The Leaked Audit That Exposes OpenAI's Real Financial Crisis

Summary of the video’s main claims and analysis

The video argues that leaked or otherwise openly verified-looking audited financial documents (confirmed by The Financial Times) reveal OpenAI is under severe financial strain despite its outsized valuation and claims that AI is a “cash machine.” It further claims OpenAI’s investor projections repeatedly move in a more pessimistic direction, with massive annual losses extending through 2028 and profitability only projected around 2029–2030.

Broken profitability promises

• OpenAI allegedly told investors it would reach roughly $175B revenue by 2030, while spending close to $200B to get there.
• The revenue target later rose to $280B by Feb 2026, but projected losses/burn allegedly grew too—eventually cited at $665B.
• The video characterizes these as “spreadsheets with good intentions,” arguing the direction stays the same: increasing losses and spending.

Audit-reported scale of losses

• In 2025, OpenAI’s revenue is claimed to have nearly quadrupled to over $13B, yet it reportedly posted a net loss of $38.5B.
• The video claims this loss exceeds what OpenAI had made in its entire earlier history combined.

“Efficiency improvements” may be misleading

• OpenAI’s defense is framed as improving cost-per-dollar (efficiency ratios).
• The video claims the ratio improved (e.g., $2.25 spent per $1 earned in 2024 down to $1.60 in 2025), but argues that this can happen alongside worsening absolute losses.
• It cites early-2026 figures where revenue growth is contrasted with very large cash burn and other large charges (including restructuring-related non-cash items), totaling a claimed $21.3B loss in a single quarter.

Internal targets reportedly missed; leadership friction

• The video claims the Wall Street Journal reported OpenAI missed internal projections for revenue and weekly active users in early 2026.
• It also alleges CFO Sarah Friar was reportedly telling people internally that OpenAI might not be “IPO-ready.”
• The video further claims Friar’s reporting line shifted—she is described as now answering to Fidji Simo (CEO of AGI Deployment)—implying internal friction over burn rate and performance.

Accounting assumptions about GPU “useful life” may inflate results

• The video disputes standard depreciation assumptions for GPUs (Nvidia-class hardware), arguing real hardware becomes obsolete faster than accounting estimates.
• It cites Michael Burry’s view that the industry may be hiding around $176B of “real expenses” between 2026–2028 by stretching depreciation periods.
• It notes counterarguments (industry firms dispute Burry), but emphasizes uncertainty and the lack of long-track records at this unprecedented scale.

“Circular financing” loops that may make spending look like demand

The video describes alleged vendor-financing / revenue-round-tripping dynamics:

• Nvidia → OpenAI → Oracle → Nvidia (chips and cloud spend moving in a loop).
• It claims Nvidia’s announced “up to $100B” commitment largely stalled (citing the Wall Street Journal), shrinking to around $30B.
• It argues that when companies pass money among themselves, record infrastructure spending can be mistaken for record end-user demand.

Microsoft–OpenAI funding loop reinforces the cycle

• The video claims Microsoft’s AI revenue is heavily tied to OpenAI, stating that ~70% of Microsoft’s AI revenue traces back to OpenAI.
• It argues Microsoft funds OpenAI so OpenAI can pay for compute, and that compute bills return to Microsoft via Azure—reinforcing the loop.

Ads-as-a-savior faces major market and growth hurdles

• The video claims OpenAI increasingly relies on advertising to hit long-term targets, after previously relying more on subscriptions.
• It says OpenAI generated $2.5B ad revenue in 2026 and projected $11B in 2027, aiming for $100B annual revenue by 2030.
• It questions feasibility: OpenAI would need weekly users far beyond current scale, and the projected U.S. chatbot ad market is argued to be too small for OpenAI’s claimed slice.

Massive power/compute requirements (“Stargate” and grid constraints)

The video argues that electricity demand is a bottleneck:

• It cites power requirements per GPU and estimates for data-center clusters.
• It claims OpenAI’s “Stargate” (SoftBank + Oracle joint venture) could spend up to $500B on data centers over 4 years, plus an additional $250B Azure commitment through 2032—totaling ~$600B compute spend through 2030.
• It alleges delays:
  • A 2-gigawatt expansion plan in Abilene, Texas was reportedly canceled.
  • Grid power connection approvals in some regions allegedly take 4–7 years, threatening timelines for 2028–2029 needs.

Funding terms include AGI/IP/IPO-dependent “strings”

• It discusses OpenAI’s March 2026 $122B funding round, valuing OpenAI at $852B.
• It claims a key portion of Amazon’s $50B commitment is contingent on OpenAI either:
  • going public by end of 2028, or
  • achieving AGI
• It also claims there are conditions tied to achieving AGI that could affect Microsoft’s exclusivity arrangement.

Capped-profit structure was effectively loosened; IPO is being pursued

• The video claims OpenAI initially positioned itself as “capped-profit” (investor returns capped at 100x), but the cap was later modified with a 20% annual increase, making it effectively meaningless by 2025.
• It says OpenAI converted into a public benefit corporation in Oct 2025 and filed IPO paperwork with the SEC in May 2026, aiming for listing as early as Q4 2026 at $852B–$1T valuation.

Real-world product performance doubts

• The video includes examples suggesting AI deployments can regress or underperform, including:
  • Klarna, which replaced human support agents with an OpenAI-built assistant and saw satisfaction drop before rehiring humans.

---

Overall conclusion

The video frames OpenAI’s situation as a multi-factor bet with several uncertain dependencies occurring simultaneously—funding arriving on schedule, depreciation accounting matching reality, circular financing remaining viable, ads materializing, and electricity staying cheap. It warns that if these bets don’t all land, confidence could collapse, likening the risk to familiar “bubble vs. dot-com crash” dynamics (even if separated by decades).

---

Presenters / contributors mentioned

• Sam Altman (mentioned)
• Ed Zitron (journalist referenced as receiving leaked documents)
• Fidji Simo (named in relation to CFO reporting)
• Sarah Friar (CFO referenced)
• Michael Burry (mentioned)
• Jensen Huang (Nvidia CEO referenced)
• Oracle (company referenced)
• Microsoft CEO (not named; admitted data center slowdown)
• SoftBank (company referenced)
• Klarna (company example)
• Financial Times (confirmed documents as real)
• The Wall Street Journal (reported missed projections and funding deal issues)
• eMarketer (market estimate referenced)