in 2030 from low levels
currently), and we think the market could reach $105 bn in the US and ~$560 bn
globally in 2035. We estimate that gross profit for AV trucking will be ~$135 bn in 2035,
and about $300 bn cumulatively over the next decade.
Considering the market more broadly for AV hardware and software & services, we
estimate that revenue associated with the industry could reach approximately $2
trillion in 2035 comprised of hardware sales (. robotaxis, consumer L3-L5 vehicles,
AV trucks, and delivery bots) and software/digital services (. robotaxi and AV trucking
services, and consumer autonomous subscriptions for L3-L5). Recognizing that a
meaningful portion of this $2 trillion of revenue associated with autonomy is
coming from business for goods/services that isn’t incremental (. buying an L3/L4
car rather than an L1/L2 car, and freight being hauled by an AV truck rather than a
person), when focusing more specifically on the piece of this market attributable to
AI (. revenue for virtual driver technology in robotaxis and AV trucks, and software
subscriptions for consumer L3-L5 autonomy), we estimate that the market will be
approximately $300 bn in 2035.
Our global forecast is informed by the views of the regional teams and company
specific outlooks for key covered companies, including views on the China market and
AV providers from Allen Chang and team detailed in their report (see link).
While a portion of AV volumes will likely be incremental demand (. as new use
cases become affordable or possible), we believe autonomy could also disrupt
existing markets in the long-term. We frame risks and scenarios including to
human-operated rideshare/taxis, trucking, and light vehicle unit sales in the United
16 April 2026 3
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
States in this report. We estimate the economic size that could potentially be
disrupted in the US is ~$440 bn.
There are already signs of change, with Waymo’s share in SF reaching 30%, 20
months after being fully launched (per Yipit), and our base case view is for 5%
cannibalization of UCAN rideshare gross bookings from AVs (and 16% in a bear
case) by 2030. For autos, the potential combination of consumers choosing to own AVs
personally longer-term, plus an increase in miles traveled enabled by the improved
experience, leads us to our base case view that the long-term average for SAAR will
remain relatively similar to the historical ~16 mn level. However, in a bear case scenario
in the long-term where all miles traveled use AV rideshare, we estimate that US
SAAR could be 3-6 mn units lower.
In terms of stocks, we highlight GOOGL, TSLA, UBER, AUR, AMZN, Pony AI, RIVN, MBLY,
LYFT, TEL, Hesai, XPeng, and Volvo Group as stocks in our coverage that are beneficiaries
of AVs and/or where we believe investor concerns about risks from AVs are overdone.
We see mixed implications for traditional auto OEMs, with risks to unit and/or profit
share if newer tech companies disrupt the market for personal transportation, but also
potential opportunities tied to software and digital services. We’d also note that
elements of competition, execution risk, valuation, and, in some cases, further capital
needs keep us on the sidelines for certain stocks that have revenue exposure to the
autonomous market (as discussed on a company by company basis in the Stocks in
Focus section of this report).
16 April 2026 4
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
The ecosystem of autonomy
Exhibit 1: Potential global autonomy ecosystem market size in 2035
Source: Goldman Sachs Global Investment Research
Exhibit 2: Illustrative autonomous ecosystem
Source: Goldman Sachs Global Investment Research
16 April 2026 5
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
PM summary
Datacenter capex from leading public hyperscalers is now approaching ~$700 bn
USD, roughly 10X the level in 2020. With spend continuing to grow, and consensus
capex estimates being revised higher yet again with the most recent set of earnings
reports, we believe the return on that capex remains a key focus for the financial
markets. We therefore examine profit pools that AI could impact, with this report
focusing on the transportation market.
The pace of autonomous technology commercialization has accelerated since our
June 2025 report, “Framing profit pools in rideshare, trucking, and insurance as AVs
move from concept to commercialization.” This acceleration is being driven by both a
growing number of deployments in the US and China, and also international expansions
into new geographies including EMEA. This has been enabled by factors including the
safety/performance from leading AV companies, and customer demand.
For example, Waymo was deployed fully driverlessly and available to the public in 5 cities
at the end of 2025 and plans to be in >15 by the end of 2026, Uber expects to be
facilitating AV trips in as many as 15 cities by the end of 2026, Pony AI expects to be in
20+ cities by the end of 2026, and WeRide plans to more than double its AV fleet by the
end of 2026.
We now estimate that the AV robotaxi market in the US will reach $19 bn in 2030, up
from $7 bn prior, and we introduce our 2035 forecast of $48 bn. When including
global markets, we estimate that the robotaxi market in 2035 could be ~$415 bn.
We assume that the gross margins of a vertically integrated operator could be 30%-50%,
implying gross profit for the market of ~$150 bn in 2035 and cumulative gross profit of
~$440 bn over the next decade for robotaxis.
Exhibit 3: Hyperscaler capex CY2020-27E
13 18 27 25
49
89
151 166
21 28 28
41
76
118
158
188
22 25 31 32
53
91
183
211
16 19
32 28 39
72
132
159
2 3 7 7 11
35
52 59
73
93
125 133
227
406
676
782
0
100
200
300
400
500
600
700
800
900
C
ap
ex
($
b
ns
)
Source: Company data, Goldman Sachs Global Investment Research
16 April 2026 6
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
Class 8 trucking is also growing, and our forecast remains similar to our prior outlook
for the market by 2030 in the US (we assume growing to $16 bn in 2030 from low levels
currently), and we think the market could reach $105 bn in the US and ~$560 bn
globally by 2035. We estimate that gross profit for AV trucking will be ~$135 bn in 2035,
and about $300 bn cumulatively over the next decade.
We expect that part of the increased adoption and deployments through 2035 will be
driven by the cost of AV trucking improving relative to a human driver. At present, we
believe the additional upfront costs to make an AV truck in the US are
~$125K-$150K, and we expect this to decline to a premium of ~$35-40K in 2035,
driven by improved hardware costs and economies of scale. With driver wages
trending higher, and as AV hardware costs are monetized over more miles, we
forecast that AV trucking costs per mile for commercial deployments to be better
than human operated trucks in 2028 in the US. There are also routes that AV trucks
can serve that are beyond what a human can do in a day given hours of driving limits.
Considering the market more broadly for AV hardware, software & services, we
estimate that revenue associated with the industry could reach approximately $2
trillion in 2035 comprised of hardware sales (. robotaxis, consumer L3-L5 vehicles,
AV trucks, and delivery bots) and software/digital services (. robotaxi and AV trucking
Exhibit 4: Robotaxi market size in $ mns, 2023-2035E
0
50,000
100,000
150,000
200,000
250,000
300,000
350,000
400,000
450,000
2023 2024 2025 2026E 2027E 2028E 2029E 2030E 2031E 2032E 2033E 2034E 2035E
R
o
b
o
ta
x
i
m
a
rk
e
t
s
iz
e
(
$
m
n
s
)
US China Europe Others
Source: Company data, Goldman Sachs Global Investment Research
Exhibit 5: Global AV trucking market size ($ mns),
2023-2035E
0
100,000
200,000
300,000
400,000
500,000
600,000
2
0
2
3
2
0
2
4
2
0
2
5
2
0
2
6
E
2
0
2
7
E
2
0
2
8
E
2
0
2
9
E
2
0
3
0
E
2
0
3
1
E
2
0
3
2
E
2
0
3
3
E
2
0
3
4
E
2
0
3
5
E
G
lo
b
a
l
A
V
t
ru
c
k
m
a
rk
e
t
s
iz
e
(
$
m
n
s
)
US China Europe Others
Source: Company data, Goldman Sachs Global Investment Research
16 April 2026 7
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
services, and consumer autonomous subscriptions for L3-L5). Recognizing that a
meaningful portion of this $2 trillion of revenue associated with autonomy is coming
from business for goods/services that isn’t incremental (. buying an L3/L4 car rather
than an L1/L2 car, and freight being hauled by an AV truck rather than a person), when
focusing more specifically on the piece of this market attributable to AI (. revenue
for virtual driver technology in robotaxis and AV trucks, and software subscriptions for
consumer L3-L5 autonomy), we estimate that the market is approximately ~$300 bn
in 2035.
Our global forecast is informed by the views of the regional teams and company
specific outlooks for key covered companies, including views on the China market and
AV providers from Allen Chang and team detailed in their report (see link).
We recognize there could be downside risks related to factors such as more restrictive
regulations than we expect and/or more difficult industry fundamentals. Upside risks
could be driven by factors such as better than expected end-user economics.
While a portion of AV volumes will likely be incremental demand (. as new use
cases become affordable or possible), we believe autonomy could also disrupt
existing markets in the long-term. We frame risks and scenarios including to
human-operated rideshare/taxis, trucking, and light vehicle unit sales in the United
States in this report. We estimate the economic size that could potentially be
disrupted in the US is ~$440 bn. This is comprised of wages for taxi/chauffeur/shuttle,
delivery and truck drivers per BLS data, an estimate for the share of bookings from
rideshare allocated to drivers, and how much auto sales could decline in a scenario
where personal transport demand is served only by AV rideshare.
There are already signs of change, with Waymo’s share in SF reaching 30%, 20
months after being fully launched (per Yipit), and our base case view is for 5%
cannibalization of UCAN rideshare gross bookings from AVs (and 16% in a bear
case) by 2030.
For autos, the potential combination of consumers choosing to own AVs personally
longer-term, plus an increase in miles traveled enabled by the improved experience,
leads us to our base case view that the long-term average for SAAR will remain relatively
Exhibit 6: Potential impact to human rideshare from
robotaxis, with a base case assumption of 5%
cannibalization/disruption and scenarios for a net positive
impact in a bull case or 16% cannibalization in a bear case
-20%
-15%
-10%
-5%
0%
5%
2025 2026E 2027E 2028E 2029E 2030E
%
d
is
ru
pt
io
n
to
h
um
an
ri
de
sh
ar
e
Base Case Bear Case Bull Case
Source: Goldman Sachs Global Investment Research
16 April 2026 8
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
similar to the historical ~16 mn level. However, in a bear case scenario where all miles
traveled use AV rideshare, we estimate that US SAAR could be 3-6 mn units lower.
We assume, for the purposes of this illustrative bear case analysis, that vehicles have a
useful life of ~400K miles, that utilization in terms of miles traveled per vehicle per year
rises given the scenario that all vehicles in operation (VIO) are shared AVs, and that
vehicle miles traveled (or VMT) ranges from unchanged to 33% higher as consumers may
use more vehicle transportation if it’s less costly and doesn’t require a driver’s attention.
