AI Companies Are Burning Cash on Free Credits to Buy Market Share, Reports Show

Top AI companies are giving away enormous amounts of computing power to win over startups and enterprise customers, according to the Wall Street Journal. Cursor, the AI-coding company, offered a 75% discount that ran through July 5, 2026, per the Journal's reporting.

Hans Ibarra, a founder building an AI-voice startup, told the Journal he's on the receiving end of a bidding war. OpenAI, Anthropic and other model makers are competing for his business with computing credits and discounts, he said.

Another founder identified only as Acker put it bluntly to the Journal: given a choice between a cheap Chinese AI model he'd have to pay for and an expensive Anthropic model offered free, he takes the free one every time. "I'm always going to pick the one for which I have free credits," he said.

The Cost Argument

Separately, Forbes has reported that AI systems can cost more than the human workers they're meant to replace once you factor in the labor needed to check and fix AI output. The premise: running large models is expensive, and the corrections required when AI gets things wrong add labor costs back on top of the compute bill.

Charles Hugh Smith, writing on his OfTwoMinds blog and syndicated by ZeroHedge, ties these two threads together into a broader claim: that AI's low or free pricing for users is a subsidy, not a sustainable price, and that the technology is "unaffordable" once those subsidies disappear. His reasoning is straightforward. Training and running large models requires enormous electricity, processing power and memory. Something that expensive to operate can't actually be cheap for the end user unless someone else is eating the cost.

Smith frames the discounting war as a race for what he calls network effect dominance, the idea that whoever locks in the most users first can eventually raise prices once switching costs make it painful for customers to leave.

What's Proven and What Isn't

The discounting and free-credit giveaways described by the Journal are documented and real. Cursor's 75% discount, the competing offers founders like Ibarra are fielding, these are on-the-record facts.

What's not documented, at least not in any of the reporting cited here, is exactly how much money OpenAI, Anthropic or their peers are losing per user, or what their actual unit economics look like once subsidies are stripped out. None of these companies have published detailed cost-per-query breakdowns. Smith's conclusion that AI is broadly "unaffordable" is an inference built from public discounting behavior and anecdotal cost comparisons, not a disclosed financial statement.

There's a legitimate counterargument worth stating plainly. Subsidized pricing to capture market share is a well-worn Silicon Valley playbook, not unique to AI. Uber lost billions for years subsidizing rides before eventually raising prices. Amazon ran razor-thin or negative margins on retail for over a decade while building logistics dominance. Streaming services burned cash on content for years before some turned profitable. A company giving away compute today doesn't prove the underlying technology is permanently uneconomic. It may just mean the market is in its land-grab phase, and prices could stabilize as chip costs fall, models get more efficient, and competition consolidates.

Both things can be true at once: the discounting is real and aggressive, and it's still an open question whether AI pricing today reflects a temporary subsidy war or a permanently unworkable cost structure.

What to Watch

Neither OpenAI, Anthropic, nor Cursor has disclosed specific loss-per-customer figures tied to these promotions. None of the companies named in the Journal's reporting have said publicly when or whether discounts will end. The real test comes if and when these companies start raising prices toward their true operating costs, and whether the customers locked in today, founders like Ibarra and Acker among them, stick around once the free credits run out.