Cloud providers are quietly becoming investors in AI startups by swapping GPU or compute credits for equity, a practice that lets startups access tens of millions in infrastructure without paying cash but can lock them into long-term contracts and complex cap‑table issues.
How compute-for-equity deals work
A cloud provider—ranging from hyperscalers like Microsoft or Google to GPU-focused firms such as CoreWeave or Lambda—agrees to provide a large block of compute credits, often worth tens of millions of dollars on paper, in exchange for equity, warrants, or a committed multi‑year spend. No cash changes hands on the startup side; the provider gets a stake in a company and a guaranteed customer for expensive GPU capacity.
The most scrutinized example is Microsoft’s roughly $13 billion commitment to OpenAI, much of which was paid in Azure compute credits redeemable only for compute on Microsoft’s infrastructure. That arrangement let OpenAI train and serve models without depleting cash while Microsoft secured exclusive cloud status and integration rights. CoreWeave has run smaller versions of this playbook, taking equity stakes in AI companies, including Cohere, as part of multi‑year GPU rental commitments. CoreWeave is Nvidia‑backed and went public in 2025.
Why startups accept compute-for-equity
For pre‑revenue AI startups with no paying customers, compute is the dominant cost: training a frontier model can exceed what many startups would raise in a seed round. Compute‑for‑equity deals answer the immediate investor question—who’s paying for GPUs—without a priced equity round, due diligence roadshows, or upfront cash. Providers’ willingness to trade compute for ownership also signals external validation that can help future fundraising.
On paper, these deals sidestep headline valuation fights because the provider is compensated through future usage and potential upside rather than a per‑share cash price today.
The valuation and dilution problem
The apparent generosity hides a key distortion: compute credits are often valued at public retail sticker price when equity is calculated, but providers commonly deliver GPU time to large customers at steep discounts off list price. A startup that receives, for example, $50 million in credits may be diluting itself against a figure that overstates the provider’s actual cash cost. That gap is why some venture investors now request raw compute contract terms before leading a priced round: they must assess what the credits really cost the provider to understand how expensive the equity grant was.
Lock-in, clawbacks and operational constraints
Compute‑for‑equity arrangements typically include committed‑spend contracts—often three to five years—specifying minimum usage to keep preferential pricing. Switching cloud providers before the term ends can trigger clawbacks, forfeiture of unvested equity tied to continued usage, or unexpected cash bills. Startups can find themselves unable to move training workloads to a rival that ships a better chip or undercuts on price without breaking the economics of the equity deal they signed.
What this does to the cap table and governance
When a portion of a company’s shares was issued in exchange for compute rather than cash, later priced rounds become harder to value cleanly. New investors must determine the real worth of earlier credits, whether the provider holds board seats or information rights, whether warrants convert at a discount to the new round, and whether the provider has any control over future infrastructure choices. Those nonstandard instruments add negotiation time and legal costs.
There is also a governance conflict: a cloud provider that is both your equity holder and the source of your compute has incentives that may not align with the startup’s, particularly if the startup’s architecture would perform better on a competitor’s chips.
When these deals make sense and how founders should approach them
Compute‑for‑equity deals are not categorically bad: for startups with no other route to the compute they need, trading equity for GPU access can be the difference between existing and folding. But founders should treat the arrangement as a priced round denominated in GPU time. Key precautions include demanding that credit valuations be benchmarked against actual market rates rather than list price, scrutinizing committed‑spend clauses as carefully as any term sheet, and understanding any governance or clawback provisions that could affect future flexibility.
Viewed accurately, the pitch sounds like free money but functions as an equity transaction tied to infrastructure. Founders and investors should negotiate it with the same care as any capital raise.
