Ed Zitron Calls AI a Con: Fact-Checking OpenAI Losses and 2027

Ed Zitron spent two and a half hours in the Diary Of A CEO hot seat defending one thesis: generative AI is, at heart, a con. The episode, published on August 27, 2026, has been watched more than 9 million times [1]. “I think generative AI is at its heart con and seeing these ultra rich ultra powerful people lie through their teeth turns my stomach,” he said. The climax is a warning that the AI industry all breaks in 2027. Tested claim by claim against official financial documents, most of the financial criticism holds and the headline numbers come straight from a real audit, but the “con” thesis and the 2027 collapse date do not survive the same data.

Method: the full 31,000-word transcript was read, every quantitative claim was tested one by one against 21 primary and credible secondary sources, and 21 of them were locked in as verbatim quotes. Status labels used: supported, supported with correction, contested, unsupported. The scorecard first.

Claim in the video	Status	What testing found
“OpenAI lost $20.9 billion last year”	Supported with correction	The 2025 operating loss was indeed $20.92 billion; the attributable net loss is larger, $38.53 billion [2]
Microsoft AI revenue $34.3 billion, $24.1 billion of it from OpenAI	Supported	$24.1 billion from a regulatory document, $34 billion a Bloomberg estimate [3]
A $200-per-month subscription subsidizes $14,000 in tokens	Supported	SemiAnalysis calculation for ChatGPT Pro 20x [4]
Uber burned a full year of token budget in three months	Supported with correction	The 2026 Claude Code budget ran out in four months, by April [5]
Driverless cars are already safer than humans	Supported	Waymo crash rates 68% lower per mile traveled (IIHS) [6]
OpenAI and Anthropic carry $1.1 trillion in compute commitments	Supported	The same figure appears in Zitron's own research [16]
OpenAI's last funding round valued it at $865 billion	Unsupported	The February 2026 round: $730 billion pre-money, $840 billion post-money [8]

The OpenAI loss number hanging on one sentence

The most quoted line of the video is one sentence: “OpenAI lost $20.9 billion last year because people can burn as many tokens as they want” [1]. The audited 2025 documents, published by Zitron himself on June 15 and confirmed by the Financial Times, show the number is real, but the label matters: $20.92 billion is the operating loss, annual costs minus $13.07 billion in revenue [2]. The net loss attributable to OpenAI is larger, $38.53 billion, because it absorbs a $41.55 billion non-cash charge from the restructuring into a for-profit corporation. So “lost $20.9 billion” is true of operations, while the fuller picture is worse than the claim. Here Zitron's narrative is politer than his numbers.

On Microsoft, Zitron cites $34.33 billion a year in AI revenue with $24.1 billion of it from OpenAI, then flips it: “so that leaves them with about $10 billion in a year when they spent 115 billion on capital expenditures” [1]. All three match the record: the $24.1 billion OpenAI contribution appears in a regulatory document and makes up roughly 70 percent of Microsoft's AI revenue, while $34 billion remains a Bloomberg estimate, not a company-announced figure [3]. Fiscal 2026 capex came in at $115.9 billion [14], and calendar-2026 capital spending is guided at about $175 billion, lowered from $190 billion after an accounting change for leases [15]. The correction is small but necessary: $175 billion is the calendar-2026 budget, not fiscal 2027, which is in fact expected to grow higher still.

The consumer subsidy claim also maps. By SemiAnalysis' calculation, a $200-per-month ChatGPT Pro 20x subscription equals $14,000 at API prices if used to the full, and OpenAI starts losing money on ChatGPT Plus once utilization passes 11.4 percent [4]. The corporate side follows the same pattern: Uber's 2026 Claude Code budget was gone by April, four months in, and COO Andrew Macdonald called token costs “harder to justify” because heavier usage shows no clear line to features actually shipped [5]. OpenAI's own research found no meaningful correlation between token consumption and revenue per employee at customer companies [10]. The subsidy is real; the financial payoff has not shown up.

Real growth on questionable foundations

None of this means the industry is empty. Nvidia closed fiscal 2026 with record revenue of $215.9 billion, up 65 percent [7]. Amazon committed $50 billion to OpenAI [12], Google prepared up to $40 billion for Anthropic [13], and a single Stargate campus in Abilene is planned at 1.2 gigawatts [11]. Zitron's own research notes that OpenAI and Anthropic carry $1.1 trillion in compute commitments contingent on continued growth [16], with 190 gigawatts of data centers globally planned or under construction according to the Sightline Climate data he cites [17]. The growth is real; what is contested is who pays for it.

This is where the title claim, “2027 is when it all breaks,” gets tested. The worry has measurable foundations: that $1.1 trillion is contingent on growth continuing [16], and the Bank for International Settlements warns a funding pullback could turn the capex boom into a prolonged investment bust [16]. Market signals exist too: OpenAI is leaning toward delaying its IPO to 2027 while chasing a $1 trillion valuation after its advisers sized the risk [9]. But between “measurable risk” and “collapse in 2027” stretches a canyon of prophecy. Status: contested. No primary data pins down a year of failure, just as no evidence charts a path to profitability.

What does not survive the con thesis

The portrait of a worthless product fails testing. ChatGPT has more than 900 million weekly users and 50 million paying subscribers, plus 9 million business seats [8]. Hallucination rates on simple summarization fell from 21.8 percent in 2021 to 0.7 percent on the best 2025 models [20], though a new generation of benchmarks remains far harder [21]. Driverless safety is not fiction: IIHS research finds Waymo 68 percent safer than human drivers per mile traveled [6]. Even the opposite of a con shows in public: Eric Schmidt was booed by graduates for mentioning AI at a commencement [18], and OpenAI shut down its Sora app in April 2026 for a reason far less flattering than deception: the product did not survive the market [19]. Industries that run cons do not usually retire their own products.

