Four Transactions Ate Two-Thirds of Q2's Venture Dollars. One of Them Was a Rocket Company Buying an AI Coding Tool.

Venture capital just posted its best half-year on record, and the honest way to read the number is that most of the industry didn't participate in it. Crunchbase's H1 2026 tally, published July 2, puts global startup investment at $510 billion — more than the whole of 2025 ($440 billion) in six months. The instinct is to call this a boom. The more useful question is: a boom for whom?

The answer, increasingly, is four checkbooks. OpenAI and Anthropic together absorbed $217 billion of the H1 total — 43 cents of every venture dollar invested anywhere in the world, across every sector, every stage. AI companies broadly took more than 70% of Q2's $205 billion, up from roughly half a year earlier. Crunchbase's own accounting attributes close to two-thirds of Q2's dollars to just four transactions.

One of those four wasn't a funding round at all. Four days after SpaceX priced its Nasdaq debut at $75 billion raised — the largest IPO in market history — the company agreed to buy Cursor parent Anysphere for $60 billion in stock, the largest acquisition of a venture-backed startup ever recorded. A rocket company had just become, inside a single week, the buyer of the fastest-growing software business in history.

The Anysphere deal is worth sitting with, because it doesn't fit the infrastructure-versus-application-layer framework most of venture capital still uses to price risk. Cursor reached $2 billion in annualized revenue in February on pure application-layer distribution — developers choosing a coding tool because it was good, not because they were locked into a cloud contract. SpaceX isn't buying that distribution to protect a compute business. It's buying the coding data to feed a training pipeline for Musk's other company, xAI, and backfilling engineering talent after every one of xAI's co-founders left by March. The line between who builds the model and who builds the product on top of it just got erased by one stock-for-stock transaction.

This changes how the concentration story should be told. The consensus read — repeated in three straight quarterly recaps now — is that a handful of foundation labs are crowding out the rest of venture capital. True, but incomplete. The labs aren't just absorbing capital; well-capitalized adjacent players are using that capital to acquire the application layer outright rather than let it compete as an independent category. If Cursor could be bought for $60 billion by whichever entity happened to have the freshest balance sheet, "application-layer moat" isn't a permanent category. It's a bid.

For the middle of the market — the Series A and B vertical AI startups in credit, compliance, healthcare, or LatAm fintech — the read-through isn't "there's no room." It's that room now has two shapes. There's the shape where you get bought by whoever wins the infrastructure lottery, and the shape where you build something specific enough — proprietary data, regulatory relationships, a workflow embedded in how an industry actually operates — that no infrastructure company's balance sheet, however large, can simply absorb you into its supply chain. The first shape is a lottery ticket. The second is a business.

The $510 billion headline will get cited all year as proof venture capital is thriving. It's proof four cap tables are thriving. Whether the rest of the industry — the thousands of funds and founders who didn't get a call from Altimeter or a term sheet from SpaceX — actually benefited from H1 2026 is a separate question, and it's the one worth asking before writing the next check.