Video summary

AI Companies Are Setting Money on Fire | OpenAI's Pre-Bailout Bailout Plan

Main summary

Key takeaways

Finance

Macro / Investment Context (AI “Bubble” Framing)

  • The video argues that AI infrastructure spending resembles an “AI bubble”: capital is deployed at scale before profitability is achieved.
  • It claims that companies and the public are becoming increasingly skeptical of AI ROI, driven by:
    • Rising data center costs
    • Opposition and project cancellations
  • The thesis is that OpenAI’s survival could become “systemic,” potentially involving:
    • Large tech partners
    • The U.S. government absorbing losses (described as a “pre-bailout bailout plan”).

Key Corporate Financial Claims (OpenAI)

Revenue and Losses

(Attributes figures to Financial Times–verified reporting via Ed Zitrin, as presented in the subtitles.)

  • 2025
    • Revenue: ~$13.0B
    • Losses: ~$38.0B (also stated as $38.5B elsewhere)
  • 2024
    • Revenue: ~$3.7B
    • Losses: ~$5.09B
  • The loss is described as nearly an 8x year-over-year increase.

Costs and Spending (2025 totals)

  • Total costs & expenses: ~$34B
  • R&D: ~$19.18B
  • Sales & marketing: ~$5.73B
  • Assets by end of 2025: ~$50B total assets, with about half in cash

Monthly Revenue

  • By end of 2025: ~$2B/month
  • Up by ~$1B/month versus end of 2024

Customer / Related-Party Payments (reported in subtitles)

  • SoftBank → OpenAI: $867M (2025)
  • Microsoft → OpenAI: $33M (2025)
  • OpenAI → Microsoft: “much more,” including:
    • ~$10.59B reportedly categorized as OpenAI R&D with Microsoft (subtitles suggest this may correspond to training)
    • Another figure framed as OpenAI expenses to Microsoft ~ $17.2B after adding additional costs (“circular deal” framing)

“Non-cash” Accounting Charge

  • Subtitles claim a large portion of the loss jump was due to a non-cash counting charge from OpenAI’s previous structure:
    • Estimated at ~$30B
    • Reported as not expected to repeat after restructuring

Implied Valuation / Capital Raises (Funding Rounds)

  • Feb 2026
    • Raised $110B more
    • ~$730B pre-money valuation
    • Named funding sources: Nvidia and Amazon
  • March 2026
    • Valuation raised to ~$852B
    • Support cited from Amazon, Nvidia, SoftBank, and Microsoft
  • The video also references:
    • OpenAI confidentially filed an S-1 with the SEC on June 8, 2026
    • Mention of IPO timing possibly shifting to 2027 (per later press)
    • Restructuring timeline spanning 2025 and finishing by Oct 2025

Revenue Goal / Profitability Target

  • OpenAI reportedly targeted > $280B revenue by 2030
  • The video argues this goal is far above current run-rate, implying a need for:
    • Very large capacity expansion and/or
    • Much stronger monetization
  • Contrast given:
    • Current run-rate by end-2025: ~$2B/month (~$24B/year)
    • To reach $280B/year, it says OpenAI would need > $23B/month average by 2030

Compute / Capacity Buildout Plans (“Utility” Model)

“Compute as a Utility” Framing

  • OpenAI is repeatedly framed as building “compute as a utility,” including the idea of an OpenAI token subscription plugged into everything.

Stargate / Data Center Projects and Partners

  • July 2025: OpenAI + Oracle
    • Develop ~4.5 GW capacity
    • Total ~5 GW
    • ~2 million chips (as cited in subtitles)
  • Wall Street Journal (as referenced in subtitles): Oracle + OpenAI $300B cloud deal
  • Nvidia investment cycle
    • Sept (implied 2025): Nvidia said it would invest up to $100B in OpenAI
      • For deployment of at least 10 GW of Nvidia systems
    • Deal “fell apart,” then Nvidia reportedly settled at ~$30B about 6 months later
    • “Circular deal” claim:
      • OpenAI would use Nvidia-backed funding to rent 22.5B GPUs from CoreWeave
      • Subtitles state Nvidia has part ownership in CoreWeave
  • AMD / Broadcom allocations
    • OpenAI to deploy ~6 GW with AMD
    • Partnered with Broadcom for ~10 GW of AI accelerators

Microsoft / Amazon Deals

Subtitles claim OpenAI:

  • Will spend $250B more with Microsoft
  • Agreed to a $38B deal with Amazon Web Services (AWS)
  • It states Microsoft and Amazon each have “tens of billions” invested in OpenAI

Updated Funding-to-Compute Details (Feb 2026 round)

  • OpenAI said it would use:
    • 3 GW for inference
    • 2 GW for training
    • On Nvidia “Ver Rubin” systems
  • Also:
    • Use 2 GW on AWS
    • Amazon said it would invest up to $50B in OpenAI

