$40 billion is a rounding error

A year ago, OpenAI raised $40 billion and it was the biggest funding round in history. It dominated headlines for weeks. Analysts debated whether any private company could justify that kind of capital. Today, that same number barely registers. OpenAI just closed $122 billion in a single round, and it was only one of four record-shattering deals in the same quarter. Q1 2026 delivered $297 billion in global startup funding, according to Crunchbase. Of that, roughly $242 billion, about 81%, went to AI companies. We have entered a phase where the numbers have genuinely lost meaning. Four deals alone accounted for 65% of all venture capital deployed on the planet in a single quarter. So what happens when capital itself becomes commodity-scale?

The quarter that broke the scale

The numbers are hard to process, even in context. OpenAI closed a $122 billion round at an $852 billion valuation, the largest private funding event in history. Anthropic raised $30 billion at a $380 billion valuation, which in any other era would have been the record. Elon Musk's xAI pulled in $20 billion before being acquired by SpaceX for $250 billion, forming a combined entity valued at $1.25 trillion. And Waymo, Alphabet's self-driving unit, raised $16 billion at a $126 billion valuation. Together, those four companies raised $188 billion in a single quarter. That is more than total global VC investment in any full year before 2021. Crunchbase reported that the quarter's total represented a 2.5x increase over Q4 2025's $118 billion. KPMG's Venture Pulse put the number even higher at $330.9 billion when including broader deal categories. Either way, we are in uncharted territory.

Strip out five deals and the market actually shrank

Here is the part that does not make headlines: the record-breaking quarter was built on a handful of mega-rounds. Remove them, and the picture inverts. Global deal count fell to 8,464 in Q1 2026, down from 10,097 the previous quarter, according to KPMG. CB Insights reported that the number of active investors globally dropped to 10,000, the lowest since Q3 2020. As venture analyst Alexandre Lazarow noted, "Strip out five deals and US VC actually shrank quarter over quarter." The market is not booming. It is bifurcating. A handful of companies are absorbing capital at a rate that inflates the headline numbers while the middle market contracts. Seed funding hit $12 billion (up 31% year over year), but the number of actual seed deals dropped by 30%. Investors are writing bigger checks to fewer companies. This is not a rising tide. It is a tsunami hitting five buildings while the rest of the neighborhood floods.

A rocket company now owns an AI lab

The SpaceX-xAI merger deserves its own reflection. In February, SpaceX acquired xAI in an all-stock deal, creating the most valuable private company in the world at a combined $1.25 trillion. The entity is now positioning for an IPO at $1.75 trillion or more. A rocket company acquired an AI lab. That sentence alone tells you how far the boundaries between industries have dissolved. Musk described the vision as "the most ambitious, vertically-integrated innovation engine on (and off) Earth," combining AI, rockets, satellite internet, and a social media platform under one roof. The logic is straightforward, even if the scale is absurd. AI needs compute. Compute needs energy. Energy is cheaper in space (in theory). Starlink provides global distribution. Everything feeds everything else. But the merger also reveals something about the current moment: at this valuation level, companies are not just competing on products. They are competing on the ability to control entire supply chains, from chips to data centers to distribution to the literal power grid.

More money does not buy more PhDs

Here is the paradox at the heart of the AI funding explosion: the barrier to building frontier AI is not capital. It is talent. A recent study from the University of Chicago documented the accelerating drain of AI researchers from academia to industry, driven by salaries that universities simply cannot match. Stanford's 2026 AI Index found persistent global shortages of doctoral-level AI talent. The "superstar market" for top researchers means that even with $122 billion in the bank, OpenAI cannot simply hire its way to faster progress. There are a finite number of people on Earth who can push the frontier of large language models, robotics, or autonomous systems. Money can build data centers, buy GPUs, and secure energy contracts. It cannot manufacture the kind of deep expertise that takes a decade to develop. This creates a strange dynamic. The companies that already have the best researchers attract the most capital, which lets them offer even higher salaries, which pulls more talent away from smaller competitors and academia. The rich get richer, not because they spend more, but because they already have what money cannot easily buy.

Jevons paradox at the investment level

When DeepSeek showed that competitive AI models could be built more cheaply, the intuitive reaction was: great, we will need less infrastructure. Nvidia's stock briefly cratered. But the opposite happened. Meta raised its AI spending guidance. Microsoft accelerated its build-out. The logic of Jevons paradox, a 160-year-old economics concept, played out in real time. William Stanley Jevons observed in 1865 that improvements in steam engine efficiency did not reduce coal consumption. Instead, they made coal-powered applications economical in more contexts, and total coal usage rose. The same pattern is playing out with AI. Cheaper models do not reduce demand for compute. They expand the universe of viable applications, which increases total compute demand. Now apply this at the investment level. Cheaper, more efficient models lower the barrier to entry. More startups experiment with AI. More enterprises adopt it. The total addressable market expands. And venture capital flows in to capture that expanded opportunity, even as the per-unit cost of intelligence falls. The result: models get cheaper, investment goes up, competition intensifies, and margins compress anyway. Efficiency creates abundance, and abundance creates its own kind of scarcity.

