Sam Altman's OpenAI vision: trillion-dollar compute deals, affiliate revenue, and the path to financial self-sufficiency

Summary

Sam Altman is repositioning OpenAI as a hyperscaler, betting the company can fund a trillion-dollar infrastructure build-out through its own revenue within years. OpenAI has signed compute deals with Broadcom, Oracle, Nvidia, AMD, SK Hynix, and others that the Financial Times valued at $1 trillion. When Ben Thompson asked how OpenAI would pay for this, Altman said OpenAI's revenue would cover it. That revenue will come from multiple sources beyond ChatGPT subscriptions.

Subscriptions and transaction revenue

OpenAI's paid subscriber base is growing, with some users paying $200 monthly, but the addressable market for standalone subscriptions is likely finite. The company is preparing to monetize its 800 million weekly active users through affiliate and transactional revenue. Altman signaled openness to ads, but only where they add genuine value. He cited Instagram as his model, where ads help users discover products they want. Search ads, by contrast, he views as attacks because organic results already show what users need, then ads fight for placement above them.

Affiliates appear inevitable. Users can already search for products in ChatGPT, get links to retailers, and complete purchases without leaving the platform. OpenAI currently captures none of that transaction value. Once Shopify and Amazon integrations mature and OpenAI begins taking a cut, it could generate billions in incremental revenue as subscriptions plateau. That monetization could switch on rapidly within months once affiliate partnerships are disclosed and live at scale.

The infrastructure bet

The trillion-dollar plan depends on execution. Stacy Rasgon at Bernstein Research highlighted structural risks. AMD has never built racks at the scale OpenAI requires, and the chips do not exist yet. More troubling is the circular dependency. OpenAI is guaranteeing to buy compute output from vendors, who are then investing their own capital to build that capacity, betting OpenAI's revenue forecasts hold. Nvidia at least receives OpenAI equity; AMD receives only future revenue from deals that may or may not materialize.

The broader market risk is starker. The entire AI money machine—Nvidia's valuation, AMD's chip bets, Oracle's infrastructure deals, and xAI's parallel compute plans—rests on the assumption that Altman's vision compounds. Brent Donnelly framed it bluntly: the stock market depends on an Ouroboros that keeps feeding itself forever. Everyone thinks they can exit before everyone else does.

What remains unclear

Altman told Thompson to wait a few months for clarity on how all these partnerships fit together. Several open questions matter. Is OpenAI building proprietary chips, its own cloud platform, or both? Is the new compute going toward bigger models and reasoning breakthroughs, or toward efficiency gains that let them serve cheaper, smaller models? If progress stagnates, does the economic value of diffusing existing capabilities through cheaper inference still justify the spend?

Altman has credibility from prior delivery. OpenAI passed a reasonable definition of the Turing test by public conception, despite hype cycles. But the trillion-dollar infrastructure bet is qualitatively different from shipping ChatGPT. It requires sustained revenue growth, market share defense against Google and Anthropic, and vendors betting billions on forecasts they cannot fully verify. The plan may be solid. Waiting months to hear it is reasonable. The market has already priced in success.