PitchBook data released this week shows AI companies raised $97 billion in venture funding through November 2025—already surpassing 2024's total. But here's the part nobody's talking about: nearly 60% of that went to just five companies. OpenAI, Anthropic, xAI, Databricks, and a handful of others are vacuuming up capital while thousands of smaller AI startups fight over scraps.
The Numbers Are Staggering
$97 billion in 11 months represents a 42% increase over 2024's full-year total. At this pace, 2025 will close with over $105 billion invested in AI companies—more than the previous three years combined.
But the concentration is extreme. OpenAI's $150 billion valuation round, Anthropic's multiple funding rounds totaling $15+ billion, xAI's $15 billion raise, Databricks' $4.8 billion Series L, and a few other mega-rounds account for roughly $57 billion. That's nearly 60% of all AI funding going to five companies.
The median AI seed round is $2.5 million, down from $3 million in 2024. Series A rounds average $12 million, also declining slightly. While mega-rounds get bigger, early-stage funding is actually contracting. That's a sign of market bifurcation—huge money for winners, less for everyone else.
Who's Actually Getting Funded
The top funding categories are: foundation model companies (OpenAI, Anthropic, xAI), AI infrastructure (Databricks, companies building tools for AI development), and vertical AI applications (companies applying AI to specific industries like healthcare, legal, or manufacturing).
Foundation models are eating the most capital because training and running frontier models is obscenely expensive. OpenAI reportedly burns through billions annually on compute alone. Anthropic's costs are similar. These companies need massive funding rounds just to stay operational.
Infrastructure companies like Databricks, Scale AI, and Weights & Biases are raising huge rounds because they're selling picks and shovels to the AI gold rush. As long as companies are building AI products, they need data platforms, labeling tools, and MLOps infrastructure.
Vertical AI is the most crowded category with the least funding per company. Thousands of startups are building "AI for X"—legal, healthcare, customer service, sales, recruiting, you name it. Most raise small seed rounds and struggle to get to Series A.
The Winner-Take-Most Dynamic
AI has turned into a winner-take-most market faster than any previous technology wave. Network effects, data advantages, and compute requirements create natural concentration.
If you're a developer building an AI product, you probably use OpenAI or Anthropic's APIs. That gives those companies data advantages, ecosystem lock-in, and revenue to fund further development. Competing models need to be dramatically better to justify switching costs.
Infrastructure follows the same pattern. If you're using Databricks for your data platform, switching to a competitor requires migrating terabytes of data and rewriting pipelines. That's painful enough that most companies stick with their initial choice.
Vertical AI applications compete more on execution than technology—who has better go-to-market, deeper industry expertise, and faster product iteration. But even there, well-funded companies can outspend smaller competitors on sales and marketing until they dominate their niche.
What Happened to the Democratization Narrative
Two years ago, the story was "AI will democratize access to capabilities that were previously reserved for tech giants." Open-source models, affordable APIs, and easy-to-use tools would let anyone build AI products.
That's partially true—it's never been easier to build an AI prototype. But turning that prototype into a sustainable business requires: massive compute, expensive talent, strong distribution, and enough capital to survive while revenue ramps. Most startups don't have those resources.
The real democratization happening is: individuals and small teams can build impressive demos. But turning demos into products that make money? That still requires traditional advantages like capital, distribution, and operational excellence.
The Valuation Insanity
OpenAI at $150+ billion. Anthropic at $183 billion (reportedly). xAI at $230 billion. These valuations are based on: projected revenue growth, strategic value to investors, and the belief that AI will eat trillions of dollars of economic activity.
But here's the uncomfortable truth: most of these companies are losing money. OpenAI's revenue is growing fast but so are costs. Anthropic is well-funded but not profitable. xAI is even earlier-stage.
The valuations assume these companies will eventually figure out business models that justify the numbers. Maybe they will—Microsoft is profitable on $200+ billion revenue, so scale does work in tech. But there's also a scenario where AI capabilities commoditize faster than these companies can build moats, and valuations compress dramatically.
Where the Money's Coming From
Traditional VCs are being crowded out of mega-rounds by sovereign wealth funds, corporate strategics, and public market investors taking late-stage private positions.
OpenAI's recent round included Microsoft, NVIDIA, SoftBank, and UAE sovereign wealth funds. Anthropic's backers include Amazon, Google, and various asset managers. These aren't traditional venture investors—they're strategic players with their own AI agendas.
That creates different dynamics. Strategic investors care about AI capabilities they can leverage, not just financial returns. Sovereign wealth funds are playing long-term geopolitical games. Asset managers are seeking pre-IPO access to high-growth companies.
For startups, it means: raise from the right investors who bring strategic value beyond capital, because checks are getting bigger but terms are getting more complex.
The Acquisition Market
Over 140 AI companies were acquired in 2025, mostly by big tech. Google acquired Neon for $1 billion. Microsoft acquired multiple smaller AI startups. Amazon has been actively buying talent and technology.
Most acquisitions are acqui-hires—big tech buying AI startups primarily for the team, not the product. A typical deal: $5-15 million for a seed-stage team of engineers with AI expertise. That's cheaper than competing for talent on the open market.
The exit multiples for AI companies that actually have revenue are all over the map. Some sell for 20-30x revenue. Others struggle to get 5x. The variance reflects uncertainty about which AI business models actually work long-term.
What This Means for Founders
If you're starting an AI company in 2026, the strategic question is: can you build something defensible without raising $50+ million? Because if you need that much capital, you're competing for investor attention with well-funded incumbents.
The viable paths seem to be: (1) Build vertical applications in unsexy industries where big tech isn't competing. (2) Create developer tools that make AI easier to use. (3) Focus on use cases that require specialized domain expertise, not just model capabilities. (4) Bootstrap with services revenue until you have enough traction to raise on attractive terms.
The "build a foundation model and compete with OpenAI" path is essentially closed unless you have unique advantages like access to proprietary data, custom hardware, or regulatory arbitrage (building in jurisdictions with different AI rules).
The Bear Case
All this funding assumes AI capabilities continue improving exponentially and enterprises adopt AI widely enough to justify the investments. If either assumption breaks, valuations collapse.
We've seen this pattern before: internet boom/bust, crypto boom/bust, and numerous other technology cycles. Huge capital flowing into a hot sector, valuations decoupling from fundamentals, eventually a correction when reality doesn't match expectations.
The bull case is that AI really is different—it's applicable to nearly every business process, capabilities are improving consistently, and we're still in early innings of a multi-decade transformation. The bear case is that AI is overhyped, current limitations prevent widespread adoption, and most applications don't generate enough value to justify the costs.
My Take
The concentration of funding into a handful of winners is probably rational even if it feels unfair. Foundation models require massive capital, and there's only room for a few winners. Infrastructure markets tend toward consolidation. The real opportunity is in vertical applications, but those require different skills than most AI founders have—deep industry knowledge, strong sales capabilities, and operational excellence.
I worry that we're funding demos, not businesses. Companies raise based on impressive technology that doesn't yet have product-market fit or clear monetization paths. Eventually, investors will demand returns, and many of these companies won't have sustainable business models.
But I'm also bullish on the aggregate impact. Even if 90% of AI startups fail, the 10% that succeed will create enormous value. That's always been true in venture capital. AI just happens to be the biggest swing we've seen since the early internet.
The next 12-18 months will be telling. Revenue growth, profitability timelines, and actual enterprise adoption will start separating winners from pretenders. Some of these $100+ billion valuations will look prescient. Others will look absurd.
Place your bets accordingly.