Follow the capex

If you want to understand where the $297 billion actually goes, stop looking at startup valuations and start looking at capital expenditure. The four major hyperscalers, Alphabet, Amazon, Meta, and Microsoft, are projected to spend between $635 billion and $700 billion on AI infrastructure in 2026, according to estimates from S&P Global and CreditSights. That is up roughly 67% from $381 billion in 2025, and more than 4x the $80 billion they spent in 2019. About 75% of that spend goes directly to chips, servers, networking equipment, and data centers. Gartner projects that global data center spending alone will hit $788 billion in 2026, growing nearly 56% year over year. McKinsey estimates that AI-related data center infrastructure will require $5.2 trillion by 2030. The breakdown: 60% for technology (chips and computing hardware), 25% for energy (power generation, transmission, cooling), and 15% for physical construction. The money is not floating in the cloud. It is being poured into concrete, copper, and silicon. The AI boom is, at its foundation, an infrastructure boom, and like all infrastructure booms, the real winners are the companies selling picks and shovels.

The dot-com comparison, and why it falls short

People keep reaching for the dot-com analogy, and on the surface the parallels are obvious. Euphoric valuations. Concentrated bets on a transformative technology. Capital flowing faster than business models can justify. But the scale difference is staggering. At its peak, the dot-com era saw roughly $100 billion in annual VC investment. Q1 2026 alone tripled that in a single quarter, concentrated overwhelmingly in one technology. There are important structural differences too. Dot-com companies were largely pre-revenue, built on speculative eyeball metrics. The current AI leaders have real, rapidly growing revenue. Anthropic's run rate hit $14 billion, growing more than 10x annually. OpenAI is reportedly on track for an IPO. These are not paper companies. But the dot-com era also had real companies with real revenue, companies like Cisco, Amazon, and eBay, that still saw their stocks crater 80-90% when sentiment shifted. Revenue does not prevent a correction. It just changes the floor. The more useful question is not whether this is a bubble, but what the distribution of outcomes looks like when the dust settles. In the dot-com era, a tiny fraction of companies captured nearly all the long-term value. Everything suggests the AI era will be even more concentrated.

What this means if you are building with a $0 budget

Here is the thing that gets lost in the spectacle of $122 billion rounds: the tools these billions produce are available to everyone. The frontier labs are in an arms race to build the most powerful models, and that race is funded by the very capital concentration described above. But the output, the actual models, are increasingly commoditized and accessible. API costs have plummeted. Open-source alternatives have narrowed the gap. The best coding assistants, reasoning engines, and multimodal tools are available for pennies per query. This is simultaneously terrifying and liberating for small builders. Terrifying because you cannot compete on infrastructure or capital. You will never train a frontier model. Liberating because you do not need to. The leverage available to a single developer with good taste and a clear problem to solve has never been higher. The capital concentration at the top is building a substrate. The opportunity for everyone else is in what gets built on top of it. But do not mistake accessibility for permanence. The companies controlling the infrastructure layer can change pricing, terms, and capabilities at any time. Building on someone else's platform has always been a tradeoff between leverage and dependency. That tradeoff is more extreme now than it has ever been.

The coordination problem

There is a subtler issue buried in the funding numbers: coordination costs scale with capital. When OpenAI had 100 employees and a few hundred million in funding, it could move fast, take risks, and pivot quickly. At $852 billion and thousands of employees, every decision involves committees, stakeholders, regulatory considerations, and reputational risk. The very capital that enables massive infrastructure investment also creates organizational gravity. This is why more funding does not automatically mean faster progress. The relationship between dollars and breakthroughs is not linear. After a certain point, additional capital produces diminishing returns on research velocity while increasing returns on infrastructure scale. You can build more data centers, but you cannot committee your way to a scientific insight. The implication is counterintuitive: the most interesting AI work over the next few years may not come from the best-funded labs. It may come from smaller teams that can still iterate quickly, take unconventional approaches, and explore directions that a $852 billion company cannot justify.

The new normal

A year from now, we will probably look back at $297 billion quarters the way we currently look at $40 billion rounds, as a quaint milestone on the way to something larger. The SpaceX-xAI entity is planning an IPO that would make it the most valuable company ever to go public. OpenAI's IPO is expected in the second half of 2026. The capital cycle has its own momentum now. But the underlying dynamics remain the same as they have been in every capital-intensive technology transition. Most of the money will be wasted. A small number of winners will capture disproportionate value. The infrastructure built with today's billions will outlast many of the companies that funded it. And the most lasting impact will come from applications that nobody is currently funding at scale. $40 billion is a rounding error now. The question is not whether that is sustainable. It is what we choose to build while the window is open.