Exhibit 7: Implied US SAAR (in mns) in a scenario where the VIO/VMT is all shared AVs by
2040
0% 2% 4% 5% 10% 15% 20% 25% 33%
12,500 10 10 10 11 11 12 12 13 13
15,000 10 10 10 11 11 12 12 13 13
20,000 10 10 10 11 11 12 12 13 13
30,000 10 10 10 11 11 12 12 13 13
40,000 10 10 10 11 11 12 12 13 13
50,000 10 10 10 11 11 12 12 13 13
75,000 10 10 10 11 11 12 12 13 13
100,000 10 10 10 11 11 12 12 13 13
150,000 10 10 10 11 11 12 12 13 13
200,000 10 10 10 11 11 12 12 13 13
% increase in VMT
m
il
e
s
p
e
r
A
V
Source: Goldman Sachs Global Investment Research
16 April 2026 9
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
AI tech capabilities and capital - gauging costs and growth
Datacenter capex from leading public hyperscalers is now approaching ~$700 bn
USD, roughly 10X the level in 2020. In fact, consensus capex estimates were revised
higher yet again with the most recent set of earnings reports, with the bottom-up
aggregation of 2026 GS capex estimates for the leading hyperscalers now 14% higher
relative to the start of 4Q25 earnings. GS analysts now expect 67% yoy growth in capex
in 2026, and 16% in 2027 (Exhibit 8).
This growth is primarily to drive AI model development and datacenter capacity, and we
take a look at recent developments in this section of the note.
One measure of progress with AI models is Epoch AI’s Capabilities Index (ECI), which is a
generalized benchmark intended to capture the overall capability of AI models by
aggregating performance across several established tests covering reasoning, coding,
math, and language. Unlike individual benchmarks which can saturate as models
approach near‑perfect scores, ECI adjusts for task difficulty and combines results to
show relative differentiation across models and over time. Higher scores indicate
broader capability, while models optimized for specific tests may perform well on
individual benchmarks but rank lower on ECI due to limited generalization.
Models launched in 2023 were scoring around the low to mid 120s range, while models
released in 2026 ( Pro, Gemini Pro, Claude Sonnet ) are scoring in the mid
to high 150s range (Exhibit 9).
Exhibit 8: Hyperscaler capex CY2020-27E
13 18 27 25
49
89
151 166
21 28 28
41
76
118
158
188
22 25 31 32
53
91
183
211
16 19
32 28 39
72
132
159
2 3 7 7 11
35
52 59
73
93
125 133
227
406
676
782
0
100
200
300
400
500
600
700
800
900
Source: Company data, Goldman Sachs Global Investment Research
16 April 2026 10
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
The Stanford Institute for Human-Centered AI (HAI) also details how AI models have
improved over time, most recently in their 2026 AI index report. Per HAI, AI model
performance has continued to trend higher overall, with AI models able to exceed
human capabilities on several metrics. However, certain tasks remain more challenging
for AI models, with the 2026 report discussing low accuracy from several models at
reading analog clocks.
In addition to foundational model performance, multiple companies are seeking to
better implement AI models and tools into practice commercially, either with in-house
technology (. Waymo, Tesla, Pony AI) or by leveraging third party tools and
technology (. from Nvidia, Qualcomm, Wayve, Mobileye, Horizon Robotics, Helm,
Applied Intuition, etc.).
Nvidia’s Alpamayo open-sourced model and tools that support reasoning, announced in
January with CES, led to increased conversation among investors on whether it would
Exhibit 9: Select AI model performance on the Epoch
Capabilities Index
GPT-4
GPT-5
Pro
Claude
Opus
Gemini Pro
Gemini
Pro
Grok 4
70
80
90
100
110
120
130
140
150
160
170
Feb-23 Aug-23 Feb-24 Aug-24 Feb-25 Aug-25 Feb-26
E
C
I
S
c
o
re
Release Date
OpenAI Anthropic Google xAI
Source: Epoch AI
Exhibit 10: Select AI Index technical performance benchmark vs. human performance
0
20
40
60
80
100
120
2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025
P
e
rf
o
rm
a
n
c
e
r
e
la
ti
v
e
t
o
t
h
e
h
u
m
a
n
b
a
s
e
li
n
e
(
%
)
Image classification (ImageNet Top-5) English language understanding (SuperGLUE)
Multitask language understanding (MMLU) PhD-level science questions (GPQA Diamond)
Agent multimodal computer use (OSWorld) Autonomous software engineering (SWE-bench Verified)
Visual reasoning (VQA) Medium-level reading comprehension (SQuAD )
Competition-level mathematics (MATH) Multimodal understanding and reasoning (MMMU)
Mathematical reasoning (AIME) Human baseline
Source: Stanford University Institute for Human-Centered AI, Goldman Sachs Global Investment Research
16 April 2026 11
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
significantly accelerate the time-to-market and increase the number of companies that
could provide L4 AV technology (and we do expect a number of companies to ultimately
offer AV products, either with their own technology or by utilizing
partnerships/suppliers). The Mercedes CLA is scheduled to be the first vehicle in the
market using Alpamayo, initially as an L2+ product this year in the US.
One key component of Alpamayo is the open-sourced reasoning vision-language-action
model (VLA). The tools also include AlpaSim (an open-source AV simulation framework
which provides realistic sensor modeling, configurable traffic behavior, and scalable
closed-loop testing) and open datasets for AVs (with 1700+ hours of driving comprised
of 300K+ clips, each ~20 seconds long, recorded in 25 countries and 2500 cities with
diverse traffic, weather, and obstacles).
The tools from Nvidia tie in with a broader debate currently in the industry on how much
of AV model development and testing can utilize simulation, and how much real-world
data is needed, although we think a mixture of both remains logical (with increasingly
sophisticated simulation tools helping to speed development cycles and future iterations
of the technology, in our opinion).
Another key theme in the industry relates to how AI is implemented into AV policy. This
includes the ability to utilize an end-to-end approach and advanced AI techniques over
legacy rules-based models. While we think most if not all AV companies are now
adopting advanced AI techniques for at least elements of their technology development,
we believe some are using more holistic approaches (as explained by Waymo, for
example, in a December 2025 blog post), while others argue for the generalization
benefits of a more pure end-to-end approach.
Finally, there is a debate on the role and need for maps. With Waymo (which has
traditionally used maps) now scaling, we think there is increasing evidence that using
maps is a workable approach, at least for commercial applications that operate in a
geofence, although others, including Tesla, have suggested that a generalized approach
without maps will allow for faster scaling longer-term.
How much does AV development cost for capex?
One of the objectives of this report is to gauge capex demand and ROI tied to AVs. The
level of capex the industry in total would need depends in part on how many different
companies develop the technology. However, one case study is from Tesla, which has
described a plan to grow its H100-equivalent GPU count to ~275K by the summer of
2026 from ~110K-125K at the end of 2025. The GS semi team estimates that Nvidia
GPUs cost $25K-$45K each (with Hopper at the low-end and Rubin at the high-end of
this), and that Nvidia hardware (including networking) can be ~60% of a project cost. We
estimate this implies that Tesla’s capex for AI infrastructure would reach at least $5 bn
and could be in the $10 bn range cumulatively by this summer. Many of the AV
companies, per our industry discussions, utilize cloud training (. AWS) rather than
their own datacenters.
16 April 2026 12
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
Robotaxi deployments and market size accelerating, aided by safety and consumer adoption
Global AV deployments are accelerating. For example, Waymo was deployed fully
driverlessly and available to the public in 5 cities at the end of 2025 and plans to be in
>15 by the end of 2026, Uber expects to be facilitating AV trips in as many as 15 cities
by the end of 2026, Pony AI expects to be in 20+ cities by the end of 2026, and WeRide
plans to more than double its AV fleet by the end of 2026.
We show current and planned deployments for Waymo, Uber, Lyft, Zoox, Pony AI, Baidu,
WeRide, Mobileye, May Mobility, and Grab in Exhibit 11 and Exhibit 12 (we consider this
a representative but not exhaustive list of deployments). We also note partnerships with
Grab and Bolt exist, similar to plans from Uber and Lyft, to deploy in unspecified cities
across several AV providers (. WeRide, May Mobility, etc).
Exhibit 11: US Robotaxi Operations
Active cities defined as locations where operations are fully launched, fully driverless, and available to the public. In development includes cities where testing &
mapping operations are currently in progress or where rides are available with safety observers/monitor onboard.
Source: Company data, Goldman Sachs Global Investment Research
16 April 2026 13
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
These deployments are driving a growing fleet and market. We estimate the fleet of
commercial AVs globally will rise from ~7K last year to ~1 mn in 2030 and ~6 mn in 2035,
implying an ~96% CAGR between 2025-2035E. These forecasts are based on input from
the regional teams and views for covered companies.
We now estimate that the AV robotaxi market in the US will reach $19 bn in 2030, up
from $7 bn prior, and we introduce our 2035 forecast of $48 bn. When including
global markets, we estimate that the robotaxi market in 2035 could be ~$415 bn.
Exhibit 12: International Robotaxi Operations (ex. China)
Active cities defined as locations where operations are fully launched, fully driverless, and available to the public. In development includes cities where testing &
mapping operations are currently in progress or where rides are available with safety observers/monitor onboard.
Source: Company data, Goldman Sachs Global Investment Research
Exhibit 13: Robotaxi fleet by region, 2023-2035E
Exhibit 14: Robotaxi market size in $ mns, 2023-2035E
0
1,000
2,000
3,000
4,000
5,000
6,000
7,000
2023 2024 2025 2026E 2027E 2028E 2029E 2030E 2031E 2032E 2033E 2034E 2035E
R
o
b
o
ta
x
i
fl
e
e
t
s
iz
e
(
0
0
0
s
)
US China Europe Others
0
50,000
100,000
150,000
200,000
250,000
300,000
350,000
400,000
450,000
2023 2024 2025 2026E 2027E 2028E 2029E 2030E 2031E 2032E 2033E 2034E 2035E
R
o
b
o
ta
x
i
m
a
rk
e
t
s
iz
e
(
$
m
n
s
)
US China Europe Others
Source: Company data, Goldman Sachs Global Investment Research
Source: Company data, Goldman Sachs Global Investment Research
16 April 2026 14
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
Safety and performance are one set of factors enabling the faster pace of AV
deployments...
