Tracking AI

in 10 charts

Marketing materials for

professional investors only,
not for onward distribution
Tracking AI
in 10 charts

Contents
Global AI 4
Expanding AI 5
Competitive AI 6
Economic AI 7
Testing AI 8
Prioritizing AI 9
Scaling AI 10
Cooling AI 11
Robotic AI 12
Boring AI 13

April 2025
Foreword

Artificial Intelligence (AI) is revolutionizing industries and reshaping the future of technology.
Even as stocks have faltered so far in 2025, global competition has spurred continued
investments into new AI models, digital infrastructure, new applications. Notably, speaking at
the World Economic Forum in January, Demis Hassabis, CEO of Google DeepMind said the
company was on track to submit AI-designed drug candidates into clinical trials this year. More mundanely,
but no less impactfully, AI agents are poised to shift from merely answering questions to taking action, as
the technology seeks to fulfill its productivity-enhancing promise.

As we delve into the intricacies of AI through ten insightful charts, we aim to provide a comprehensive
overview of its current state and future potential. The data presented in these charts not only underscores
the transformative power of AI but also serves as a guide for strategic investment decisions. As AI
continues to advance, it is crucial for professional investors to stay informed about its trajectory and
implications. This collection of charts provides valuable insights into the dynamic world of AI, offering a
clear understanding of its current landscape and future potential.

Chart collection by portfolio managers Daniel Ernst and Sam Brasser

3 Robeco | Tracking AI in 10 charts

 10
09 08 07 06 05 04 03 02 01

Global AI
Figure 1: Notable AI models by country Over the last few years there has been a
1,OE+27 Cambrian explosion in AI model
Grok 3
development. Although OpenAI’s ChatGPT
was among the first to market, they were
not alone. Researchers at Epoch AI have
1,OE+27 tracked more than 200 notable AI models
Training FLOPs log scale

Claude 3.7 Claude 3.7 trained with at least 100 billion trillion (1023)
GPT 4 Doubao pro
floating point operations (FLOPs). While
1,OE+27 DeepSeek captured the market’s attention
Uama 3.3
Deepseek v3 earlier this year, there are dozens of
competing AI models in China including
Yi Large Doubao, the model created by TikTok
1,OE+27 GPT 3.5
parent ByteDance, and Yi, released by AI
Alpha Go Zero
pioneer Kai-Fu Lee’s 01.AI. Similarly in the
US, AI models released by large tech firms
1,OE+27 like Alphabet and Meta compete with
6-3-2017 19-7-2018 1-12-2019 14-4-2021 27-8-2022 9-1-2024 23-5-2025 startups from Anthropic’s Claude to X.AI’s
Grok. Looking ahead, the AI race is shifting
China Finland France Germany Japan Korea UAE UK US
from base models toward applications that
Source: Epoch AI, March 2025 make use of this emerging technology.

4 Robeco | Tracking AI in 10 charts

 09
10 08 07 06 05 04 03 02 01

Expanding AI
“Every cloud has a silicon lining” might be an Figure 2: Cloud computing capital spending trends
apt update to the well-known adage. While 400
investors had been concerned that the pace of
investment into AI infrastructure might slow 350
down, during earnings calls in the first months
300
of 2025, the largest cloud computing providers
raised their outlooks for capital spending this 250
USD billions

year. The message from Alphabet, Amazon,

and Microsoft was consistent, with each 200
company noting that demand for AI and cloud
150
computing services outstripped their capacity
to provide them. Morgan Stanley estimates 100
that global cloud computing capital spending
50
will reach USD 353 billion in 2025, a figure
that is nearly USD 38 billion greater than their 0
previous estimate. Those investments should 2023 2024 2025 2026
provide continued support for a wide range of
Amazon AWS Alphabet Meta Microsoft
technology suppliers from Nvidia’s AI
accelerators to SK Hynix’s memory chips. Source: Morgan Stanley, February 2025

5 Robeco | Tracking AI in 10 charts

 08
10 09 07 06 05 04 03 02 01

Competitive AI
According to a survey by Application Program Interface (API) developer Kong, AI
model usage within the enterprise is already making use of the diverse set of model
options. Unsurprisingly, OpenAI’s GPT models lead the pack, but with 27% share
among respondents, it is not the dominant choice. Competitive models from large
technology platforms are not far behind with Microsoft’s Azure AI taking an 18% share
and Google’s Gemini holding 17% of the market so far. With AI model development
moving rapidly, it is likely shifts in market share are still to come.

