OpenAI projects $278B negative FCF through 2030

Opening

OpenAI’s own planning deck shows ~$278 billion of negative free cash flow from 2026 through 2030, the Financial Times reported Friday. Reuters relays the FT account of a company presentation: negative FCF of $278 billion over 2026–2030 while securing compute to train and run models. Revenue is projected to rise from $36 billion in 2026 to $350 billion in 2030 (~$840 billion cumulative through decade-end). Compute and infrastructure spend is pegged at about $856 billion by end-2030 — the largest expense line. The FT said the firm raised $122 billion in March at an $852 billion valuation and is on track to exhaust that cash by 2028. Reuters could not immediately reach OpenAI for comment outside business hours.

The numbers

−$278 billion projected FCF, 2026–2030 — FT (company presentation).

~$856 billion compute/infra spend by end-2030 vs ~$840 billion cumulative revenue — FT.

Revenue path $36 billion (2026) → $350 billion (2030) — FT.

$122 billion raised March at $852 billion valuation; cash runway to ~2028 — FT via Reuters.

Why it matters

The load-bearing claim is that multi-year AI revenue growth still leaves OpenAI deeply cash-negative through the decade because compute/infra outlays roughly match cumulative revenue — so IPO or pre-IPO capital has to plug a structural hole, not a one-quarter miss. @JaredKubin framed the same print as compute and revenue “almost cancel,” with the full −$278B burn as “everything else,” and ~$150B of IPO/pre-IPO proceeds needed to plug the gap.

Positioning

AI capex durability — The deck assumes hyperscale compute spend continues at a scale that dominates the P&L; a break would be sequential hyperscaler or lab guidance that cuts those capacity dollars, not a headline about “burn.”

Connects to

Baker pacing / margins — Whether “pace the frontier” reallocates spend inside this burn path or changes its size (discussion).

SoftBank–Arm margin loan — Same week’s financing plumbing around the AI stack (news).