Axios: Open-Source AI Push Rattles OpenAI-Anthropic VC Bet

The venture math on this AI cycle has rested on one uncomfortable fact, and Axios's Dan Primack laid it out plainly: OpenAI and Anthropic between them absorbed more than 60% of every venture dollar committed to US startups in the first half of 2026, per PitchBook. LPs stopped treating that concentration as a problem once VCs reframed the pitch as access rather than selection. The bet only works if those two names keep the lead.

That is why the open-source shift Primack flags matters more than a normal news cycle would suggest. Moonshot's Kimi K3 landed two weeks ago, described as powerful and cheaper. Nvidia CEO Jensen Huang told Axios that 'open-source models that are excellent should be used,' then signed a public letter backing open weights alongside Microsoft, Meta and Andreessen Horowitz. Those signatories mostly benefit if open weights drive more compute, more ads and more distribution, so read the alignment with that in mind. OpenAI and Anthropic pushed back on safety grounds and were promptly accused of regulatory capture.

The other data point Primack leans on is SpaceX, whose stock closed Friday 49% below its all-time intraday high, 23% under its initial trades and 15% under the IPO price. He uses it as a momentum warning for the private AI mega-rounds priced off a similar cannot-afford-not-to-own-it logic. If that is what happens to a beloved private name once it hits public markets, the marks LPs are carrying on OpenAI and Anthropic at cost are not obviously safe.

The honest caveat is that the last time an open-weight release spooked this trade, when DeepSeek arrived in early 2025 and was called an extinction-level event for frontier labs, OpenAI and Anthropic responded with stronger models and significantly higher revenues. What the reporting does not tell you is how LPs are actually pricing this concentration risk today, whether any 2026 fundraise has been restructured because of it, or how Kimi K3 benchmarks against the closed frontier on tasks paying customers care about. The story worth watching is not whether open source wins, but whether the second-tier labs that lost this cycle's funding lottery become a bid target for enterprise buyers who would rather not pay OpenAI's price.