Nvidia offers startups GPU credits in exchange for a share of future revenue

Nvidia on July 2, 2026 announced a new model for accessing AI equipment: the company will issue developer token credits in exchange for a share of their future revenue, according to Bloomberg. The program is aimed at startups that cannot afford upfront GPU infrastructure payments, and is changing how young AI teams obtain computational resources for model training.

How the program works

Token credits are notional units that provide access to Nvidia GPU capacity. A startup receives resources now and pays later—with a share of sales when the product generates revenue. The structure resembles tech accelerator models: a team receives something valuable (money or infrastructure) and in return commits to sharing future success.

Bloomberg does not disclose program details: the exact revenue percentage, agreement terms, and the type of capacity provided—whether Nvidia's own cloud GPUs or partner cluster infrastructure—remain unknown. Which teams are eligible for the program and what the entry threshold is are also unspecified.

For early-stage AI startups, GPU access is one of the primary financial barriers. Renting a single Nvidia H100 card in the cloud costs $2 to $5 per hour. Training a serious language or multimodal model requires hundreds of such cards over several weeks. A typical early-stage startup can spend $50,000–200,000 per month on compute alone—before generating any revenue. Revenue sharing eliminates this barrier: instead of upfront payment, a share of future growth that materializes only upon real success.

Nvidia already offered startups favorable terms through Inception and NCA Credits programs, but the revenue sharing model represents a structurally different approach: it's not a discount, but direct financial partnership.

Why is Nvidia doing this now?

Competition in the AI infrastructure market has intensified significantly in 2025–2026. Amazon, Google, and Microsoft are actively investing in their own chips—AWS Trainium, Google TPU, and Microsoft Maia. Meta is developing MTIA. Independent manufacturers—AMD with its MI300X series, along with startups Groq and Cerebras—are also gaining market share.

Nvidia, maintaining around 80% of the AI accelerator market, seeks to maximize the number of developers on its CUDA platform. Ecosystem lock-in is a strategic asset: switching to alternative architectures requires rewriting code, retraining teams, and losing accumulated expertise.

The revenue sharing program also allows Nvidia to monetize spare capacity without cutting market prices and build a portfolio of equity instruments in promising AI companies—effectively becoming a strategic investor rather than just a hardware supplier.

What this means

If the model scales, it could change the standard path for early-stage AI startups: instead of raising a venture round to pay for servers, direct agreements with chip manufacturers. Nvidia transforms from GPU supplier to growth partner. For startups, this lowers the barrier to entry; for Nvidia, it's a long-term bet on the next generation of AI companies.

*Meta is recognized as an extremist organization and is banned in the RF.