Safety across AVs is improving, and we think autonomous technology in some situations
already eclipses human performance. Waymo, for example, provides safety data on its
website and notes that across its operating cities, Waymo has 83% fewer accidents with
an airbag deployment and 82% fewer injury causing accidents. A past study from Swiss
Re showed Waymo had an 88% reduction in property damage claims and 92% reduction
in bodily injury claims relative to a human benchmark, while a recent audit from TUV
SUD validated Waymo’s safety case, per the company.
Exhibit 15: US Robotaxi total fleet size new vs old,
2023-2030E
Exhibit 16: US robotaxi fleet and market size, 2023-2035E
0
10,000
20,000
30,000
40,000
50,000
60,000
70,000
2023 2024 2025 2026E 2027E 2028E 2029E 2030E
U
S
R
o
b
o
ta
x
i
to
ta
l
fl
e
e
t
s
iz
e
New Old
0
50
100
150
200
250
0
10,000
20,000
30,000
40,000
50,000
60,000
2023 2024 2025 2026E2027E2028E2029E2030E2031E2032E2033E2034E2035E
R
o
b
o
ta
x
i
fl
e
e
t
s
iz
e
(
0
0
0
s
)
R
o
b
o
ta
x
i
m
a
rk
e
t
s
iz
e
(
$
m
n
s
)
Robotaxi market size Robotaxi fleet
Source: Company data, Goldman Sachs Global Investment Research
Source: Company data, Goldman Sachs Global Investment Research
Exhibit 17: US Robotaxi market size in $ mns, 2023-2030E
new vs old
0
2,000
4,000
6,000
8,000
10,000
12,000
14,000
16,000
18,000
20,000
2023 2024 2025 2026E 2027E 2028E 2029E 2030E
U
S
R
o
b
o
ta
x
i
m
a
rk
e
t
s
iz
e
(
$
m
n
s
)
New Old
Source: Company data, Goldman Sachs Global Investment Research
16 April 2026 15
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
We also compiled NHTSA accident reports for Waymo and Tesla compared to AV miles
traveled between July 2025 and March 2026 (with Waymo across a range of US cities,
and Tesla for its fleet in Austin that has a mix of unsupervised and monitored vehicles
with Tesla having removed safety monitors from some vehicles in January). We’d note
that crashes reported to NHTSA under federal regulations include smaller incidents that
human drivers are unlikely to report (. damage due to riding over a curb, bumping a
pole backing out of a parking spot, etc). Waymo’s miles between accidents has been
trending higher in recent months. Tesla’s performance has been more varied, and there
were no accidents reported by Tesla in February or March (we think a key reason for the
larger variation in Tesla is the still small fleet size).
...with consumer demand another key driver
We believe consumer interest and demand for AVs is also a key factor in the market
growth, as shown by growing penetration rates as a percent of Uber and Lyft monthly
active users (MAUs), solid Google search traffic in SF, and the rising number of Waymo
deployments.
Exhibit 18: Waymo airbag deployed accidents vs human
benchmark in its key active markets
0
1
2
Austin LA Phoenix SF Blended total
In
c
id
e
n
ts
p
e
r
m
il
li
o
n
m
il
e
s
Waymo Benchmark
Source: Company data
Exhibit 19: Waymo and Tesla miles between accidents
0
50,000
100,000
150,000
200,000
250,000
300,000
350,000
400,000
M
il
e
s
b
e
tw
e
e
n
a
c
c
id
e
n
ts
Waymo miles between accidents Tesla miles between accidents
March is partial month data through March 16th; Tesla didn’t report any
accidents in August, February, or March
Source: NHTSA, Company data, Goldman Sachs Global Investment Research
16 April 2026 16
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
Importantly, while a portion of the Waymo app users are still likely using it to experience
an AV (potentially while traveling to a city that offers AV rides), retention rates of the app
have improved.
Exhibit 20: Waymo US users as % of Uber & Lyft
Exhibit 21: Google trends - SF
%
%
%
%
%
%
%
%
%
Ja
n-
22
M
ar
-2
2
M
ay
-2
2
Ju
l-2
2
S
ep
-2
2
N
ov
-2
2
Ja
n-
23
M
ar
-2
3
M
ay
-2
3
Ju
l-2
3
S
ep
-2
3
N
ov
-2
3
Ja
n-
24
M
ar
-2
4
M
ay
-2
4
Ju
l-2
4
S
ep
-2
4
N
ov
-2
4
Ja
n-
25
M
ar
-2
5
M
ay
-2
5
Ju
l-2
5
S
ep
-2
5
N
ov
-2
5
Ja
n-
26
M
ar
-2
6
% of Uber MAUs % of Lyft MAUs
0
20
40
60
80
100
120
Ja
n-
12
A
ug
-1
2
M
ar
-1
3
O
ct
-1
3
M
ay
-1
4
D
ec
-1
4
Ju
l-1
5
F
eb
-1
6
S
ep
-1
6
A
pr
-1
7
N
ov
-1
7
Ju
n-
18
Ja
n-
19
A
ug
-1
9
M
ar
-2
0
O
ct
-2
0
M
ay
-2
1
D
ec
-2
1
Ju
l-2
2
F
eb
-2
3
S
ep
-2
3
A
pr
-2
4
N
ov
-2
4
Ju
n-
25
Ja
n-
26
Uber Lyft Waymo
Green line represents partnership city launch; red line represents Waymo One
Exclusive service launch
Source: SensorTower, Data compiled by Goldman Sachs Global Investment
Research
Source: Goldman Sachs Global Investment Research, Google trends
(
Exhibit 22: Waymo US growth trajectory
Exhibit 23: MAU composition of Waymo users in the USA
0%
50%
100%
150%
200%
250%
300%
350%
400%
0
200,000
400,000
600,000
800,000
1,000,000
1,200,000
1,400,000
1,600,000
1,800,000
Ja
n-
22
M
ar
-2
2
M
ay
-2
2
Ju
l-2
2
S
ep
-2
2
N
ov
-2
2
Ja
n-
23
M
ar
-2
3
M
ay
-2
3
Ju
l-2
3
S
ep
-2
3
N
ov
-2
3
Ja
n-
24
M
ar
-2
4
M
ay
-2
4
Ju
l-2
4
S
ep
-2
4
N
ov
-2
4
Ja
n-
25
M
ar
-2
5
M
ay
-2
5
Ju
l-2
5
S
ep
-2
5
N
ov
-2
5
Ja
n-
26
M
ar
-2
6
MAUs YoY Growth
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Ja
n-
24
M
ar
-2
4
M
ay
-2
4
Ju
l-2
4
S
ep
-2
4
N
ov
-2
4
Ja
n-
25
M
ar
-2
5
M
ay
-2
5
Ju
l-2
5
S
ep
-2
5
N
ov
-2
5
Ja
n-
26
M
ar
-2
6
SF LA PHX MIA DAL HOU SAT ORL Other
Green line represents partnership city launch; red line represents Waymo One
Exclusive service launch
Source: SensorTower, Data compiled by Goldman Sachs Global Investment
Research
Source: SensorTower, Data compiled by Goldman Sachs Global Investment
Research
16 April 2026 17
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
AV profitability improving
We expect the cost of AVs to come down over time, enabled by purpose-built hardware
and scale, putting the industry on a path to costs per vehicle that are well under the
historical >$100K range. For example, while Waymo’s 5th gen AV has 29 cameras, the
6th Gen is down to 13 (along with 4 lidar units and 6 radar sensors). Similarly, Tesla
believes its most recent vehicles (HW4 compute) have the necessary hardware to work
as robotaxis (they do not use radar or lidar), and these vehicles start with prices for
consumers in the US (prior to any incentives) in the mid $30K USD range. Although we
recognize that market economics in China are different, purpose-built AV costs in China
have already reached the sub $40K USD range, and our team expects costs to decline
further over the next 10 years.
We updated our illustrative cost model for a vertically integrated AV rideshare provider
from our 2025 report, and assume COGS per mile are sub $1 in 2035 (Exhibit 28).
Specifically, we assume: 1) depreciation costs per mile could decline from ~$ in
Exhibit 24: Retention cohort data - benchmarking Waymo
vs. Uber/Lyft
October 2025 Cohort
Exhibit 25: Waymo US retention trends over time
0%
5%
10%
15%
20%
25%
30%
Oct-25 Nov-25 Dec-25 Jan-26 Feb-26 Mar-26
Uber Lyft Waymo
%
%
% %
%
%
%
%
%
%
%
%
%
%
%
Ja
n-
25
F
eb
-2
5
M
ar
-2
5
A
pr
-2
5
M
ay
-2
5
Ju
n-
25
Ju
l-2
5
A
ug
-2
5
S
ep
-2
5
O
ct
-2
5
N
ov
-2
5
D
ec
-2
5
Ja
n-
26
F
eb
-2
6
M
ar
-2
6
Day 30 Day 90
+63bps
Source: SensorTower, Data compiled by Goldman Sachs Global Investment
Research
Source: SensorTower, Data compiled by Goldman Sachs Global Investment
Research
Exhibit 26: US user session count over time
per user
Exhibit 27: Rideshare usage penetration %
Ja
n-
22
M
ar
-2
2
M
ay
-2
2
Ju
l-2
2
S
ep
-2
2
N
ov
-2
2
Ja
n-
23
M
ar
-2
3
M
ay
-2
3
Ju
l-2
3
S
ep
-2
3
N
ov
-2
3
Ja
n-
24
M
ar
-2
4
M
ay
-2
4
Ju
l-2
4
S
ep
-2
4
N
ov
-2
4
Ja
n-
25
M
ar
-2
5
M
ay
-2
5
Ju
l-2
5
S
ep
-2
5
N
ov
-2
5
Ja
n-
26
M
ar
-2
6
Uber Lyft Waymo
%
%
%
%
%
%
%
%
%
%
%
Ja
n-
22
M
ar
-2
2
M
ay
-2
2
Ju
l-2
2
S
ep
-2
2
N
ov
-2
2
Ja
n-
23
M
ar
-2
3
M
ay
-2
3
Ju
l-2
3
S
ep
-2
3
N
ov
-2
3
Ja
n-
24
M
ar
-2
4
M
ay
-2
4
Ju
l-2
4
S
ep
-2
4
N
ov
-2
4
Ja
n-
25
M
ar
-2
5
M
ay
-2
5
Ju
l-2
5
S
ep
-2
5
N
ov
-2
5
Ja
n-
26
M
ar
-2
6
Uber Lyft Waymo
Session count over time = number of times a user opens an application in a
given month
Source: SensorTower, Data compiled by Goldman Sachs Global Investment
Research
Usage penetration = percentage of users with one or more sessions in the app
during the month
Source: SensorTower, Data compiled by Goldman Sachs Global Investment
Research
16 April 2026 18
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
2025 to ~$ in 2035 for a representative AV; 2) insurance costs will decline from
~$ in 2025 (a premium to human-driven rideshare of about $) to about $
per mile in 2035; 3) wages for remote operators per mile will decline to $ in 2035
from $ in 2025, driven by a higher ratio of vehicles per operator (from 6 cars to 1
operator in 2025 to 26 to 1 in 2035). Note that we have updated our cost model
compared to our 2025 report to better capture empty miles.