Figure 3: Enterprise IT adoption of AI models

OpenAI GPT 27%

Microsoft Azure AI 18%
Google Gemini 17%
Meta Llama 8%
Amazon Bedrock 7%
Claude 7%
Cohere 5%
Mistral 4%
Other 4%
None 3%

Source: Kong, September 2024

6 Robeco | Tracking AI in 10 charts

 07
10 09 08 06 05 04 03 02 01

Economic AI
Since DeepSeek's launch in January 2025, the market has been driven by a combination of advances in both software and
concerned that lower cost technology would undermine the semiconductors. Although the price of AI-dependent graphical
economics of AI models produced by US technology leaders. processor units (GPUs) has risen, the amount of data those
Although subsequent research has found DeepSeek’s cost semiconductors process has jumped exponentially. As a result,
advantage was highly exaggerated, the trend toward the price per floating point operation per second of an Nvidia
increasingly efficient technology is an industry standard. For GPU has also fallen by more than 99% in the last ten years.
instance, over a 30-year period, the cost of a transistor, a solar While the chart of GPU price performance below appears to
module, and a lithium-ion battery each fell by more than 99%. show slowing progress, even in the last two years the cost of a
The cost of sequencing DNA fell by more than 99% in just a single calculation has fallen by nearly 75%.
decade between 2003 and 2013. For AI, price improvements are

Figure 4: Price per Nvidia GPU floating point operation per second
USD 0.70

USD 0.60

USD 0.50

USD 0.40

USD 0.30

USD 0.20

USD 0.10

USD 0.00
2015 2016 2017 2018 2019 2020 2021 2022 2023 2024

Source: Epoch AI, February 2025

7 Robeco | Tracking AI in 10 charts

 06
10 09 08 07 05 04 03 02 01

Testing AI
A key goal for AI is to achieve what researchers call Figure 5: OpenAI model score on ARC AGI test
‘artificial general intelligence’ (AGI). While there is some 100%
debate as to what that means, the ARC Foundation
90%
(which developed an AGI test) defines the concept as
“a system capable of efficiently acquiring new skills and 80%
solving novel problems for which it was neither explicitly 70%
designed nor trained.” Until last year, most AI models
60%
scored poorly on ARC’s AGI test, but that changed with
the release of OpenAI’s so-called reasoning models. In 50%
September 2024, Open AI’s o1 preview model scored 40%
21% on the AGI test, and in December a highly tuned
version of their o3 model scored 87.5%. That 30%
performance comes at a price, according to ARC, the 20%
cost per task of these highly tuned reasoning models
10%
approaches USD 20. However, in January DeepSeek
released a model that while only scoring 20.5%, 0%
2019 2020 2021 2022 2023 2024
apparently achieved that result for a mere USD 0.05
per task. Source: ARC Foundation, December 2024

8 Robeco | Tracking AI in 10 charts

 05
10 09 08 07 06 04 03 02 01

Prioritizing AI
Amid continued questions concerning the Figure 6: CIO prioritization of AI/ML project spend growth
return on investment of AI spending, 18%
corporate chief information officers (CIOs)
are putting the technology at the top their 16%
priority lists. According to Morgan Stanley’s
14%
4Q24 CIO survey, 16% of CIOs cite AI as the
segment receiving the greatest budget 12%
increase in 2025. The survey also noted
that security concerns remain top of mind, 10%
with 12% of CIOs planning to allocate the 8%
largest budget increase to the segment this
year. Within AI, applications that enable AI 6%
to take action rather than merely answer
4%
questions are showing promise. These AI
agents are working their way into existing 2%
software to automate tasks that often
involve data entry or the manual process of 0%
2018 2019 2020 2021 2022 2023 2024
integrating information from multiple
applications. Source: Morgan Stanley, January 2025

