Startups raised a record $510B in H1 2026. Two companies took 43%

Global startup funding hit $510 billion in the first half of 2026, the biggest half-year total ever recorded, according to Crunchbase data published this month. That is more than the $440 billion invested across all of 2025. It is also a deeply misleading number if you read it as a startup boom, because $217 billion of it, 43 percent of everything raised on the planet, went to exactly two companies: OpenAI and Anthropic.

The numbers

The half splits into two very different quarters. Q1 delivered roughly $305 billion, a record quarter on its own, as TechCrunch reported, anchored by OpenAI’s financing, which Crunchbase called the largest venture deal ever at north of $100 billion and a post-money valuation reported above $840 billion. Q2 added roughly $205 billion across more than 5,000 companies, and again one deal did the heavy lifting: Anthropic’s $65 billion round was close to a third of the entire quarter’s global total, stacked on top of the roughly $30 billion the company raised earlier in the year, per Crunchbase’s monthly recaps.

Zoom out to the category and the tilt gets steeper. Crunchbase’s data shows more than 70 percent of Q2 global funding went to AI companies, up from just under 50 percent a year earlier. Global seed funding in Q2 was about $12 billion. That is roughly 6 cents of every venture dollar in the quarter reaching the bottom of the funnel, in the middle of the loudest funding headline the industry has ever printed.

Mega-rounds wearing a trenchcoat

So is $510 billion breadth, or a handful of mega-rounds in a trenchcoat? Do the subtraction. Remove OpenAI and Anthropic and the first half comes to about $293 billion for every other startup on earth. That is a healthy market, comfortably ahead of the post-2022 trough, but it is not a new paradigm, and it would not be generating record headlines on its own. The record lives almost entirely inside two cap tables.

It also matters what kind of money this is. Rounds of $65 billion and $100 billion are not venture capital in any classical sense. They are compute procurement financed through equity, closer to project finance for data centers than to backing a team and a prototype, a structure we unpacked in our breakdown of the AI capital stack. The investors are sovereign funds, chipmakers, and hyperscalers with strategic exposure, not traditional VCs sizing a fund-returner. Counting those checks in the same column as a $4 million seed round tells you very little about the health of startup formation.

The signal

The honest read: this is not a startup funding boom, it is a capital-intensity story about two frontier labs, with a decent mid-market underneath it. The bull case for the concentration is that the revenue at the top is real and compounding at unprecedented speed, with AI companies reaching $100 million ARR in under 18 months. The bear case is that when 43 percent of global venture flows to two loss-making companies whose spending plans assume ever-cheaper compute and ever-growing demand, the system’s failure modes concentrate too. We laid out the evidence on both sides of the bubble question earlier this year, and this data sharpens rather than settles it.

One genuinely encouraging detail in the same report: exits soared. Crunchbase logged record IPO and M&A activity in Q2, which is the release valve this market has been missing. Capital that comes back can be recycled; capital that only goes in, at ever-higher marks, is how bubbles are built.

What the framing hides

Caveats worth holding onto. Mega-round totals often include staged commitments and tranches rather than cash actually wired, so the headline overstates money at work today. Crunchbase’s seed-stage data lags by design, meaning early-stage totals will revise upward and the concentration ratio will soften somewhat. Different trackers (PitchBook and CB Insights close their books on different rules) will print different totals for the same half. And records measured in nominal dollars flatter the present: the comparison every allocator actually cares about, returns per dollar deployed, will not be knowable for years.

What to watch

Three things through year-end. First, the concentration ratio in Q3: if the two-lab share keeps climbing, the trenchcoat thesis hardens. Second, whether either lab returns to market within 12 months, which would say more about burn rates than about demand. Third, whether the exit window stays open; a record half of money in only works if money keeps coming out. The $510 billion headline will age well or badly depending almost entirely on decisions made inside two boardrooms, and that, not the record itself, is the story.