Overall, we assume that the gross margins of a vertically integrated operator could be
30%-50%, implying cumulative global gross profit of ~$440 bn over the next decade for
robotaxis.
How much of profits are for a “virtual driver”
We also frame potential revenue for a “virtual driver” business model where companies
could provide their software for a per mile fee to robotaxi fleet operators. For context
on how revenue/costs are currently allocated, Lyft committed to paying human drivers
at least 70% of rider payments per week after external fees such as commercial
insurance (we estimate ~$ per mile) are subtracted, and Lyft estimates that there
are ~$ of expenses associated with operating the car for the human driver (.
fuel costs, maintenance, cleaning, and depreciation). If we assumed these types of
economics for a virtual driver, and applied that to the full number of miles we expect
Exhibit 28: Illustrative cost per mile for Robtaxis by
region, 2023-2035E
Exhibit 29: Robotaxi gross margin by region, 2023-2035E
$
$
$
$
$
$
$
$
2
0
2
3
2
0
2
4
2
0
2
5
2
0
2
6
E
2
0
2
7
E
2
0
2
8
E
2
0
2
9
E
2
0
3
0
E
2
0
3
1
E
2
0
3
2
E
2
0
3
3
E
2
0
3
4
E
2
0
3
5
E
Il
lu
s
tr
a
ti
v
e
C
O
G
S
p
e
r
m
il
e
US Global
-200%
-150%
-100%
-50%
0%
50%
100%
2
0
2
3
2
0
2
4
2
0
2
5
2
0
2
6
E
2
0
2
7
E
2
0
2
8
E
2
0
2
9
E
2
0
3
0
E
2
0
3
1
E
2
0
3
2
E
2
0
3
3
E
2
0
3
4
E
2
0
3
5
E
Il
lu
s
tr
a
ti
v
e
g
ro
s
s
m
a
rg
in
US Global
Source: Company data, Goldman Sachs Global Investment Research, AAA
Source: Goldman Sachs Global Investment Research
Exhibit 30: Potential gross profit pool by region from
Robotaxis in $ mns, 2023-2035E
-20,000
0
20,000
40,000
60,000
80,000
100,000
120,000
140,000
160,000
2
0
2
3
2
0
2
4
2
0
2
5
2
0
2
6
E
2
0
2
7
E
2
0
2
8
E
2
0
2
9
E
2
0
3
0
E
2
0
3
1
E
2
0
3
2
E
2
0
3
3
E
2
0
3
4
E
2
0
3
5
E
R
o
b
o
ta
x
i
g
ro
s
s
p
ro
fi
t
p
o
o
l
($
m
n
s
)
US China Europe Others
Source: Goldman Sachs Global Investment Research
16 April 2026 19
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
robotaxis to drive in 2030 and 2035, the theoretical global TAM for a virtual driver
would be $12 bn and $106 bn respectively. Note that this is illustrative, and we expect,
at least over the near to intermediate term, a majority of AVs to be owned and operated
by tech stack providers like Waymo, Pony AI, and Tesla.
Regulations
As it relates to regulations for autonomous vehicles, in the US, there is no specific federal
law at present that requires a human driver, although there are rules (. FMVSS) on
what vehicle features are mandatory. While the current administration is taking steps to
work towards a federal framework (. the SELF DRIVE Act) and adjusting the FMVSS
rules to allow for more flexibility in robotaxi/AV design, regulations for AVs are still set
largely on a state-by-state and local basis. Additionally, the US DOT recently solicited
public comment for potential actions (. becoming a signatory to, adopting portions
of, or declining to join) on the upcoming UN Global Technical Regulation on
AVs/Automated Driving Systems. At the state level currently, some states (. California)
require several permitting and reporting obligations, and others (. Texas) have more
limited rules. We note that presently, in order to deploy AVs that don’t meet full FMVSS
requirements (such as purpose-built AV with no steering wheel and gearshifts), approval
Exhibit 31: Potential market for a robotaxi “virtual driver”
$ bns 2030 2035
US $6 $16
China $5 $31
Europe $1 $23
Others $0 $36
Total $12 $106
Source: Company data, Goldman Sachs Global Investment Research
Remote assistance 101
We have received some investor questions on remote assistance, and how this works. While companies can
make different choices on how to implement this, most companies use it as a phone a friend type role
for unusual situations or extra advice, not to regularly remote control them. AVs only implement the
advice coming from remote assistance if they can do so safely. For example, as noted by Waymo, if an
emergency vehicle is occupying a lane with its lights on and appears parked at the entrance of a lane, and
the robotaxi is uncertain if the lane is closed, it could query a remote assistance agent to verify if the lane is
open. In another example, if there is a narrow two-way street and an AV needs to make way for a large
vehicle (. truck), the remote assistance agent could provide a potential temporary pull over point that
may be outside of the vehicle’s traditional operating domain (. a private driveway). Several AV
companies responded to a US Senate questionnaire linked here on how they use remote assistance.
For context on how remote assistance has scaled, typically, in an early deployment, the ratio of vehicles to
remote assistant is low, per our industry discussions. But as the geofenced areas and software are proven
out, the ratio can go into the double digits. Pony AI has said they’re now at about 30:1 (vehicles to remote
assistant) in some locations. Similarly, Waymo disclosed about 70 people typically on shift as remote
assistants, and they have a fleet of >3K vehicles (although not all those vehicles are operating
continuously).
16 April 2026 20
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
is needed and there is a limit of 2,500 vehicles per year for up to 2 years (or a maximum
of 5,000 in total) that don’t comply.
Unlike the commercial operations already underway in the US, a small series of piloting
projects and research tests are being conducted on a country by country level in Europe,
., in the UK, Germany, France, and Luxembourg, with no existing public driverless
operations. However, Europe is well-positioned for L4 commercial deployment with a
robust legal foundation already in place: an EU-wide type-approval regime provides a
clear pathway for L4 vehicles, while the UK is actively rolling out its Automated Vehicles
Act 2024 — with the first pilots of automated passenger services set to launch on UK
roads in 1H26. A global effort would accelerate regulatory convergence as well: A new
global ADS regulation covering L3/L4 will be submitted to the UNECE in June 2026; if
adopted, it would enter into force immediately. Both the EU and UK are positioned to
align with this framework, while the UK will also finalize a series of calls for evidence and
consultations ahead of full implementation of its Act by the end of 2027. This represents
a positive trajectory for robotaxi operators: the new UN regulation is purpose-built for
L4 driverless use cases, where companies can pursue a single regulatory strategy across
major markets – making the path to commercial deployment clearer and more
achievable.
In China, AV testing and deployment is allowed under permits from the government
across many cities (. Beijing and Shanghai) with specific safety standards that must be
met. There is policy at the national level (. the Notice on pilot work of intelligent
connected vehicle access and road access), as well as local and/or city level polices that
must be followed. Similar to the US, China regulations and deployment polices are more
formally carried out at the local level. Please see section 11 of Allen Chang’s May 2025
note, “China’s Robotaxi market - the road to commercialization” for more details on
policies for robotaxis in China.
Autonomous last mile delivery bots in focus
Another application of AVs that we believe will continue to scale is the use of last mile autonomous delivery
bots. These last mile delivery bots are smaller than light vehicles and you can see them operating on
sidewalks and college campuses in some cities. Several last mile delivery bot companies partner with third
party platform providers ( Uber Eats and DoorDash), food companies (. Chipotle, Shake Shack),
convenience stores ( 7-Eleven), and parcel/package delivery companies. Notably, the AVs that are
currently deployed in major cities generally operate within smaller radii (generally under 3 miles). Looking
across some of the key last mile delivery bot players (. Serve Robotics, Starship Tech, Coco Robotics, and
Neolix) we believe last mile delivery bot revenue was a couple hundred million dollars with a fleet size
between 20-30K globally in 2025. We expect this market to grow to $40-$50 bn in 2035 which assumes a
fleet size between mn delivery bots globally. Some key assumptions to our forecast include an
increasing miles per trip from ~ to 3 miles and $4-5 of revenue per trip.
16 April 2026 21
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
Exhibit 32: We continue to expect growth in the last
mile delivery bot market
0
200,000
400,000
600,000
800,000
1,000,000
1,200,000
1,400,000
$0
$5,000
$10,000
$15,000
$20,000
$25,000
$30,000
$35,000
$40,000
$45,000
$50,000
2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035
F
le
e
t
s
iz
e
D
e
li
v
e
ry
b
o
t
re
v
e
n
u
e
(
in
$
m
n
)
Last mile delivery revenue ($mn) Fleet size
Source: Goldman Sachs Global Investment Research, Company data
16 April 2026 22
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
Consumer AVs - partially and fully autonomous vehicles poised to grow
The vast majority of the global car parc and miles traveled come from personally owned
vehicles rather than shared mobility/rideshare, making consumer applications of
autonomous capabilities a particularly large potential market for the long-term.
Numerous auto OEMs are focused on bringing either L3 (which is situationally eyes-off
autonomy, such as on a highway in nice weather) or L4 technology (which is fully eyes-off
and hands-off driving within a geofenced area like a part of a city) to consumer vehicles.
Currently L4 AV technology is only available in commercial vehicles (. rideshare
applications). However, as operational design domains (ODDs) expand into wider areas
and the need for remote assistance declines, we think L4 technology can be utilized by
consumers (potentially with a subscription for remote assistance). While some AV
vehicles have expensive hardware (. Waymo), we do not believe vehicle costs will be a
long-term barrier to adoption. The cost of AVs in China even with sensors (including
lidar) and compute are already in the sub $40K USD range (less than the average price of
a new vehicle in the US at ~$45-50K), and Tesla believes its current consumer vehicles
(with prices in the US starting in the mid $30K range) have L4 capable hardware.
