Startups raised a record $510 billion in six months. Two AI companies took 43% of it.

Global venture funding hit $510 billion in the first half of 2026 — more than the $440 billion invested in all of 2025 combined, and a record for any half-year on record. Nearly 40 AI startups reached unicorn status in those six months, at valuations ranging from $1 billion to $41 billion. Every instinct trained on headlines says boom, and the word is not wrong. But the desk's job is never to stop at the headline number; it is to check the instinct against the distribution underneath it. And the distribution here tells a very different, much stranger story than 'the market is hot.'

The number under the number

OpenAI and Anthropic alone accounted for $217 billion — roughly 43% of all startup funding in the half. Read that again, because it does not compute on first pass. Nearly half of every venture dollar deployed anywhere on the planet, across every sector — biotech, climate, fintech, defense, consumer, all of it — went to two AI companies. That is not a rising tide lifting all boats. It is a gravity well bending the entire capital landscape toward two points. The 'record funding' story and the 'historic concentration' story are built from exactly the same data; which one you tell depends entirely on whether you are one of the two raising, or one of the thousands competing for the 57% that's left.

For a founder outside the top handful, 'record year' can be the most misleading phrase in the market. The aggregate is up and your odds may be down at the same time, because the aggregate is being carried by rounds you will never be in. That is the reality the topline obscures, and it is the reality that actually governs most people trying to raise in this environment.

The mega-rounds themselves are real and worth naming, because their scale recalibrates what 'a lot of money' means. Prometheus, co-founded by Jeff Bezos, closed a $12 billion Series B — a Series B — led by JPMorgan Chase and BlackRock, at a $41 billion valuation, on total funding of $18.2 billion. Together AI raised $800 million at an $8.3 billion post-money valuation, with Aramco Ventures leading. DeepSeek raised $7.4 billion at north of $50 billion. Two years ago any one of these would have been the defining round of its year. In H1 2026 they are line items in a list, checks written at a scale that would have been unthinkable, now written like routine.

Nearly half of every venture dollar on earth went to two companies. That's not a tide. It's a gravity well.

What it signals — two truths at once

The concentration signals two things simultaneously, and holding both is the whole discipline of reading it correctly. The first is conviction. The largest and most sophisticated financial institutions on the planet — JPMorgan, BlackRock, sovereign-linked funds — are now underwriting frontier AI directly, at the top of the capital stack, with checks sized for infrastructure rather than experiments. That is a serious, considered vote that this is foundational technology and not a fad, from institutions whose entire business is being right about where durable value accrues. It deserves weight.

The second truth is fragility, and it is the inseparable other side of the first. When a category's capital is concentrated to this degree, the health of the entire market becomes hostage to a very small number of outcomes. If the top two companies deliver revenue and products commensurate with these valuations, the concentration will be remembered as foresight — the market correctly identifying the winners early. If either stumbles — a capability plateau, a regulatory shock, a business model that doesn't close — the correction will not stay politely contained to them. Everything priced off the assumption of their success reprices with them. Concentration is efficient on the way up and merciless on the way down, and you do not get to choose which direction you experience it from.

The skeptic's footnote

Record funding is an input, not an outcome, and the two are constantly confused. None of these valuations is verifiable against durable revenue from the outside — private companies disclose what flatters them — and 'raised the most' has never once, in the history of venture capital, been synonymous with 'built the most valuable business.' The dot-com peak was also a record-funding era, right up until it wasn't. The number that will actually matter in a year is not the $510 billion raised in this one; it is what fraction of that capital produced companies that can stand on their own revenue without the next round. That fraction is unknowable today and decisive tomorrow. Standard reminder, and standing disclaimer below: we report on these markets; we don't advise anyone into them.