The offers from growing AI-sales armies at companies such as OpenAI and Anthropic are so rich that some early-stage startup founders say they won’t need to raise money as soon as they expected, and others have been able to play AI companies off one another. Startups have received offers that in some cases amounted to more than $3 million in credits from multiple companies for cloud computing and tokens, the central units used to measure and charge for AI usage, founders say. That is the size of the median U.S. seed round, according to PitchBook.

Alphabet’s GOOGL 1.82%increase; up pointing triangle Google Cloud is giving some startups up to $500,000 in cloud credits and early access to Gemini models. It also occasionally offers special access to DeepMind engineers, a Google spokesman said. Microsoft and Amazon Web Services also offer startups special perks.

The pitched battle for business users comes as AI companies seek lasting streams of revenue. They hope that by winning startups as customers early in the life of new companies, their tools will become integral to the venture’s growth over time.  

OpenAI and Anthropic are offering a string of promotions and one-time bonuses, even as both companies face enormous pressure to improve their margins ahead of expected initial public offerings. They also face competition from increasingly powerful “open weight,” or free models, as well as cheaper ones, many of which were developed in China.

The token deals available to founders “directly correlate to the scale you can grow your product,” said Ibarra, co-founder of Dialogus. “If you’re not getting this deal, you will need to raise money to buy those.”

Anthropic’s revenue skyrocketed late last year as millions of new users tapped their Claude Code and Cowork software to autonomously complete a range of tasks. Claude’s viral popularity helped launch the “agentic” AI era, in which top AI companies are increasingly focused on building tools that customers can use to complete long-running knowledge-work tasks, such as coding and deep research.

For months, OpenAI struggled to match the strength of Anthropic’s coding-focused models and products, giving its younger rival the advantage in the lucrative enterprise market. Companies initially nudged employees to use AI more in their work, but soon some saw the bills as prohibitively high.

OpenAI’s fortunes began to change after the March release of a new model, called GPT-5.4, that matched many of Anthropic’s capabilities. The company has since deployed its salespeople to sell its Codex tool, which is powered by its GPT model, to startups across Silicon Valley, oftentimes offering volume discounts and other sweeteners to win new customers.

Semianalysis, an AI-infrastructure data and consulting firm, recently published research showing how heavily the companies are subsidizing power users.

Subscribers to Anthropic’s Claude Max plan, which costs $200 a month, are able to burn tokens worth $8,000 in their usage-based plans administered through an application programming interface, or API, which allows them to integrate Anthropic’s technology into their products. Maximum use of OpenAI’s ChatGPT Pro 20x plan, which also costs $200 a month, can burn tokens worth $14,000.

In their quest to secure new business customers, Anthropic and OpenAI have zeroed in on startups participating in Y Combinator, the Silicon Valley institution that launched Airbnb and Stripe. In May, Sam Altman announced that OpenAI would give $2 million in token credits to every startup participating in the accelerator program in exchange for equity in those companies.

Y Combinator runs four cohorts a year. Poppy Lynch for WSJ

Around the same time, Anthropic began offering Y Combinator startups $500,000 in free credits, a sharp increase from the $30,000 it previously offered, an Anthropic spokeswoman said. Anthropic’s offer doesn’t require startups to give up equity.

Soon afterward, in recent weeks, OpenAI adjusted its deal, offering startups $500,000 in free credits—no equity required—with an optional additional $1.5 million in credits in exchange for equity, according to people familiar with the matter.

The back-and-forth reflects the intense battle the companies are in to sway young startups that could become large customers in the future. Model providers hope that by offering these companies discounts, they can lock them into their ecosystem. 

At an event hosted to kick off the summer season of Y Combinator’s program, representatives from OpenAI and Anthropic, among others, met with startup founders and offered advice about making the most of their token usage, including by embracing loop engineering, or teaching AI agents to repeat a task until they have achieved their assigned goal.

Touchmark, an AI startup that was accepted by Y Combinator in May, was immediately granted $1 million in token credits from OpenAI and Anthropic before the accelerator even kicked off its summer session.  

For Ilia Bolgov, co-founder of Touchmark, the credits meant “quite a lot of time to go all-in on tokenmaxxing,” a term for using as many tokens as possible, he said. “It’s hard to imagine productivity now without these deals.”

The credits represent a massive potential investment on behalf of the model providers. Y Combinator runs four cohorts a year, with recent cohorts enrolling about 200 companies each, meaning OpenAI and Anthropic could offer up to $800 million in combined AI credits in the next year.

“The world of AI is being powered by OpenAI and Anthropic because they are giving startups the money to pay for it,” said Christopher Acker, co-founder of SuperPenguin, a firm that helps companies track their AI spending.

“If I’m choosing between a really cheap Chinese model that I actually have to pay for, and a very expensive Anthropic model that I don’t have to pay for, I’m going to pick the Anthropic model,” Acker said. “I’m always going to pick the one for which I have free credits.”