We expect the L3 and L4 mix of global vehicle shipments to grow in the coming years,
although we think the vast majority of L4 vehicle shipments for the near-term will
remain for commercial use cases.
The growing fleet of L3-L5 capable vehicles creates an opportunity for software
revenue. We frame potential market revenue with implied monthly ASPs ranging from
$25 to $200 for L3-5 autonomy software using our estimated global L3-5 installed base
of consumer vehicles (ex. robotaxis). This suggests the software market for consumer
vehicles associated with L3-5 technology could reach $50 bn+ in 2035. Please note
this would include software revenue being amortized from up-front purchases (and we
think some OEMs may require or embed software costs into upfront purchase prices on
at least certain trims to offset added costs), and also that these are global ASPs (.
pricing is likely lower in China but higher in the US). Other digital services (such as
predicative maintenance, insurance) would not be captured in this L3-L5 AV software
market.
Exhibit 33: L3-5 ADAS penetration rates for key regions and
globally, 2020-2040E
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
L
3
-5
A
D
A
S
p
e
n
e
tr
a
ti
o
n
US China Europe Global
Source: Wards, Goldman Sachs Global Investment Research
16 April 2026 23
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
Separately, our industry discussions and recent data from Wards suggests that the mix of
L2 capable vehicles (. vehicles that require supervision but can incorporate multiple
features like lane centering, lane changes, emergency braking, adaptive cruise control) in
the US and Europe is now very high (. ~80%), and we raised our L2 mix assumptions as
a percent of new vehicles sold compared to our prior AV report. There would likely be
additional software revenue from L2 features (FSD, for example, is currently L2), but we
expect this market to be increasingly competitive longer-term (with some OEMs in China
now providing L2 technology as a standard feature).
Addressing risks to personal ownership from AVs
There is a view from some investors that the proliferation of AVs will materially lower the
size of the personal vehicle market, given that personal vehicles are not used the
majority of the time. While this is a possibility, our base case view is that this is too
negative due to economic and use considerations. First, the cost of owning and
operating a personal vehicle in the US was $ to $ per mile in 2025 according to
AAA (at 10K and 15K miles traveled annually, respectively). This remains well below the
cost of rideshare at >$2 nationally. Second, while we expect the cost of AVs (and, in turn,
the price of rides with AVs) to fall (potentially to $1 or less long-term), these types of
economics would be enabled by hardware costs that are on par with or below the
current price of consumer vehicles. If the cost and convenience of an AV improves
enough in the long-term (., over 10-20 years), especially if users can sleep in them
while traveling, we see this being something many users would prefer to own (perhaps
with a monthly subscription to access remote assistance). Importantly, this aligns with
the ambitions of several OEMs. Waymo has said it’s potentially open to selling AVs to
individuals, Tesla’s plan is to allow for unsupervised personal autonomy in the long-term,
and GM, Rivian, and Ford have discussed L3/L4 efforts for personal autonomy.
Exhibit 34: Software revenue for L3-5 autonomy offerings on consumer vehicles globally ($ bns)
2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035
$25 $0 $0 $0 $1 $3 $5 $8 $11 $16 $21 $28
$50 $0 $0 $1 $3 $5 $9 $15 $22 $32 $43 $56
$75 $0 $0 $1 $4 $8 $14 $23 $34 $48 $64 $84
$100 $0 $0 $2 $5 $10 $18 $30 $45 $63 $86 $112
$125 $0 $0 $2 $6 $13 $23 $38 $56 $79 $107 $140
$150 $0 $0 $3 $8 $15 $28 $45 $67 $95 $129 $168
$200 $0 $1 $4 $10 $21 $37 $60 $90 $127 $172 $224
Year
Im
p
li
e
d
M
o
n
th
ly
A
S
P
Source: Goldman Sachs Global Investment Research
16 April 2026 24
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
1. 进群福利：进群即领2000+份汽车干货，直接打包下载
2. 每日分享：5+份精选汽车行业干货
3. 报告查找：群里直接咨询，免费协助查找
4. 不定期行业一线专家免费业务经验直播分享
5. 严禁广告：仅限行业报告交流，禁止一切无关信息
6. 严禁未经群里私加群友
3W AUTO汽车研习院——文库网站
电脑端访问体验更佳：
1. 读报告——
• 数万份汽车行业精选资料
• 重点关注电动化、智能化和全球化
• 涵盖技术、市场、供应链、组织与人才等15大主题
2. 查数据——
• 免费访问数万份研报数据截图
• 国产乘用车批发、零售、终端数据
• 国产乘用车出口、海关进出口、重点海外市场销量数据
• 重点品牌OTA数据等
3. 学课程——
• 合作机构和专家，座舱、战略等课程陆续上线
4. 找专家——
• 标准化专家访谈音频持续更新
• 定制化专家访谈需求匹配
• 长期招募一线业务骨干型专家
免责申明：
1. 本资料来源于网络公开渠
道，版权归属版权方；
2. 本资料仅限会员学习使用，
如他用请联系版权方；
3. 会员费用作为信息收集整
理及运营之必须费用；
4. 如侵犯您的合法权益，请
联系下方客服微信将及时
删除。
免费汽车干货群
扫码进群 长期有效
微信扫码 海量资料
汽车文库网站
微信扫码 合作联系
To better address this risk, we show an analysis of how much the US fleet size (currently
there are ~300 mn vehicles in operation) could theoretically change in 10 years using our
new AV forecast (we estimate there will be a fleet of about 3 mn AVs in 2035 on roads in
the US). Assuming various increases in utilization of these AVs compared to traditional
miles traveled (ranging from 25%-700%) and also potential changes in how many vehicle
miles are traveled (from 0% to 5% to capture a higher level of miles, especially from L3-5
vehicles), it suggests a negligible impact on the fleet size as a base case over the next 10
years (Exhibit 36).
Separately, we also show a scenario of what SAAR could be if the fleet was 100% shared
L4/5 AVs to meet miles traveled demand for the US population in the “What gets
disrupted” section of this report, and in that bear case scenario, SAAR could settle in the
10-13 mn range (with a smaller fleet that turns over every 2-3 years given the increase in
miles traveled per year) compared to the historical 16 mn per year average.
Exhibit 35: US AV revenue per mile (cost to consumer) vs
cost per mile of a personally owned vehicle
$
$
$
$
$
$
$
$
$
C
o
s
t
p
e
r
m
il
e
AV revenue per mile Average car cost per mile - 10k/year
Average car cost per mile - 15k/year
Source: AAA, Goldman Sachs Global Investment Research
Exhibit 36: Potential impact to base case 2035 US VIO from
increased utilization within the AV fleet
0% 1% 2% 3% 4% 5%
25% 0% 1% 2% 3% 4% 5%
50% 0% 1% 2% 3% 4% 5%
75% -1% 0% 1% 2% 3% 4%
100% -1% 0% 1% 2% 3% 4%
200% -2% -1% 0% 1% 2% 3%
300% -3% -2% -1% 0% 1% 2%
400% -4% -3% -2% -1% 0% 1%
500% -5% -4% -3% -2% -1% 0%
600% -6% -5% -4% -3% -2% -1%
700% -7% -6% -5% -4% -3% -2%
% increase in VMT
%
i
n
c
re
a
s
e
i
n
u
ti
li
z
a
ti
o
n
Source: Goldman Sachs Global Investment Research, US Bureau of Transportation
Statistics (BTS)
16 April 2026 25
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
AV trucking market is currently small but now poised to ramp
We now estimate that the AV trucking market in the US will reach $16 bn in 2030,
slightly down from $18 bn prior (as volume in 2026 is more likely to be filled with
upfitting than OEM lineside integration), and we introduce our 2035 forecast of
$105 bn. When including global markets, we estimate that the AV trucking market
in 2035 could be ~$560 bn. This is enabled by the US AV truck fleet growing from low
levels today, to ~24K in 2030, and ~161K in 2035, and a global fleet growing from a few
thousand today (primarily in China) to ~ mn as well as increasing miles per truck as
ODDs expand. AV trucking companies in North America, including Aurora, Kodiak,
Waabi, and Plus plan to scale commercially in 2026/2027, with Aurora expecting to
expand from 10 AV trucks at the end of 2025 to >200 by the end of 2026 (and we
assume to over a thousand starting in 2028).
We expect that part of the increased adoption and deployments through 2035 will be
driven by the cost of AV trucking improving relative to a human driver. At present, we
believe the additional upfront costs to make an AV truck in the US are
~$125K-$150K, and we expect this to decline to a premium of ~$35-40K in 2035,
driven by improved hardware costs and economies of scale.
Exhibit 37: Global AV trucking fleet by region (000s),
2023-2035E
Exhibit 38: Global AV trucking market size ($ mns),
2023-2035E
0
200
400
600
800
1,000
1,200
1,400
2
0
2
3
2
0
2
4
2
0
2
5
2
0
2
6
E
2
0
2
7
E
2
0
2
8
E
2
0
2
9
E
2
0
3
0
E
2
0
3
1
E
2
0
3
2
E
2
0
3
3
E
2
0
3
4
E
2
0
3
5
E
G
lo
b
a
l
A
V
t
ru
c
k
f
le
e
t
(0
0
0
s
)
US China Europe Others
0
100,000
200,000
300,000
400,000
500,000
600,000
2
0
2
3
2
0
2
4
2
0
2
5
2
0
2
6
E
2
0
2
7
E
2
0
2
8
E
2
0
2
9
E
2
0
3
0
E
2
0
3
1
E
2
0
3
2
E
2
0
3
3
E
2
0
3
4
E
2
0
3
5
E
G
lo
b
a
l
A
V
t
ru
c
k
m
a
rk
e
t
s
iz
e
(
$
m
n
s
)
US China Europe Others
Source: Company data, Goldman Sachs Global Investment Research
Source: Company data, Goldman Sachs Global Investment Research
Exhibit 39: US AV trucking market size ($ mns), new vs old,
2025-2030E
$0
$2,000
$4,000
$6,000
$8,000
$10,000
$12,000
$14,000
$16,000
$18,000
$20,000
2025 2026E 2027E 2028E 2029E 2030E
U
S
A
V
t
ru
c
k
in
g
m
a
rk
e
t
s
iz
e
(
$
m
n
s
)
New Old
Source: Company data, Goldman Sachs Global Investment Research
16 April 2026 26
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
On driving costs, when factoring in remote operations as well as expected savings from
fuel and, over the longer-term, insurance, as well as typical truck costs such as
maintenance, depreciation, and tolls, we expect the cost per mile for an AV truck to
decrease from ~$ in 2025 to ~$ in 2035. By comparison, we expect the
like-for-like cost for a human-driven truck (including tolls, maintenance, depreciation
costs, etc.) to increase from ~$ in 2025 to $, driven in part by rising driver
wages. Driver wages and benefits are currently about $ per mile, per ATRI.
