Record $510B flowed into startups in H1 2026 - two labs took 43% of it

$510 billion flowed into startups globally in the first half of 2026. It is a record. It is also a story about two companies.

OpenAI and Anthropic together took $217 billion of it - 43% of all venture funding, full stop. More than 70% of Q2 capital went to AI companies. The market is not broadening. It is concentrating at a rate the venture asset class has not seen before.

What the number actually says

The headline is a record. The structure underneath it is a funnel. Capital is not distributing across the AI category. It is pooling at two points: the frontier labs that make the models, and the infrastructure that serves them. Applied AI - the layer where most founders actually operate - got the remainder.

That remainder is still large in absolute terms. But the comp set has changed. When a seed investor prices your round, the question is no longer "is AI a good bet." It is "does this survive in a market where two companies can buy compute and talent at a scale you cannot."

What it means for your company

If you are raising, three things shifted.

First, your comparable is no longer "AI startups." It is the specific wedge you own inside a market dominated by two buyers. An investor who passed on you a year ago citing "AI risk" will pass again citing "concentration risk" unless you show why your wedge is not displaceable by a model release or a hyperscaler bundle.

Second, the exit market came back. Q2 notched one of the strongest periods for venture-backed exits in years. That matters for you because it means the buyers who were on the sidelines - the strategics, the late-stage funds - are writing checks again. Your Series B has a path it did not have in 2025.

Third, the talent market is the real cost. When two labs absorb that much capital, they absorb that much talent. Your hiring plan should assume senior AI engineering is scarcer and pricier than your last budget cycle reflected.

The decision it forces

This is a capital-markets story, but it forces a product decision. The question is not whether you can raise. It is what you own that a model release cannot flatten.

If your product is a thin wrapper over a frontier model, the concentration number is a threat. Two companies now have the capital to ship the feature you built, for free, inside the model. If your product is a workflow that gets better with usage, encodes domain logic, or owns a distribution surface the labs cannot reach, the concentration number is a tailwind - it means the substrate keeps getting cheaper while your moat is elsewhere.

You need to be able to answer, in one sentence, what you own beyond the model call. If you cannot, that is the work.

Three things to do this week

1. Re-cut your pitch around the wedge. Remove "AI-powered" from the first slide. Lead with the workflow, the data, or the distribution the labs cannot replicate. The model is the substrate, not the product.

1. Stress-test against a free-feature scenario. Run the exercise: if OpenAI shipped your core feature into the model tomorrow for free, what is left? Whatever is left is your real company. Build the narrative around that.

1. Re-price your round against the right comp. Stop benchmarking against the 2025 AI-seed median. Benchmark against applied-AI companies that raised after the concentration became visible. The valuation math is different.

The catch

The record hides a bifurcation. Most of the H1 capital went to a handful of mega-rounds. The median round size for applied AI did not move the way the headline implies. If you are pre-revenue, the bar is higher, not lower, than it was - because the capital that is flowing is flowing to proof, not promise.

Bottom line

The money is there. It is going to two places first. Your job is to show why the part that is left should go to you - and to build the thing that keeps it there after the next model drops. The decision is what you own. The deadline is now, because the comp set is already moving.

Source