Compute Spending Scale and Scaling Back

  • Earlier ambition (Oct 2025, Altman quoted in subtitles):
    • ~$1.4T compute capacity
    • Commitments totaling ~30 GW over “many years”
  • CNBC (Feb 2026, referenced in subtitles):
    • OpenAI targeted ~$600B total compute spend by 2030
    • Projected 2030 revenue > $280B
  • Europe / US project changes:
    • Stargate UK and Norway: progress reportedly ceased in April 2026
    • Texas Stargate expansion with Oracle: reportedly abandoned due to financing challenges and changing demand forecasts (Bloomberg-style language in subtitles)

Other Financial Instruments / Equity Deals Mentioned

Warrants / Equity-Linked Arrangements

  • AMD deal (Oct 2025):
    • AMD issued OpenAI a warrant for up to 160M shares
    • Rationale implied: AMD equity could rise as OpenAI buys more AMD GPUs
    • Subtitles say an AMD SEC filing indicated none of the warrant shares had vested/exercisable yet

Broadcom / Chip Financing Uncertainty

  • Subtitles claim Broadcom and OpenAI faced financing hurdles for the AI chip announced in 2025.

Timeline Drift / “Softened” Language

  • OpenAI + Broadcom planned an inference chip called “Jalapeno”
    • Initial deployment by end of 2026
    • Expand “in the years ahead”
    • Earlier wording cited 10 GW deployment complete by 2029, with subtitles suggesting later language indicates backtracking risk.

Performance / Metrics Discussed (and Potential “Gaming”)

“Compute Tracks Revenue” Claim

  • OpenAI CFO claim (Jan 2026, via subtitles):
    • Available compute grew 3x YoY
    • ~9.5x from 2023 to 2025
      • 0.2 GW (2023) → 0.6 GW (2024) → 1.9 GW (2025)
    • Revenue similarly grew:
      • ~$2B ARR (2023) → ~$6B (2024) → > $20B (2025)

Critique of ARR (Annualized Recurring Revenue)

  • Subtitles (via “Zitrin”) state:
    • ARR is an estimate based on a partial-year run rate extrapolated to 12 months
    • It can be misleading/abused by:
      • Choosing strong months
      • Ignoring seasonality

Explicit Recommendations / Cautions (as Stated)

  • The video offers no direct buy/sell investment recommendations
  • Instead, it repeatedly cautions about:
    • Doubts about AI ROI and claims AI usage may be scaled back
    • Revenue metrics such as ARR being potentially manipulated
    • Risk of roadmap slippage due to vague contract language (e.g., “no assurance transaction will be completed,” “negotiations,” etc.)
    • The idea that OpenAI’s losses could require absorption by broader stakeholders, including government

Disclosures / Disclaimers

  • Subtitles mention discussion about the video host/supporting their work.
  • No clear “not financial advice” disclaimer appears in the provided text.
  • No formal regulatory disclaimer is included in the subtitles.

Tickers / Entities / Instruments Mentioned

Companies / Entities (major)

  • OpenAI, Nvidia, Microsoft, Amazon / AWS, AMD, Oracle, SoftBank, Broadcom
  • Also mentioned: Intel (“ask Intel”)

Infrastructure / Partners

  • CoreWeave

Regulatory / Filing / IPO Instrument

  • S-1 (SEC registration)

Systems / Chips

  • Nvidia “Ver Rubin” systems
  • AI chip name: “Jalapeno”

Market Research

  • Sensor Tower

Note: No ETF/bond/commodity tickers or price levels were included in the subtitles.

Framework / Methodology Items Mentioned

  • “Utility model” framework (conceptual)
    • Intelligence like electricity/water:
      • customers buy access via subscription/token
      • access is “plugged into everything” and runs continuously
  • “Revenue compute tracking” framework (reported claim by OpenAI CFO)
    • Revenue growth scales with available compute capacity (GW)
  • ARR calculation caveat (metrics framework)
    • ARR extrapolates a partial-year run rate to a full year
    • Criticism: can ignore seasonality and be “gamed”

Presenters / Sources Mentioned

  • Ed Zitrin (author/critic; “Better Offline”; blog figures cited)
  • Financial Times (verification of Zitrin’s reporting)
  • Wall Street Journal (Oracle cloud deal referenced)
  • New York Times (possible IPO timing to 2027 referenced)
  • Bloomberg (framing around financing challenges referenced)
  • CNBC (compute spend target and scaling back referenced)
  • SEC (confidential S-1 submission referenced)
  • Sam Altman (OpenAI CEO; quoted)
  • Jensen Huang (Nvidia CEO; quoted)
  • Gabe Newell and Valve (mentioned historically)
  • Peter Thiel (mentioned in biographical context)

Original video