Overall, we expect these improving cost dynamics and scaling up of the AV trucking fleet
to drive an increase in the global gross profit pool for AV trucking from close to zero in
2025 to ~$135 bn in 2035.
We show our expectations for COGS savings from an AV relative to a human-driven truck
in Exhibit 43. One of the key drivers of the improvement in AV cost per mile is an
increase in miles traveled per AV, as AVs aren’t subject to hours of use limits like humans
(. in the US, up to 11 hours of driving per day for human-driven trucks, with a break of
at least 30 minutes after 8 hours).
Exhibit 40: AV trucking illustrative COGS per mile,
2023-2035E
Exhibit 41: AV trucking gross profit margin, 2023-2035E
$
$
$
$
$
$
$
$
$
2
0
2
3
2
0
2
4
2
0
2
5
2
0
2
6
E
2
0
2
7
E
2
0
2
8
E
2
0
2
9
E
2
0
3
0
E
2
0
3
1
E
2
0
3
2
E
2
0
3
3
E
2
0
3
4
E
2
0
3
5
E
Il
lu
s
tr
a
ti
v
e
C
O
G
S
p
e
r
m
il
e
US Global
-160%
-140%
-120%
-100%
-80%
-60%
-40%
-20%
0%
20%
40%
60%
2
0
2
3
2
0
2
4
2
0
2
5
2
0
2
6
E
2
0
2
7
E
2
0
2
8
E
2
0
2
9
E
2
0
3
0
E
2
0
3
1
E
2
0
3
2
E
2
0
3
3
E
2
0
3
4
E
2
0
3
5
E
G
ro
s
s
p
ro
fi
t
m
a
rg
in
US Global
Source: ATRI, Company data, Goldman Sachs Global Investment Research
Source: Goldman Sachs Global Investment Research, Company data
Exhibit 42: AV trucking gross profit pool by region,
2023-2035E
-20,000
0
20,000
40,000
60,000
80,000
100,000
120,000
140,000
160,000
2
0
2
3
2
0
2
4
2
0
2
5
2
0
2
6
E
2
0
2
7
E
2
0
2
8
E
2
0
2
9
E
2
0
3
0
E
2
0
3
1
E
2
0
3
2
E
2
0
3
3
E
2
0
3
4
E
2
0
3
5
E
G
lo
b
a
l
A
V
t
ru
c
k
g
ro
s
s
p
ro
fi
t
p
o
o
l
($
m
n
s
)
US China Europe Others
Source: Company data, Goldman Sachs Global Investment Research
16 April 2026 27
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
For virtual driver models, similar to what Aurora plans to launch in partnership with
Aumovio, we expect costs in the US (including hardware depreciation, cost of remote
operation, etc.) to decline from >$1 in 2027 (when Aurora expects to launch with
Aumovio) to ~$ in 2035, as remote operators can supervise more vehicles and AV
hardware costs decline. We expect revenue per mile in the US to decrease from $ in
2027 to $ in 2035, driven by increased competition and better costs (implying solid
gross margin potential for a virtual driver). We estimate that the revenue potential for
virtual driver technology will increase from $ bn to ~$24 bn from 2027-2035
assuming all AV trucks in the US were to use a virtual driver model (although we believe
that not all AVs will operate with the virtual driver model). Across the global fleet, the
potential revenue under a similar assumption would be ~$116 bn in 2035, increasing
from a relatively small level in 2027.
Regulations
In terms of regulations for AV trucks, similar to robotaxi and passenger AVs, AV trucks
must also comply with the relevant FMVSS standards at the federal level, with
AV-specific regulations set at the state level. Presently, per Aurora, 40 states implicitly or
explicitly allow the deployment of driverless AVs including trucks.
The regulatory landscape for autonomous commercial vehicles in the EU draws from
international frameworks (Vienna Convention, UNECE) and EU instruments (AI Act,
General Safety Regulation) rather than harmonized legislation focused on autonomous
driving. These establish a type-approval framework and mandate rigorous active safety
features but currently limit widespread deployment of autonomy levels 4 (high driving
automation) and 5 (full driving automation); an automation scale defined by SAE
International. Specific regulations differ between nations. Germany and France have
enacted legislation permitting autonomous vehicles for road freight transport to operate
on public roads without a human driver as long as they are within defined operational
areas and with remote “technical supervisors”. Italy, the Netherlands, Finland, and
Sweden, among others, remain focused on controlled testing rather than commercial
deployment. We note that in controlled environments such as mines/mining, the
regulatory burden is lower than for public roads, allowing for specific private use
Exhibit 43: Incremental AV trucking cost (savings) per mile
vs human driven truck, 2025-2035E
-$
-$
$
$
$
$
$
$
$
$
2
0
2
5
2
0
2
6
E
2
0
2
7
E
2
0
2
8
E
2
0
2
9
E
2
0
3
0
E
2
0
3
1
E
2
0
3
2
E
2
0
3
3
E
2
0
3
4
E
2
0
3
5
E
In
c
re
m
e
n
ta
l
A
V
c
o
s
t
(s
a
v
in
g
s
)
p
e
r
m
il
e
v
s
h
u
m
a
n
US China Europe
Savings relative to human driven trucks are denoted by negative values, positive
values represent incremental cost
Source: Company data, ATRI, Goldman Sachs Global Investment Research
16 April 2026 28
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
deployment. Within the UK, the same Automated Vehicles Act 2024 that regulates
robotaxis/passenger AVs also regulates AV trucks.
For China, the rules for AV trucks are similar to AVs, with government permitting and
approvals for deployments in cities needed.
16 April 2026 29
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
What is at risk of disruption - human driven rideshare/trucking, and auto sales
While a portion of AV volumes will likely be incremental demand (. as new use
cases become affordable or possible), we believe autonomy could also disrupt
existing markets in the long-term. The two main vectors of risk relate to human-driven
taxi/rideshare and trucking, and lost light vehicle sales. Importantly, there is a
shortage of drivers in the US Class 8 trucking market currently (per ACT) and the
ATA projects a driver shortage over the coming decade, and we therefore think the
disruption may be more about lost opportunities for new employment as the driver
base turns over rather than direct job losses.
Fewer human-driven commercial miles in taxi/rideshare and trucking market
Per data from the US BLS, as of 2024, ~ mn people were employed as heavy and
tractor-trailer truck drivers in the US. The same data lists a median annual wage of
~$, implying annual labor costs of $128 bn.
Per data from the US BLS, as of 2024, ~448K people are employed as taxi drivers,
chauffeurs, or shuttle drivers in the US. The BLS lists a median annual wage of ~$,
implying annual labor costs of ~$16 bn. Additionally, the BLS estimates that ~ mn
people in the US are employed as delivery drivers with a median annual wage of $,
which equates to $65 bn in annual cost. We note that these datapoints do not include
rideshare drivers given many of these drivers work part-time, can work multiple jobs,
and don’t have a set annual salary. However, based on gross bookings in 2025 across
human driven rideshare in the US (~$64 bn) and estimated driver payouts excluding fees
(. commercial insurance and vehicle costs like fuel) of 40%-50% of gross bookings, we
estimate the total payments to US rideshare drivers in 2025 to be ~$30 bn.
We discuss the potential impact to Uber and Lyft, and how much disruption there
could be for human-driven ridehail, in the following section, “Analyzing the
Potential Impact of Robotaxis on Uber & Lyft”.
Lost vehicle sales: Downside scenario of ~3-6 mn lower US SAAR, although if
consumers opt to own AVs there may be no disruption to unit volumes
We show the potential implied US auto SAAR in a scenario (which is not our base case)
where all vehicles on the road are shared AVs in 2040 in Exhibit 44. We assume, for the
purposes of this illustrative bear case analysis, that vehicles have a useful life of ~400K
miles, that utilization in terms of miles traveled per vehicle per year rises given the
scenario that all vehicles in operation (VIO) are shared AVs, and that vehicle miles
traveled (or VMT) ranges from unchanged to 33% higher, as consumers may use more
vehicle transportation if it is less costly and doesn’t require a driver’s attention. Under
such a scenario, SAAR could be 3-6 mn units lower than the current steady state of ~16
mn. In such a scenario, assuming a mid to high single digit EBIT margin and average ASP
of ~$50K, the potential disruption to revenue could be $100-$300 bn, and lost EBIT
could be ~$5 bn to ~$25 bn.
16 April 2026 30
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
Exhibit 44: Implied US SAAR (in mns) in a scenario where the VIO/VMT is all shared AVs by
2040
0% 2% 4% 5% 10% 15% 20% 25% 33%
12,500 10 10 10 11 11 12 12 13 13
15,000 10 10 10 11 11 12 12 13 13
20,000 10 10 10 11 11 12 12 13 13
30,000 10 10 10 11 11 12 12 13 13
40,000 10 10 10 11 11 12 12 13 13
50,000 10 10 10 11 11 12 12 13 13
75,000 10 10 10 11 11 12 12 13 13
100,000 10 10 10 11 11 12 12 13 13
150,000 10 10 10 11 11 12 12 13 13
200,000 10 10 10 11 11 12 12 13 13
% increase in VMT
m
il
e
s
p
e
r
A
V
Source: Goldman Sachs Global Investment Research
16 April 2026 31
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
Analyzing the Potential Impact of Robotaxis on Uber & Lyft
We view the rise of autonomous vehicles as having the opportunity to drive
incremental industry growth, rather than be a zero-sum game. Within the rideshare
industry, several friction points exist which could be solved by robotaxis today and on a
longer-term time horizon (with lower prices - Exhibit 45). While autonomous vehicles
provide an alternative to human rideshare today, we do not view the AV market as being
fully disruptive of the industry (on the supply side), given a few key factors:
Operating zones. Despite Waymo expanding its commercial operating zone by more n
than 10x over the last four years, the company still operates on a small scale.
