wsj.com web signal

OpenAI, Anthropic Battle for Startups With Free Compute Credits

TL;DR

  • Anthropic offers eligible startups up to $100,000 in Claude credits with no equity required, via direct or VC-partner tracks.
  • OpenAI's Grove cohort pays $50,000 to selected teams in a 5-week San Francisco program, with a smaller Ramp-linked tier of $2,500.
  • The push signals frontier labs are competing on customer acquisition, not just token pricing, as enterprise revenue becomes the durable prize.

The API layer is turning into a customer-acquisition channel, and the Wall Street Journal frames the current credit programs as a race by the frontier labs to grab durable enterprise revenue before token prices commoditize the underlying product.

The numbers, as tracked across the current program pages, are meaningful for an early team. Anthropic's startup program advertises up to $100,000 in Claude credits with no equity requirement, split across a rolling direct track of up to $25,000, a VC-partner tier that stretches the total higher, and an Anthology fund with Menlo Ventures that adds another $25,000. OpenAI's Grove cohort in San Francisco is smaller and more selective at $50,000 across a 5-week program, with a separate Ramp-linked Founder Stack tier of up to $2,500 for teams that arrive without a VC referral.

Why this matters if you are not a founder chasing free tokens: the labs are subsidizing usage in a competitive land grab whose end state is not settled. Anthropic and OpenAI already sit at the top of the enterprise AI stack, with one recent tally putting their combined share of AI startup revenue near 89 percent, and by capturing startups early they hope to convert credit spend into revenue that persists once the free tier runs out. The trade for a founder is that your default model gets embedded before you have really evaluated the alternatives, and the switching cost shows up later.

The honest caveat is that the specific dollar figures move constantly and program terms shift month to month; much of what is public relies on tracker sites and founder self-reports rather than fully disclosed lab data. What the reporting does not give you is a clean picture of conversion, how many credit recipients turn into meaningful paying accounts, or how much of each lab's disclosed enterprise growth is genuine revenue rather than credits burning down.

The interesting thing to watch is not the sticker amount but the terms attached to the larger tiers, whether that is data sharing, referral funnels, or a right of first refusal. That is where the real customer acquisition cost lives, and it is what will determine whether these programs are a genuine gift to the founder or a very cheap option for the lab writing the check.