The data is in, and in 2025, global funding to AI companies surged to $225.8 billion, nearly doubling the $114.4 billion raised in 2024,. Perhaps more telling than the raw dollar amount is the market share: AI companies captured 48% of all venture funding in 2025, the highest percentage on record.
If 2023 was the year of the awakening and 2024 was the year of experimentation, 2025 will be remembered as the year AI swallowed the venture capital industry whole. One graph that stood out for me in the State of Startups 2025 report by Carta, is the share of AI companies out of various categories.
We have reached a tipping point where AI is no longer a vertical on a landscape map; it is the horizontal layer powering the entire startup ecosystem. From the “Smart Money” VCs focusing their portfolios almost exclusively on AI to the massive capital expenditures of hyperscalers, the message is clear: investors are betting on a winner-take-all dynamic where few other sectors matter.
Here is a breakdown of where that capital went in 2025, from the massive foundation models to the agents changing how we work.
1. Foundation Models: The Era of Megarounds
The foundation model layer remains the most capital-intensive segment of the stack. In 2025, foundation model companies raised $80 billion, representing 40% of all global AI funding. This is an area almost exclusively reserved for mega funds and strategics doing mega rounds (xAI’s recent $20 billion round or Anthropic’s recent $10 billion round at $350 billion valuations are good examples of this).
This capital is heavily concentrated at the top. The two dominant players, OpenAI and Anthropic, alone captured 14% of global venture investment this year. OpenAI closed the year valued at $500 billion, making it the most valuable private company of all time, while Anthropic commanded a $350 billion valuation.
The dynamic here is clear: scale requires capital. In AI, 58% of funding came from megarounds of $500 million or more. As we look toward 2026, the question remains whether equity markets can sustain this voracious appetite for compute, or if partnerships will be required to bridge the gap.
2. Infrastructure: The Physical Bottleneck and the Inference Land Grab
While models grab the headlines, the physical infrastructure required to run them attracted massive attention in 2025. Investors have realised that AI’s bottlenecks are increasingly physical: power, chips, and data centres. The AI infrastructure layer includes Cloud, Data Centres, and Virtualisation companies that can provide the physical and virtual environments necessary to run massive AI workloads, AI Chips and Processors as well as storage, international and other infrastructure.
Significant capital flowed into this layer, with companies like Crusoe raising a $1.4 billion Series E and Lambda securing a $1.5 billion Series E to support virtualisation and data centre needs. The hyperscalers committed an estimated $300 billion-plus to CapEx in 2025, filtering down to startups solving energy and compute constraints.
A critical shift is occurring from training to inference. As Rob May notes in his post “The Inference land grab begins“, training is a one-time CapEx event, but inference is the recurring, scaling OpEx. This shift triggered massive consolidation, highlighted by Nvidia’s aggressive $20 billion play for Groq to secure an “inference moat,” effectively forcing remaining chip startups to rethink their survival strategies.
3. Agentic AI: From Chatbots to System 2 Thinking
Moving up the stack, 2025 saw the rise of “Agentic AI” systems capable of multi-step, “System 2” thinking, to borrow a term from Daniel Kahneman’s book Thinking, Fast and Slow. System 2 AI agents represent a transition of AI from fast, intuitive responses to slow, deliberate reasoning. Rather than simple request-response loops, the new Agentic AI startups will be able to plan, make decisions under uncertainty and be capable of strategic thinking, as described by Rob May in “The inference land grab begins“.
Data on headcount growth reveals where the build momentum is: AI agent tool libraries and AI agent browser infrastructure were among the fastest-growing markets for talent in 2025. Smart Money VCs (the top performing funds) have concentrated their bets here, with Coding AI agents and End-to-end software development agents ranking as top markets for deal activity.
This sector is producing high-velocity startups. For instance, Anysphere, the maker of the AI coding editor Cursor, raised a $2.3 billion Series D, signalling the market’s hunger for agents that can execute complex professional workflows, Sierra, the company automating customer support with voice agents, etc.
4. Applications: The $100 Trillion Opportunity
While infrastructure builds the rails, the application layer is where the economic value is beginning to materialize. Enterprise AI revenue reached $37 billion in 2025, growing more than 3x year-over-year.
Investors are looking for startups that layer proprietary data over foundation models to solve “big, painful workflows”. We are seeing the emergence of a “layer cake” investment thesis:
- Vertical Applications: Companies like Clio (legal tech) raising $500 million and OpenEvidence (healthcare) raising massive rounds at high valuations per employee.
- Robotics: Physical AI is having a moment, with robotics companies raising a record $40.7 billion in 2025. Startups like Figure and Physical Intelligence are commanding unicorn valuations by applying AI models to real-world machinery.
What will 2026 bring to AI investors?
Most likely, you can expect the same or greater volume of investments in AI, but a higher concentration of recipients and less winners.
It’s also likely that will start seeing more acquisitions in this space as the foundation models continue to aggressively compete on both talent and revenue streams. If OpenAI ends up going public in 2026 (a move that is under works right now), we might see
As we close the books on 2025, the venture landscape has been fundamentally reshaped. The U.S. continues to dominate, capturing 70% of global venture funding, largely driven by these massive AI rounds.
For founders and investors, the lesson of 2025 is that AI is not a feature, it is the new baseline. Whether it is solving the energy crisis for data centers, building agents that code, or deploying robots in warehouses, the “AI Land Grab” is over. We are now in the era of deployment.
I'm a former general partner at google ventures, head of Google for Entrepreneurs in Europe and founding head of Campus London, Google's first physical hub for startups.
I'm also the founder of Techbikers, a non-profit bringing together the startup ecosystem on cycling challenges in support of Room to Read. Since inception in 2012 we've built 11 schools and 50 libraries in the developing world.
- Weekly Firgun Newsletter – March 6 2026 - March 6, 2026
- Weekly Firgun Newsletter – February 27, 2026 - February 27, 2026
- The Rise of the Personal AI Agent - February 26, 2026