Relative to the square mileage of the top 25 metro markets (based on the Census
Bureau’s definition of metro markets & regions ranked by population), Waymo’s
operating zone today represents <5% of this opportunity (Exhibit 46).
Operational design & domain limitations. While AV technology and fleet n
capabilities continue to improve, scaling operations often takes time due to ongoing
city-mapping and ODD constraints. For example, Waymo announced Austin testing
in August 2023 and did not fully roll out to the public (through their partnership with
Uber) until early 2025. It is likely that in markets with more severe climates/regions
where weather conditions & other driving patterns fluctuate, the commercial launch
(and scaling) could take a longer amount of time.
Supply relative to human rideshare networks. By 2030, we expect that the total n
industry robotaxi fleet will reach 62,750 vehicles and further scale to 200,000+ by
2035. A hybrid network of robotaxis and human drivers will likely be needed to meet
trip demand (as we estimate that bn+ UCAN trips will be completed in 2030). For
context, there are 6 mn+ Uber driver MAUs today (per SensorTower) which
completed ~ bn trips during 2025. We acknowledge that MAUs are not active
daily and that utilization rates are often lower than robotaxis. In markets where
Waymo operates exclusively through its own distribution network, wait times are
often longer than traditional rideshare platforms (albeit have continued to improve
as markets mature).
16 April 2026 32
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
Longer-term, as autonomous fleets grow (both within existing and new markets),
we acknowledge that AVs could cannibalize a portion of rides that have historically
been completed by human drivers. Today, a majority of rideshare drivers complete
trips for supplemental income (. 94% of Lyft drivers drive less than 20 hours/week
(link)), but we believe that over time, the emergence of robotaxis could provide new
channels to generate supplemental income, including through the monetization of
personal assets (previously outlined here).
Robotaxi deployments to date have been concentrated in markets with large
populations, and we believe that in the medium term, gross bookings disrupted by
AVs will largely be tied to GBs generated in the top 20 metro markets. In existing
markets, Waymo reaches ~17% of the addressable US rider population today, and our
expected launch markets in the US are largely oriented around larger metro markets. To
help frame the US impact from larger markets, we would note that Uber highlighted
~30% of US rideshare gross bookings come from top 20 markets (including NYC).
Exhibit 45: Autonomous vehicles have the opportunity to
solve a number of key friction points...
%
Exhibit 46: ... but today, AV operations are limited to a set
operating zone (as seen at Waymo)
Waymo operating area (sq. miles)
4%
13%
15%
20%
20%
24%
29%
34%
45%
0% 10% 20% 30% 40% 50%
Other
Downloaded too many apps
Not enough green rides
Safety
Not enough autonomous / driverless cars
Not enough competition
Drivers cancel on me too often
Wait times too long
Prices
180
272
433
900
1,215
%
%
%
%
%
%
0
200
400
600
800
1,000
1,200
1,400
2022 2023 2024 2025 2026 YTD
PHX SF LA MIA ORL ATL DAL AUS HOU SA NASH % Coverage
Consumer survey of biggest issues with rideshare in 2026
Source: Obi, Data compiled by Goldman Sachs Global Investment Research
As of 4/7/2026. % coverage represents Waymo operating zone as % of top 25
metro market zones
Source: Company data, Census Bureau, Goldman Sachs Global Investment
Research
16 April 2026 33
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
In terms of outlining the potential disruption of rideshare gross bookings derived from
human drivers, we frame three potential scenarios:
Base case. Expect autonomous trips to continue to scale at a meaningful rate - n
albeit remaining a minority of the overall GB mix within top 20 cities, given that
consumer habits evolve slowly, and the ramp within cities (in terms of overall mix)
takes time to scale. For example, according to 3P data sources, Waymo’s market
share reached ~30% (per Yipit) in San Francisco after 20 months of being fully
launched.
Bear case. Over the next five years, assumes that autonomous rides will become a n
majority of top 20 city rideshare bookings and that 100% of trips completed by a
robotaxi will disrupt trips historically taken by human drivers. These operating
assumptions would imply that market share dynamics across all operating zones
more closely align with trends seen in San Francisco by 2029.
Bull case. Frames the case that the overall industry will be accretive to overall n
rideshare industry growth rates, with 60% of AV GBs driving incremental growth.
Given that each market is unique (both in terms of user behavior and adoption
trends, etc.), our bull case assumes that disruption trends seen in other markets
(Waymo and Lyft in San Francisco) are unique, as each market differs in terms of
consumer adoption and market share trends.
We believe that autonomous vehicles will be accretive on a longer-term basis, but in
the medium-term, could be cannibalistic to the overall industry (based on our
forecasts today). Our current forecast implies that by 2030, trips completed by AVs will
become more affordable in comparison to rides completed by human drivers, which we
Exhibit 47: Metro cities where Waymo currently operates account for 17% of the addressable
rideshare user opportunity
%
17%
29%
0%
5%
10%
15%
20%
25%
30%
35%
P
H
X S
F LA
A
U
S
A
TL M
IA
N
A
S
H
D
A
L
H
O
U
S
A
O
R
L
To
ta
l
LV
TP
A
S
D
D
C
N
O B
A
P
G
H
S
TL
D
E
N
P
H
I
D
T
M
N
To
ta
l
Existing Markets Expected Launch Markets
Addressable population % based on metro population / total addressable users
Source: Census Bureau, Goldman Sachs Global Investment Research
16 April 2026 34
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
believe will uplift overall industry growth rates. Over the next 10 years, elements of
affordability could lead to increased: a) penetration into the addressable rider market; &
b) trip frequency by riders who have historically used rideshare platforms for select use
cases.
The disruption of bookings generated by human drivers does not necessarily mean
that traditional rideshare companies will be fully impacted. In fact, we believe that
Uber and Lyft are well positioned to play a role in the AV industry given a few key factors:
Hybrid networks drive a seamless consumer experience. Supply availability is n
crucial to maintain appropriate demand service levels/ETAs, and we are of the view
that hybrid networks that combine AVs and human drivers will produce the best
consumer experience for riders and the highest utilization rates for ridesharing
vehicles. Specifically, in markets where Waymo has deployed its fleet on the Uber
app, the average trips per vehicle per day (TpVD) is 30% higher relative to other
markets (. Los Angeles, where Waymo offers rides exclusively through the Waymo
app).
Building a DTC rideshare network is capital intensive. Economic incentives matter, n
and building on-demand local networks requires significant capital investments. As
one measure of the losses incurred to reach scale, we note that UBER’s accumulated
deficit (negative cumulative retained earnings) peaked at $(33)bn in 2022. This is
with UBER adopting an asset-light approach and not owning the supply outright;
UBER also did not need to invest to compete/displace a scaled mobile-first
incumbent in on-demand personal mobility (as is the case for challengers today).
Investments & partnerships made within the AV industry highlight a n
commitment toward scaling exposure to autonomous vehicles. Uber and Lyft
have announced a number of partnerships aimed at scaling both their AV fleet
(which will be primarily owned and operated by third parties). We would also note
that Uber will continue to make investments toward driving an improved and
enhanced suite of services & capabilities aimed at building and commercializing AVs
Exhibit 48: Our bear case implies that AVs cannibalization
could reach $19 bn by 2030, but potentially be additive to
the industry overall in a bull case scenario...
$mn
Exhibit 49: ... which would imply at worse 16%
cannibalization, with a minimal base case impact (to
human rideshare GBs) implied, with potential upside
optionality in an all boats rising scenario
%
$(20,000)
$(15,000)
$(10,000)
$(5,000)
$-
$5,000
2025 2026E 2027E 2028E 2029E 2030E
Base Case Bear Case Bull Case
-20%
-15%
-10%
-5%
0%
5%
2025 2026E 2027E 2028E 2029E 2030E
Base Case Bear Case Bull Case
Source: Goldman Sachs Global Investment Research
Source: Goldman Sachs Global Investment Research
16 April 2026 35
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
(previously outlined here).
On the back of these factors, we would frame both Uber and Lyft as being well
positioned to capitalize on the growing AV opportunity, and primarily operate as an
asset-light-third-party marketplace (. a demand aggregator) for AV fleet operators
to plug their supply into as a way to generate demand. Relative to Waymo (which
accounts for a majority of AV rideshare bookings today), we believe that Uber and Lyft
will act primarily as AV facilitators.
In our view, the key unlock toward driving AV bookings onto Uber/Lyft will be from
scaling the addressable fleet outside of Waymo/Tesla. Our forecast implies that
>90% of the US robotaxi fleet today is associated with Waymo/Tesla. As AV fleets
outside of the two operators ramp over the next 5 years, we expect a large portion of
the “other” bucket (Exhibit 51) to offer rides through traditional rideshare platforms.
Demand generation by platforms which do not have a first mover advantage or other
elements of differentiation could have difficulties scaling a DTC offering, leading to these
players placing their operations onto Uber/Lyft.
Our fleet forecast implies that 30%+ of the US AV fleet will be from parties outside
of Waymo/Tesla in 2030. On the back of the addressable fleet scaling and AV
partnerships already announced, we believe that the two platforms could help facilitate
30%+ of the total US AV rideshare bookings by 2030. Our forecasts could prove to be
conservative as we do not make assumptions surrounding additional partnerships
between Waymo and Uber/Lyft.
Exhibit 50: Uber and Lyft AV Partnerships
As of April 2026
Source: Company data, Data compiled by Goldman Sachs Global Investment Research
16 April 2026 36
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
Exhibit 51: We believe that players outside of
Tesla/Waymo will emerge in the US...
Addressable commercial fleet (US)
Exhibit 52: ... and drive market share gains for Uber/Lyft
over time
Non Uber/Lyft AV market share %
2,032
12,532
38,350
91,750
156,250
208,638
0
50,000
100,000
150,000
200,000
250,000
2025 2026 2027 2028E 2029E 2030E 2031E 2032E 2033E 2034E 2035E
Waymo Tesla Other
89%
62%
45%
67%
50%
75%
60%
0%
20%
40%
60%
80%
100%
120%
2025 2026E 2027E 2028E 2029E 2030E 2031E 2032E 2033E 2034E 2035E
Bull Base Bear
Source: Goldman Sachs Global Investment Research
Source: Goldman Sachs Global Investment Research
16 April 2026 37
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
Spotlight on Waymo: Framing Operating Assumptions and the Path to $8 bn+ Gross Bookings by
2030E
Waymo initially launched in Phoenix during 2018 and began serving fully
autonomous commercial rides in 2019. Since their launch, the company now operates
in 11 cities across the US, and expects to lay the groundwork in 20 markets this year
(link). Within these markets, Waymo serves over 500k paid rides weekly (as of March
2026), a rate which has more than doubled since April 2025. This ramp resulted in over
15 mn rides being completed during 2025, serviced by a fleet of >1,800 vehicles (GSe).
