Sponsored by
AI Spotlight · Business & Strategy
The Great AI Giveaway: Why Compute Is Suddenly Free
OpenAI and Anthropic are handing startups millions in free tokens and compute, even as both companies race to fix their margins before an IPO. Here is what is actually being offered, and the catch nobody says out loud.
Hans Ibarra, a founder building an AI voice startup, recently found himself holding competing offers worth more than $3 million in cloud computing and token credits, roughly the size of an average US seed round, according to a Wall Street Journal report. He is not an outlier. Across Silicon Valley, early-stage founders are fielding a wave of aggressive compute offers from Anthropic, OpenAI, and the major cloud providers, all fighting for a slice of the same small pool of AI-native companies.
The scale of the discounts is unusual even by Silicon Valley standards. Cursor, the AI coding company recently acquired by Elon Musk's SpaceX, offered a 75 percent discount that ran through July 5 [1]. Cloud providers are matching the intensity: a Google Cloud spokesperson confirmed the company hands out up to $500,000 in credits, early access to Gemini models, and occasional access to DeepMind engineers, while Microsoft and Amazon Web Services offer comparable perks of their own.
This issue breaks down why this giveaway is happening now, exactly what startups are being offered, and why the "free" part of free compute usually has a price tag attached later.
* * *
Why now
The pitched battle comes as AI companies chase lasting revenue
The timing is not a coincidence. Both OpenAI and Anthropic need to improve their margins ahead of expected IPOs, and the fastest way to demonstrate durable enterprise revenue is to get as many high-growth startups building on your platform as early as possible, before their architecture, evals, and vendor relationships harden into something expensive to switch. At the same time, both companies face growing pressure from cheaper open-weight models, many of them out of China, which makes locking in customers now more urgent than optimizing margin per customer today.
Anthropic's own numbers illustrate why this land grab matters commercially. The company's revenue reportedly surged fivefold late last year on the strength of Claude Code and its Cowork product, pushing the company toward a valuation near $1 trillion, while OpenAI reportedly did not catch up on developer traction until it shipped GPT-5.4 in March and began actively selling its Codex tool directly to startups [1].
|
The two forces driving the giveaway IPO pressure: Both OpenAI and Anthropic need visible, durable enterprise revenue growth before going public. Cheaper competition: Open-weight models, particularly from China, are compressing prices and making early lock-in more valuable than short-term margin. |
* * *
The bidding war
Inside the escalating offers for Y Combinator startups
Y Combinator companies have become the single most contested prize in this fight, and the escalation over just a few months shows how seriously each company is taking it. In May, Sam Altman announced $2 million in token credits for YC startups in exchange for equity stakes. Anthropic countered almost immediately with $500,000 in credits and no equity requirement at all, a dramatic jump from its previous offer of just $30,000.
OpenAI then matched Anthropic's no-equity offer at $500,000, while also keeping an optional $1.5 million package available in exchange for shares. With roughly four YC cohorts a year of about 200 companies each, the two firms alone could hand out up to $800 million in credits combined across a single year [1].
| Offer | Value | Equity required? |
|---|---|---|
| OpenAI (initial, May) | $2,000,000 | Yes |
| Anthropic (counter) | $500,000 | No |
| OpenAI (revised) | $500,000 + optional $1.5M | No / Optional |
The direction of that escalation is notable on its own: within a single quarter, both companies moved from equity-for-credits deals toward no-equity offers, effectively treating startup adoption as a pure distribution cost rather than an investment they need a stake in.
How AI-Era Pricing Is Reshaping Finance Operations
Usage-based and hybrid pricing models are changing how B2B companies generate revenue — and creating new headaches for the finance teams behind them.
Tabs co-founder Rebecca Schwartz and PwC Partner Amit Dhir sat down to unpack exactly what that means in practice: how pricing model decisions ripple into revenue recognition, forecasting, and financial ops — and what it takes to scale without piling on manual work.
Watch the on-demand recording to get practical frameworks, real-world examples, and a clear path to operationalizing usage-based revenue — including a forward-looking take on how AI will reshape financial workflows. If your team is navigating pricing complexity heading into the back half of the year, this is worth an hour.
Beyond the two leaders
Cloud providers are running the same playbook at scale
This is not purely a two-horse race between model makers. The major cloud providers are running parallel programs aimed at the exact same pool of startups, often stacking on top of model credits rather than competing directly with them.
|
Up to $350,000 in cloud credits for AI-focused startups, plus early access to Gemini models and occasional access to DeepMind engineers. Non-AI startups are eligible for up to $200,000. |
|
Microsoft for Startups Founders Hub Immediate access to $25,000 in Azure credits with no VC referral required, scaling up to $150,000 to $200,000 as the startup grows, alongside developer tooling and technical support. |
|
Standard credits above $100,000, rising to as much as $300,000 for AI-focused startups building on Amazon Bedrock or SageMaker infrastructure. |
The consistent thread across every one of these programs, as noted in independent analysis of the WSJ report, is the underlying goal: pull startups into a specific ecosystem early and make the tools too embedded to leave later [1].
* * *
The real cost
Free tokens are financing, not a gift
The practical risk for founders is straightforward but easy to ignore while credits are flowing. A lot of AI startups do not know their real gross margin until the credits run out, because the product looks efficient and the customer acquisition cost looks clean, right up until the subsidy disappears. Free tokens can accelerate a prototype, but they can also quietly shape architecture, evals, observability, and data-retention choices long before a team has measured the true post-credit cost of running that same workload at full price.
This reflects a familiar platform pattern: early defaults tend to become durable revenue for the platform, not the startup. If a team builds its prompts, fine-tuning plan, logging, and customer commitments entirely around one provider's stack while credits are covering the bill, switching providers later can end up costing far more than the original credit headline ever suggested.
|
Before building on subsidized credits, check — What the actual per-token cost becomes once the credits expire. — How portable your prompts, evals, and fine-tuning setup are to a different provider. — What data-retention terms you agreed to in exchange for the discounted access. |
None of this means the credits are not worth taking. For a cash-constrained early-stage team, $500,000 in no-equity compute is a legitimate runway extension. The mistake is treating it as free rather than as temporary financing that needs to be modeled and repaid in the form of a harder migration later.
* * *
|
Before you go If you run a startup, have you taken free AI credits, and did you check what happens when they run out? Hit reply with one sentence. The most interesting answers will shape a follow-up issue on how founders are managing the post-credit cost cliff. |
Until next time,
AI Spotlight
Practical AI, translated into real work, once a week.