9 Robeco | Tracking AI in 10 charts

 04
10 09 08 07 06 05 03 02 01

Scaling AI
One of the biggest questions facing AI-related companies is the their Now Assist agent, and Workday reporting that over 30% of
return on investment. Beyond providers of foundational customer expansion deals included an AI solution. Encouragingly,
technologies like Nvidia’s AI accelerators, and cloud computing it’s not only larger tech companies leveraging the technology,
services from Amazon, Google and Microsoft, data on AI revenue but also startups. According to the payments processor Stripe,
has been inconsistent. Established enterprise software providers in 2024, the median time for AI startups to reach annualized
have noted rising adoption, with ServiceNow for instance revenues of USD 5 million was 24 months, compared to 37
reporting that 12% of its customer base had begun to employ months for Software as a Service (SaaS) startups in 2018.

Figure 7: SaaS and AI startups’ median time to annualized revenue milestone

5
24 37
Annualized revenue (USD mln)

11
1
15

0
0 5 10 15 20 25 30 35 40
Months
Top 100 AI companies (by revenue) on Stripe 2024 Top 100 SaaS companies (by revenue) on Stripe 2018

Source: Stripe, February 2025. The value of your investments may fluctuate. Past results are no guarantee
of future performance.

10 Robeco | Tracking AI in 10 charts

 03
10 09 08 07 06 05 04 02 01

Cooling AI
Data centers use both air and liquid cooling to Figure 8: Worldwide revenues of data center cooling methods
prevent overheating and optimize server 18 35%
performance. But changes in server cooling
16
are needed considering how AI servers have 30%
much higher thermal density. Moreover, these 14
25%
AI servers have higher power density, uneven 12
heat distribution and substantial load 20%
USD billion
10
variations. Air cooling methods are not
adequate anymore and a shift to liquid cooling 8 15%
is underway. As AI computing usage increases, 6
the demand for liquid cooling has surged 10%
4
substantially. Technology consultancy firm
5%
Omdia forecasts that worldwide data center 2
cooling sales will double from USD 8 billion in 0 0%
2024 to USD 16 billion in 2028. The share of 2021 2022 2023 2024 2025E 2026E 2027E 2028E
liquid cooling is forecast to increase from 17%
Air cooling Liquid cooling Share of liquid cooling (RHS)
in 2024 to 33% in 2028.
Source: Omdia, October 2024. The value of your investments may fluctuate. Past results are
no guarantee of future performance.

11 Robeco | Tracking AI in 10 charts

 02
10 09 08 07 06 05 04 03 01

Robotic AI
While the steady growth of industrial robotics produces an advancements in mechanical engineering are enabling more
unexciting chart, behind-the-scenes innovations tell another dexterous and more mobile robotics. CB Insights reports that
story. Over the last five years, the global installed base of the venture investment into industrial humanoid robotics tripled
industrial robots has expanded at an 11.3% CAGR reaching in 2024 to USD 1.2 billion. With these trends in mind, Nvidia CEO
nearly 4.7 million in 2024. Robots are not only getting more Jensen Huang, speaking at the Consumer Electronics Show in
plentiful, they are getting smarter too. For instance Google's January, declared robotics as the next ‘multi-trillion-dollar’
DeepMind has built a model that enables robots to learn on the opportunity.
job rather than follow pre-programmed instructions. Further,