While Waymo expects to lay the groundwork for ride-hailing operations in 20+
cities during 2026, our assumptions around launch timing remains relatively
conservative given the level of predictability/visibility surrounding the regulatory
environment. For example, while self-driving cars can be tested in Washington ., the
vehicles cannot complete rides in a fully autonomous manner (link). In terms of our
assumptions surrounding specific launch markets, the company has announced a
number of test cities across US and international markets - all of which vary in terms of
traffic patterns, weather/climate and rideshare demand trends. Our current launch
assumptions revolve around markets which have: a) a top 50 MSA (Metropolitan
Statistical Area) by population; & b) testing currently underway.
By the end of 2026, we expect Waymo will have commercial operations live in 15
markets and ramp to 49 markets globally by 2030. For the purposes of our analysis,
we assume that all future city launches (excl. Tokyo) will be offered exclusively through
the Waymo One service. Additionally, relative to our previous report framing the Waymo
opportunity (link), we take the opportunity to adjust our trip and fleet assumptions to be
an annualized estimate rather than run-rate values.
Exhibit 53: Waymo Major Announcements Timeline
Source: Company data, Goldman Sachs Global Investment Research
16 April 2026 38
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
http://%C2%A0Additionally, relative to our previous report framing the Waymo opportunity (link), we take the opportunity to adjust our trip and fleet assumptions to be an annualized estimate rather than run-rate values.
We see a clear pathway for the platform to reach an annualized rate of 975K paid
weekly rides in 2026, which implies an exit rate well ahead of the company’s goal (1
mn paid weekly rides by the end of 2026). Outside of expanding into new markets, our
forecasts assume that Waymo will scale its operating fleet (while also improving
utilization rates) in existing markets, given the likely expansion of operating zones.
Fleet Assumptions
Based on prior disclosures, Waymo’s active fleet is >3,000 vehicles (March 2026), which
we expect to reach 5,000, on average, in 2026. The company has publicly disclosed
plans to build over 2,000 more vehicles through 2026, including Jaguar I-PACE AVs and
building partnerships with OEMs including Zeekr and Toyota to drive further vehicle
deployments. By 2030, we expect Waymo’s average total fleet to reach 30,000+ vehicles
in 2030, and 95,000+ by 2035, with growth coming both from: a) the expansion of fleets
within existing markets; & b) introducing fleets into new markets.
Exhibit 54: Expect Waymo to reach 75 cities globally by 2035
City ramp
Waymo Ramp Operations Type Launch Year Key
(1) Phoenix 2020 Waymo One Exclusive
(2) San Francisco 2023 Uber + Waymo One
(3) Los Angeles 2024 Uber Exclusive
(4) Austin 2025 Lyft + Waymo One
(5) Atlanta 2025 GO + Waymo One (TBD)
(6) Miami 2026
(7) Dallas 2026
(8) Houston 2026
(9) San Antonio 2026
(10) Orlando 2026
(11) Nashville* 2026
(12) London 2026
(13) Las Vegas 2026
(14) Tampa 2026
(15) San Diego 2026
(16) Washington DC 2027
(17) New Orleans 2027
(18) Baltimore 2027
(19) Pittsburgh 2027
(20) St. Louis 2027
(21) Denver 2027
(22) Philadelphia 2027
(23) Detroit 2027
(24) Minneapolis 2027
(25) Tokyo** 2027
City (26) - City (33) 2028
City (34) - City (41) 2029
City (42) - City (49) 2030
City (50) - City (55) 2031
City (56) - City (60) 2032
City (61) - City (65) 2033
City (66) - City (70) 2034
City (71) - City (75) 2035
*Nashville available through Waymo One App today (April 2025)
**Tokyo access method unspecified
Source: Company data, Goldman Sachs Global Investment Research
16 April 2026 39
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
In the US, there have been meaningful operational improvements over the last 3
years, and we estimate that the average trips per vehicle per day has improved from
9 in 2023 to 24 in 2025. Even as we forecast the company scaling to complete 40 TpVD
in the US, we still expect that Waymo’s total US fleet will grow to 25,000+ vehicles by
2030 to meet rider demand.
Given the expected launch of London in 2026, we also take the opportunity to frame
our fleet assumptions in international markets. The expansion would come 7 years
after Waymo’s initial commercial launch in Phoenix. In our view, international markets
could scale in a similar (albeit, likely slower, given the limited data) manner relative to US
markets. Specifically, we expect that utilization rates within international markets will lag
behind what the US has historically achieved by 5-6 years (Exhibit 57).
As international markets are introduced in 2026/2027+, a growing percentage of
Waymo’s total paid rides will come from these markets, and we forecast international
trip volume to account for 13% of total trips in 2030 (and contribute $1 bn in gross
bookings). Based on the TpVD rates framed below, this would imply that Waymo’s
international fleet will reach an average of 6,000+ vehicles by 2030.
Exhibit 55: Over the last 3 years, we believe that Waymo
has become more efficient at vehicle deployments &
increasing utilization rates...
City TpVD
Exhibit 56: ... but even with these improvements, there is
still a need to scale fleets both within existing & new
markets
US fleet (GSe)
9
15
24
0
5
10
15
20
25
30
2023 2024 2025
PHX SF LA AUS ATL Total
232 869
1,813
5,079
9,329
13,852
19,145
25,910
0%
50%
100%
150%
200%
250%
300%
0
5,000
10,000
15,000
20,000
25,000
30,000
2023 2024 2025 2026E 2027E 2028E 2029E 2030E
Existing New YoY Growth
Source: Company data, Goldman Sachs Global Investment Research
Existing markets defined as markets where Waymo has commercial operations
(as of April 2026)
Source: Company data, Goldman Sachs Global Investment Research
16 April 2026 40
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
Trip Assumptions
On the back of growing fleet supply (both within existing and new markets), we
model Waymo reaching 960K annualized US paid weekly trips during 2026. Our
growth assumptions are anchored around the company: a) expanding its operating
zones in existing markets; b) launching operations in new regions; & c) increasing rider
frequency among existing riders. In recent years, particularly in markets such as Phoenix
and San Francisco, Waymo has successfully expanded its service area (Exhibit 46) - and
we believe that Waymo’s operating zone will continue to expand in the coming years.
In addition to expanded operating zones, a meaningful unlock toward driving trip
volume, rider adoption, and gross bookings is anchored around airport trips. Airport
trips compose a meaningful portion of rideshare gross bookings today - for example,
during 2025, Uber generated 15% of total mobility GBs from trips that started at, or
were completed at an airport.
Over the next 5-10 years, with the broader AV industry expected to offer riders more
affordable trips (lower ASPs on a per mile basis) in comparison to human drivers, the
industry is likely to see both new riders come onto the various platforms, and existing
riders increase the frequency of their rides.
Exhibit 57: Expect international market utilization rates to
lag the US, but improve over time...
Regional TpVD
Exhibit 58: ... and Waymo’s international fleet to reach
6,000+ by 2030
Intl fleet
E 2035E
9
30
40
5
21
35
0
5
10
15
20
25
30
35
40
45
20
23
20
24
20
25
20
26
E
20
27
E
20
28
E
20
29
E
20
30
E
20
31
E
20
32
E
20
33
E
20
34
E
20
35
E
US Intl
429 1,500
2,387
4,432
6,874
9,634
14,468
20,408
27,053
33,022
0%
50%
100%
150%
200%
250%
300%
0
5,000
10,000
15,000
20,000
25,000
30,000
35,000
2026E 2027E 2028E 2029E 2030E 2031E 2032E 2033E 2034E 2035E
Intl YoY Growth
Source: Company data, Goldman Sachs Global Investment Research
Source: Goldman Sachs Global Investment Research
16 April 2026 41
Goldman Sachs Global AVs
c4
5a
43
53
0f
60
4d
12
bc
b9
a8
2b
5a
a6
b9
f6
【价值目录】网整理：
While visibility into the timing (and launch) of new markets remains low, we believe
that expanding into international markets will play a key role in driving trip growth.
In terms of the regulatory landscape for autonomous commercial vehicles - international
frameworks (incl. the Vienna Convention, UNECE) and EU instruments (AI Act, General
Safety Regulation) have a type-approval framework and mandate rigorous active safety
features, which could limit widespread deployment of L4/L5 autonomy. In our view, this
will likely limit the near-term scaling of Waymo in international markets over the next 5
years (. we expect international markets to account for 10% of total Waymo trips in
2029). That said, by 2035, we expect for international trips to account for ~30% of
Waymo’s total paid trip volume.
Longer-term, we view international markets to be a stimulant to trip growth, and in the
near-term, we see the announced markets as having the potential to increase awareness
in robotaxis. Specifically in tourist destinations such as London and Tokyo,
tourism/novelty effects could remain (previously highlighted in Exhibit 24) and also drive
awareness of the service and eventual adoption as Waymo expands into new markets.
Exhibit 59: Relative to 2025, we expect Waymo to triple its US paid trips in 2026, and for this
to further double in 2027 from the 2026 level
Annualized weekly paid trips (US, 000s)
15 93 300
960
1,959
3,297
5,026
7,074
9,206
11,343
13,286
15,300
17,594
0
2,000
4,000
6,000
8,000
10,000
12,000
14,000
16,000
18,000
20,000
20
23
20
24
20
25
20
26
E
20
27
E
20
28
E
20
29
E
20
30
E
20
31
E
20
32
E
20
33
E
20
34
E
20
35
E
Existing New
May:
10
May:
50
Aug:
100
Oct:
150
Dec:
175
Feb:
200
Apr:
250
Dec:
400
Mar:
500
Company Target:
1,000
(Year end, US + intl)
Existing markets defined as US markets which already have launched commercial operations (April 2026). Blue boxes are paid
weekly trip run-rates reported by Waymo
Source: Company data, Goldman Sachs Global Investment Research
16 April 2026 42
Goldman Sachs Global AVs
c4
5a
43
53
0
