OpenAI and Anthropic just hit a wall. The two AI giants are making sudden, drastic moves - killing popular products, overhauling pricing, scrambling for profitability - as they race toward what could be the biggest IPOs in tech history. The culprit? AI agents are burning through compute resources faster than anyone anticipated, and the pressure to deliver returns on hundreds of billions in investment has never been more intense. This is the AI industry's make-or-break moment.
The AI boom is hitting its most critical inflection point yet. OpenAI and Anthropic, the two companies that define the frontier of artificial intelligence, are suddenly making the kind of desperate moves that signal something fundamental has changed. They're not just optimizing anymore - they're fighting for survival.
Last month, OpenAI did something almost unthinkable: it killed Sora, its highly-anticipated video generation product, and walked away from a $1 billion licensing deal with Disney in the process. The reason, according to reporting from The Verge? Sora cost too much to run, and OpenAI needed every available GPU for Codex, its AI coding agent. When you're ditching billion-dollar partnerships to preserve compute resources, something's gone seriously wrong with your business model.
Then last week, Anthropic dropped its own bombshell. The company announced it would no longer allow Claude users to run OpenClaw - the popular open-source agent framework - on standard subscription plans. Instead, users got pushed onto pay-as-you-go pricing that costs substantially more. The Verge reported the move was essentially a ban disguised as a pricing change, and the developer community erupted.
What's driving these sudden pivots? AI agents. Products like Claude Code, Cowork, and OpenAI's Codex have become the killer apps these companies desperately needed - real tools that developers and enterprises actually want to pay for. But there's a brutal catch: agents consume compute at rates 10 to 20 times higher than standard chatbot queries. When someone uses an agent to write code or automate workflows, it's burning through tokens - and therefore GPU cycles and electricity - far faster than these companies ever modeled in their projections.
"The way people are using agents is burning tokens at a rate way faster than these companies anticipated," said Hayden Field, senior AI reporter at The Verge, on the Decoder podcast. "And that's causing them to make hard decisions."
Those hard decisions are coming at the worst possible time. Both OpenAI and Anthropic are barreling toward IPOs that would rank among the largest in tech history. OpenAI just raised another $122 billion at an $850 billion valuation, according to previous Verge reporting. But that astronomical valuation comes with astronomical expectations.
This week, The Wall Street Journal leaked financial projections from both companies that paint a picture of almost incomprehensible growth - hundreds of billions in revenue and profitability by the end of the decade. To hit those numbers, everything has to go exactly right. Instead, compute costs are spiraling, popular products are getting axed, and the path to profitability looks increasingly treacherous.
The infrastructure costs alone are staggering. The Wall Street Journal reported on the spiraling expense of building and running AI systems at this scale - data centers, chips, energy, cooling systems. Every agent query chips away at margins that don't really exist yet. OpenAI and Anthropic have raised hundreds of billions in capital, but they're also linked to even greater amounts of forward-looking investment in infrastructure. At some point, the profits have to materialize, or the bubble pops.
The internal chaos is starting to show. Wired reported that OpenAI COO Fidji Simo is taking a leave of absence amid executive shake-ups, while The Verge noted that the vibes inside OpenAI have shifted noticeably. When you're making brutal resource allocation decisions - choosing which products live and die based purely on GPU availability - company culture takes a hit.
OpenAI even acquired TBPN, a smaller AI company, as The Verge reported, in what looks like an acqui-hire to bring in talent as the company restructures. These aren't the moves of a confident market leader - they're the moves of a company under immense pressure to prove its business model works.
The market pressure isn't just coming from investors. Public sentiment toward AI has soured dramatically. An NBC News poll showed voters like AI less than ICE, the immigration enforcement agency. When your technology is less popular than one of the most controversial government agencies, you've got a perception problem that makes going public even harder.
What makes this moment so critical is that AI agents actually work. They're not vaporware or overhyped demos - developers are using Claude Code and Codex to ship real products. Enterprises are adopting these tools. The product-market fit is there. But the economics might not be. If the cost of delivering the service exceeds what customers will pay, even great products fail.
Anthropic's pricing overhaul is a direct admission of this problem. By forcing agent users onto pay-as-you-go plans, the company is essentially saying: we can't afford to let you use our best features at a flat subscription rate. That's not how successful SaaS businesses typically operate. It's a Band-Aid on a business model that's bleeding cash.
The compromises keep stacking up. What other products will OpenAI kill to preserve compute for profitable use cases? How much will Anthropic have to charge to actually make money on agent queries? And can either company hit the growth projections they've promised investors while simultaneously restricting access to their most valuable features?
The AI industry's existential moment has arrived. OpenAI and Anthropic built products people actually want to use, but they're discovering that success at scale might be economically impossible under their current models. The pressure to go public and deliver returns on hundreds of billions in investment is forcing both companies into increasingly desperate moves - killing products, overhauling pricing, rationing compute. Whether they can navigate this monetization cliff without crashing and burning will determine not just their own fates, but the trajectory of the entire AI industry. The next few months will reveal whether these companies can turn astronomical valuations into actual profits, or whether we're watching the beginning of one of tech's most spectacular flameouts.
