Cloudflare's ceiling moves to $350k, Google's credits still won't pay for Claude
Cloudflare's tiers moved, Google's credits exclude third-party models, and AWS lists an account requirement that quietly disqualifies applicants.
For the first issue we did the boring thing: opened every top-tier program page and compared it with what we had on file. Three findings were worth writing down — one real change, and two rules that quietly decide how much a program is actually worth to you.
Cloudflare's top tier went from $250,000 to $350,000, and the $25,000 rung disappeared
When Cloudflare last described the program in a blog post, in April 2025, it had four tiers: $5,000, $25,000, $100,000 and $250,000. The current program page lists three: $10,000, $100,000 and $350,000. The entry tier doubled, the top tier is up 40%, and the middle rung is gone.
The number most people will quote at you is still $250,000, because that is the figure in the announcement post that search engines have had eighteen months to rank. The higher number is on the program page itself.
Losing the $25,000 tier matters more than gaining $100,000 at the top. That rung used to be reachable with an active profile and under $1M raised, no partner required. Now the self-serve path stops at $10,000, and everything above it needs funding from one of Cloudflare's affiliated partners. If you are bootstrapped, your ceiling did go up — from $5,000 to $10,000 — and the next step up got further away.
Our Cloudflare card carries the current tiers. Note the expiry: twelve months, no extensions, which is short for this size of allocation.
Google's $350,000 will not pay for a third-party model. AWS's $200,000 will.
Google's AI startup track reaches $350,000 by covering all of your usage up to $250,000 in year one, then 20% of usage up to another $100,000 in year two. The program page also says which models the credits apply to: Google's own, Gemini and Gemma. Third-party models are billed directly and are not covered.
AWS goes the other way. Activate credits are redeemable on third-party models through Bedrock, which includes Anthropic's and Meta's.
So if your product runs on Claude or another model you did not train, a ceiling that is 75% larger can be worth less to you than the smaller one. The comparison that matters is not the headline number, it is the headline number multiplied by the share of your bill the credits are allowed to touch. A team spending most of its inference budget on a third-party model gets close to nothing from the AI-first tier's ceiling and should be reading the AWS terms instead.
This is the same trap we wrote about in comparing cloud credit programs: the ceiling is a marketing number, the exclusions are the product.
AWS lists a pre-Series B cap and a paid-plan account requirement
The Activate credits page states the eligibility bar as:
- pre-Series B;
- founded within the last ten years;
- an AWS account on a paid tier plan;
- either new to Activate credits, or asking for more than you previously received.
The account condition is the one that catches people. An application from an account still on the free plan does not meet the bar as written, and that is an easy thing to fix before applying rather than after being turned down. The pre-Series B cap is worth planning around too — it is a door that closes on announcement day, not on the day the money lands.
Both details are now on our AWS Activate card. If you are timing applications around a raise, credit sequencing covers which programs to file before the round is public.
What we changed in the database
Nothing on the three cards above needed a correction to its headline figure — the amounts we publish matched the provider pages. What we added were the two rules that change what those figures mean: Google's exclusion of third-party models, and the AWS account condition. You can see every program's rules on the full list.