Figure 9: Industrial robot installations and operational installed base

700 7,000

600 6,000

500 5,000

400 4,000

300 3,000

200 2,000

100 1,000

0 0
2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025E 2026E 2027E

Robot installations Operational robots

Source: International Federation of Robotics, October 2024

12 Robeco | Tracking AI in 10 charts

 01
10 09 08 07 06 05 04 03 02

Boring AI
In business there is an old joke of a consultant Figure 10: Leading use cases for AI in the workplace
who borrows your watch to tell you the time,
something you could have easily done Financial planning
yourself. Workplace AI tools appear to be Risk management
following the same playbook. For all the Predictive behavior
excitement about what the future of AI may Brainstorming
hold, and for all the fear about what risks the
Fraud detection
technology may pose, AI today is very often
assisting rather mundane tasks. According to a Email generation
recent survey published by the Wharton School Presentation creation
of Business, weekly users of AI tools in the Customer/competitor research
workplace rose from 37% in 2023 to 73% in Support/help desk
2024. At 64%, the most popular use case for AI
Marketing content
in the workplace is the writing and editing of
documents or proposals. Building on the Document/meeting summaries
previously popular buzzword, 'big data', 62% of Data analysis
survey respondents reporting using AI to help Document writing/editing
with data analysis. Notably, 94% of purchasing
0% 10% 20% 30% 40% 50% 60% 70%
managers reported using AI in 2024, up from
50% last year. Source: Wharton University, October 2024

13 Robeco | Tracking AI in 10 charts

STRATEGIES

Robeco's AI-related strategies

INVESTING IN AI COMPANIES: INVESTING WITH THE HELP OF AI:

Digital Innovations strategy Dynamic Theme Machine

AI is eliminating slack, accelerating productivity, and reducing The Dynamic Theme Machine is a quantitative thematic
costs in diverse sectors. Robeco's Digital Innovations strategy strategy that uses artificial intelligence to detect and invest in
capitalizes on the human ingenuity and technological emerging and established themes. The strategy seeks to
innovation that are transforming business and reshaping the capitalize on long-term trends, themes, or macroeconomic
world around us. It seeks to capture the value creation potential developments that are expected to impact the economy,
derived from companies supplying foundational technology society, and the environment. It rotates through themes,
and enabling the modernization of enterprise and industry. buying attractive ones and selling those in decline.

The investment universe spans a broad range of sectors Some of the leading themes (as per Q1 2025) in this strategy:
such as:
• Digital advertising and media streaming
• Semiconductors • AI and cloud technologies
• computing hardware • Cancer treatments
• Cloud infrastructure • Mental health and neurology
• cybersecurity • Biological research and technology
• Software
• Robotics
• Precision engineering
• Information technology services

14 Robeco | Tracking AI in 10 charts

IMPORTANT INFORMATION

This information refers only to general information about

Robeco Holding B.V. and/or its related, affiliated and
subsidiary companies, (“Robeco”), Robeco’s approach,
strategies and capabilities. This document is solely
intended for professional investors, defined as investors
qualifying as professional clients, who have requested to
be treated as professional clients or who are authorized to
receive such information under any applicable laws. Unless
otherwise stated, the data and information reported is
sourced from Robeco, is, to the best knowledge of Robeco,
accurate at the time of publication and comes without any
warranties of any kind. Any opinion expressed is solely
Robeco’s opinion, it is not a factual statement, and is
subject to change, and in no way constitutes investment
advice. This document is intended only to provide an
overview of Robeco’s approach and strategies. It is not a
substitute for a prospectus or any other legal document
concerning any specific financial instrument. The data,
information, and opinions contained herein do not
constitute and, under no circumstances, may be construed
as an offer or an invitation or a recommendation to make
investments or divestments or a solicitation to buy, sell, or
subscribe for financial instruments or as financial, legal,
tax, or investment research advice or as an invitation or to
make any other use of it. All rights relating to the
information in this document are and will remain the
property of Robeco. This material may not be copied or
used with the public. No part of this document may be
reproduced, or published in any form or by any means
without Robeco’s prior written permission

15 Robeco | Tracking AI in 10 charts

For more information
please visit our website