The next chapter of the Microsoft–OpenAI partnership

Original Article ↗ Hacker News Discussion ↗

Deal structure & Microsoft’s position

  • New terms: Microsoft’s stake drops to ~27% at a ~$500B OpenAI valuation, while OpenAI commits to an additional $250B of Azure spend and extends Microsoft’s IP rights over models/products through 2032, including “post‑AGI” models.
  • Some see this as a loss of prior advantages (e.g., losing compute exclusivity / right of first refusal); others argue the locked‑in $250B Azure revenue and ongoing IP rights are a strong win, especially if OpenAI never reaches AGI under the contract’s definition.
  • A common interpretation: Microsoft is de‑risking a very speculative bet while ensuring it still benefits if OpenAI succeeds.

AGI definition, declaration & “expert panel”

  • The clause that AGI must be “declared” by OpenAI and then verified by an “independent expert panel” is widely mocked and seen as fundamentally political, not scientific.
  • Commenters note prior reporting that Microsoft and OpenAI once tied AGI to $100B in profit, calling this Goodharted, financially motivated, and reminiscent of Tesla’s “Full Self Driving” rebranding.
  • Many emphasize that AGI has no agreed‑upon technical definition; any panel’s judgment will depend on who sits on it and their incentives.

Is AGI near?

  • Views range from “AGI is already here in a minimal sense” to “we’re nowhere close and LLMs are just advanced pattern matchers.”
  • Arguments against nearness: lack of robust reasoning, inability to handle out‑of‑distribution tasks, failure on long‑horizon autonomy, and the fact that even self‑driving remains brittle.
  • Others say previous timelines for LLMs were badly wrong in the conservative direction, so it’s honest to admit “we don’t know,” though most still doubt short (<5–10 year) timelines.

Non‑profit mission, governance & “greatest theft”

  • The recapitalization into a PBC and unified traditional equity is described by several as effectively stripping the original non‑profit of control and converting a “for humanity” charter into a $500B private asset.
  • Some call it “the greatest theft from mankind,” arguing the non‑profit has handed over a unique public asset to private shareholders with minimal public accountability.

Profitability, compute commitments & bubble fears

  • OpenAI is said to be committed to roughly $1.4T in compute (Azure, Oracle, NVIDIA, etc.) while currently earning on the order of ~$10B/year in revenue; many doubt any realistic path to pay for this.
  • Multiple commenters compare the situation to dot‑com, NFTs, or Enron‑style financial engineering: capital recycling between hyperscalers and labs to pump valuations.
  • Concern is voiced that LLMs are not yet profitable enough to justify this scale, raising risk of a major AI bubble and broader economic fallout, including energy/climate impacts.

Cloud, open weights & competition

  • The revised deal lets OpenAI:
    • Use other clouds for non‑API products.
    • Jointly develop products with third parties.
    • Release some “open‑weight” models below certain capability thresholds.
  • This is read as a loosening of Microsoft’s stranglehold and a response to pressure from competitors (Anthropic, Google, open‑weight players, Chinese models).
  • Some think even frontier‑quality open weights wouldn’t kill OpenAI’s business but could be used to block competitors’ service‑layer moats.

Consumer hardware & government/defense angle

  • Excluding consumer hardware from Microsoft’s IP rights and prior Jony Ive involvement fuel speculation about AI wearables or post‑phone devices; others are skeptical given the difficulty of that market.
  • A new clause explicitly allowing OpenAI to serve US national security customers on any cloud raises concern that “unaligned” or lightly aligned models will be tailored for military and surveillance use as a major revenue stream.

Broader sentiment on hype & terminology

  • Many see “AGI” in these documents as a pure business lever: a contractual milestone and investor story more than a coherent technical concept.
  • Comparisons to Tesla FSD, marketing‑driven redefinitions of “AI,” and prior hype cycles are frequent. Some are simply waiting for the AI/AGI bubble to pop; others think we’re still early in a long, messy boom.