Anthropic and OpenAI at the Edge: The Race for Profit Becomes a Question of Survival

The world's largest AI companies face a choice: become profitable or burn out along with their investors' money. Commentators call it the "monetization cliff": the moment when years of losses must finally give way to a real business.

Anthropic and OpenAI are built on hundreds of billions of dollars in venture funding. At the same time, they have launched multibillion-dollar programs to build data centers, buy chips and expand infrastructure. The logic is simple: at some point profit has to appear, or the bubble bursts. Most CEOs in the industry make no secret of it: some players will fail spectacularly, some will survive, but the sums invested are too large for anyone to walk away voluntarily.

AI agents became the tipping point. Products such as Anthropic's Claude Code and Cowork, along with the open-source OpenClaw and OpenAI's Codex, changed how much compute companies think they need. Agents are valuable to users right now: they carry out tasks autonomously, write code and manage workflows. But that value has a price: agents burn through tokens many times faster than ordinary chatbot queries. The companies were not ready for this rate of consumption.

It is already forcing hard decisions. A month ago, OpenAI abruptly shut down Sora, its video generation app, despite a licensing deal with Disney worth $1 billion. The reason was prosaic: the service demanded too much compute, and Codex needs it. Product profitability mattered more than a high-profile partnership.

Anthropic acted similarly but hit a different audience. The company announced that users who run the OpenClaw framework heavily through a standard subscription can no longer do so on the previous terms. They are being moved to a pay-per-use model, which is substantially more expensive. In effect, the company has introduced price discrimination for its most resource-hungry customers.

Both moves are symptoms of the same ailment. The companies must manage a shortage of compute at a time when demand is growing faster than infrastructure. The agentic use case turned out to be far more expensive than planned.

On the horizon are the two largest IPOs in the history of the tech industry. Anthropic and OpenAI are preparing to go public, and the pressure on profitability metrics has never been higher. This week, financial projections for both companies came to light: hundreds of billions of dollars in revenue and profitability by the end of the decade. But those very figures raise the central question: are they realistic?

The answer depends on several variables. Will the companies be able to sustain their current growth rate? Will enough enterprise customers be willing to pay for agents at the necessary scale? Will competition from open models erode the price premium on which these forecasts rest?

What is happening is not just management decisions at two startups. It is the moment when the AI industry stops being a story about technology and becomes a story about money. Every product shutdown and every tightening of pricing policy signals to the market that these companies can no longer afford unprofitable growth. Investors expect a return, the market expects profit, and the entire developer ecosystem is watching to see who stays standing and who goes over the edge.