A few corrections remain. The “$865 billion” figure for the last funding round does not match the official record of the February 2026 round, $730 billion pre-money or $840 billion post-money [8]. Uber's token budget ran out in four months, not three [5]. The claim of a failed $500 million GPT-5 training run found no primary source and remains unverified. Zitron's reading of “annualized run rate” figures deserves the skepticism he gives it, since Microsoft rarely publishes final AI numbers, but that criticism stands on reporting practice, not proof of fraud [3].

The limits of this test deserve an honest record: OpenAI's audited documents were obtained through Zitron's own publication, confirmed by the Financial Times, rather than an S-1 filing that does not exist yet; the $34 billion Microsoft AI figure remains a Bloomberg estimate; and the 190-gigawatt figure comes from secondary coverage [2][3][17]. The conclusion: Zitron's financial criticism is the most durable part of this video. The numbers are genuine, and the structure he describes, cross-subsidy, contingent commitments, an IPO waiting for momentum, is documented in official filings. The word “con” requires intent to deceive, and the available evidence better fits an industry with subsidized demand and selectively reported metrics. The most exposed point this video reveals is not product quality but who actually pays the bill.

Sources:

[1] https://www.youtube.com/watch?v=Lf5oqGOCRCM — DOAC: The Man Who Calls BS On AI - Ed Zitron (2026-08-27, 2h27m, 9,007,093 views)

[2] https://www.techtimes.com/articles/318496/20260616/openai-lost-385-billion-2025-audited-financials-expose-17b-azure-dependency.htm — TechTimes: OpenAI Lost $38.5 Billion in 2025 - Audited Financials (2026-06-16)

[3] https://www.briefs.co/news/openai-payments-made-up-most-of-microsoft-s-ai-revenue — Briefs.co: OpenAI Payments Made Up Most of Microsoft's AI Revenue (2026-08-05)

[4] https://www.techspot.com/news/112759-openai-anthropic-cant-afford-have-everyone-use-ai.html — TechSpot: A $200 ChatGPT subscription could cost OpenAI $14,000 at full utilization (2026-06-14)

[5] https://qz.com/uber-coo-ai-spending-harder-justify-052626 — Quartz: Uber COO Andrew Macdonald says AI spending hard to justify (2026-05-26)

[6] https://www.iihs.org/news/detail/waymos-driverless-cars-crash-less-often-than-people — IIHS: Waymo's driverless cars crash less often than people (2026-07-23)

[7] https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-fourth-quarter-and-fiscal-2026 — NVIDIA Newsroom: Financial Results for Fourth Quarter and Fiscal 2026 (2026-02-25)

[8] https://openai.com/index/scaling-ai-for-everyone — OpenAI: Scaling AI for everyone (2026-02-27)

[9] https://qz.com/openai-ipo-delay-2027-valuation-062626 — Quartz: OpenAI may delay its IPO until next year (2026-06-26)

[10] https://fortune.com/2026/08/13/buried-in-openais-latest-research-no-correlation-between-ai-use-and-revenue-per-employee — Fortune: Buried in OpenAI's latest research - no correlation between AI use and revenue per employee (2026-08-13)

[11] https://www.crusoe.ai/resources/newsroom/crusoe-expands-ai-data-center-campus-in-abilene-to-1-2-gigawatts — Crusoe: Crusoe expands AI data center campus in Abilene to 1.2 gigawatts (2025-03-18)

[12] https://www.aboutamazon.com/news/aws/amazon-open-ai-strategic-partnership-investment — Amazon: OpenAI and Amazon announce strategic partnership

[13] https://www.cnbc.com/2026/04/24/google-to-invest-up-to-40-billion-in-anthropic-as-search-giant-spreads-its-ai-bets.html — CNBC: Google to invest up to $40 billion in Anthropic (2026-04-24)

[14] https://roboinvestor.ai/research/msft — RoboInvestor: Microsoft MSFT FY2026 10-K - $115.9B Capex, Azure Up 41% (2026-07-30)

[15] https://www.pymnts.com/earnings/2026/microsoft-extends-data-center-lifespans-to-soften-ai-buildout-costs — PYMNTS: Microsoft Extends Data Center Lifespans to Soften AI Buildout Costs (2026-07-29)

[16] https://www.wheresyoured.at/the-ai-industry-is-losing — Ed Zitron: The AI Industry Is Losing (2026-06-30)

[17] https://www.startuphub.ai/ai-news/market-research/2026/ed-zitron-big-tech-ai-boom-is-built-on-a-scandalous-lie — StartupHub.ai: Ed Zitron on Bloomberg - Big Tech AI Boom Is Built on a Scandalous Lie (2026-07-31)

[18] https://www.bbc.com/news/articles/ce8pqd54qneo — BBC: Former Google CEO Eric Schmidt booed by graduates at mention of AI (2026-05-18)

[19] https://help.openai.com/en/articles/20001152-what-to-know-about-the-sora-discontinuation — OpenAI Help Center: What to know about the Sora discontinuation

[20] https://aionx.co/ai-comparisons/ai-chatbot-hallucination-rates — AIonX: AI chatbot hallucination rates - Vectara HHEM benchmark data (updated 2026-09-16)

[21] https://www.vectara.com/blog/introducing-the-next-generation-of-vectaras-hallucination-leaderboard — Vectara: Introducing the next generation of Vectara's Hallucination Leaderboard (2025-11-19)