OpenAI, Anthropic, and Google are competing for startups with credit packages topping $3M

Some early-stage AI startups are receiving credit offers worth more than $3 million from multiple providers competing for their business, a figure that matches the median U.S. seed round, according to PitchBook data cited by the Wall Street Journal. The credits cover cloud computing and tokens, the unit AI providers use to measure and bill for model usage.

The competition is accelerating. OpenAI and Anthropic have both deployed expanded sales teams and are running promotions to land new enterprise customers. Google Cloud is offering select startups up to $500,000 in cloud credits, early access to Gemini models, and occasional introductions to DeepMind engineers, a company spokesman told the Journal. Microsoft and Amazon Web Services are also running startup credit programs, the Journal reported.

Why providers are moving aggressively now

The logic behind the credit push is straightforward: win a startup's infrastructure dependency early, and retain it as the company scales into a paying enterprise account. The Journal reported that some founders say the offers are rich enough to defer their next fundraising round, because credits effectively substitute for the capital they would otherwise raise to cover compute costs.

The competitive intensity also reflects pressure building on the providers themselves. OpenAI and Anthropic both face margin challenges ahead of anticipated initial public offerings. At the same time, open-weight models and cheaper alternatives, including several developed in China, are eroding the pricing floor, the Journal reported. Discounting through credits lets providers compete on cost without formally cutting list prices.

Cursor, the AI coding tool acquired by SpaceX, offered a 75% discount through July 5, according to the Journal, illustrating how the credit dynamic is cascading beyond pure cloud infrastructure into application-layer tools.

The lock-in risk behind the free compute

For IT and procurement teams evaluating AI vendors, the credit environment creates a specific evaluation problem. A startup or enterprise team that builds deeply on one provider's token API, fine-tuning environment, or proprietary model features during a credit window faces real switching friction when that window closes. Token pricing and API structures vary enough across providers that migrating a production workload is not trivial.

Founders who spoke with the Journal described actively playing providers against each other to improve terms, a negotiating posture that enterprise procurement teams can replicate. The key variable is how quickly a specific workload would accumulate platform-specific dependencies, such as provider-only embeddings formats, proprietary agent frameworks, or fine-tuned model weights stored in a vendor's ecosystem.

What this means for your team

• Audit current AI workloads for provider-specific dependencies before accepting or renewing large credit packages; free compute now can translate into high switching costs at contract renewal.
• Use competing offers as leverage: the Journal reported that startups are already negotiating providers against each other, and enterprise procurement teams have more volume leverage to do the same.
• Build portability checkpoints into any multi-year AI platform agreement, particularly around model weights, embeddings, and agent framework compatibility.
• Track the IPO timelines of key AI providers: the Journal notes that margin pressure ahead of public offerings is a primary driver of current discount depth, suggesting the current terms may not persist once those milestones pass.