Oracle is underwater on its $300B OpenAI deal

Original Article ↗ Hacker News Discussion ↗

Perception of the Oracle–OpenAI Deal

  • Many see the “$300B” plan (massive capex over years for OpenAI capacity) as irrational relative to OpenAI’s current ~$20B revenue and lack of profit.
  • Commenters stress Oracle gets little or no IP: it’s mostly buying Nvidia boxes, racking them, cooling them, and earning a modest markup.
  • Counterparty risk is a core concern: Oracle may build and finance infrastructure and then not get paid if OpenAI stumbles.
  • Others argue that as a cloud provider Oracle is “selling shovels” and could in theory re-sell GPU capacity to other AI users, but skeptics doubt there will be enough profitable demand for a 10x datacenter build-out.

AI Bubble, Overcapacity, and Money Destruction

  • Strong sentiment that AI resembles a speculative bubble, like crypto or dot-com, with huge valuations built on projections of 50–75% annual growth for years.
  • Some argue AI infra is a way to “burn off” excess money created in the last decade; others push back, noting you can destroy wealth but not the money supply.
  • There’s concern of a coming GPU glut: once subsidies and loss-leading free tiers end, demand and pricing might not sustain current capex, leaving “$300B of shovels” earning far less than expected.

Oracle’s Core Business and Survival

  • Several note Oracle’s legacy database business still “prints money” from locked-in customers; few new firms choose Oracle, but existing deployments are sticky and expensive to replace.
  • This leads to a split view: for some, Oracle is the weak link when the AI bubble bursts; for others, the DB cash cow plus Chapter 11–style restructuring means the company survives even if the AI bet fails.

Market Reaction and Valuation Debate

  • Oracle’s stock spike on the OpenAI announcement and subsequent drop are seen as classic hype-and-cooldown; tying a $300B multi-year plan to a few months of price action is viewed as flimsy.
  • Some argue “underwater” based on lost market cap is rhetorical; real judgment must wait on actual returns.
  • Thread devolves into broader arguments about shorting, “skin in the game,” bubble talk vs. actionable insight, and whether tech firms should return excess cash via dividends/buybacks rather than mega-bets.

Competition and AI Economics

  • Multiple comments suggest Google may outlast or out-execute OpenAI: it has huge profits, its own chips (TPUs), the search/crawler data pipeline, and can wait out others.
  • Others counter that LLMs are increasingly commoditized; brand and adoption (ChatGPT) may matter more than marginal model quality.
  • A major open question: can AI chat ever be profitably monetized (especially with ad models) at the compute cost levels implied by these infrastructure builds? Many commenters say this remains unclear or unlikely at present.