Today we're talking about Sam Altman. Someone I don't know personally and someone I'd probably like to meet. He's a complicated guy bearing a pressure that would drive anyone mad: he runs the most-watched company in the global tech sector at the moment of greatest expectation and greatest public exposure in recent history. Every word of his moves markets. Every decision of his has consequences for millions of people. Every mistake he makes hits the front page before he's finished his morning coffee. Even so, his public exposure is excessive, with plenty of light and plenty of shadow. Is Sam Altman a walking oxymoron? In a quick glance at his public activity we're going to see whether he's a reliable guy or not. Without malice. With data.
May 2023, the Senate and Brussels Play
On 16 May 2023, Sam Altman testifies before the US Senate Judiciary Subcommittee on the regulation of artificial intelligence. The testimony is on record. Altman says, verbatim, sentences that translate roughly like this: "I think if this technology goes wrong, it can go quite wrong. We want to work with the government to prevent that from happening. We believe that regulatory intervention by governments will be critical to mitigate the risks of increasingly powerful models."
That same week, in Brussels, OpenAI's lobbying team was pressuring the European Union to soften the strictest clauses of the AI Act, then still under negotiation. Specifically, the clauses that were going to classify large-scale general-purpose models (GPAI) as high-risk systems with hard operational obligations. Time published the documentation in June 2023: internal OpenAI papers shared with European authorities arguing that "we disagree" with GPT-4 being automatically considered high-risk and proposing amendments that would lighten the obligations.
"Asking for regulation" in public and "blocking it" in private isn't an apparent contradiction. It's a coherent strategy. General regulation, in the abstract, is good for OpenAI: it raises the entry cost for competitors, legitimizes the established players, gives the appearance of responsibility. Specific regulation, with concrete operational obligations and quantified sanctions, is bad for OpenAI: it limits deployment, raises costs, exposes liability. Asking for the first before the Senate while blocking the second in Brussels is playing the game of public responsibility-talk with the right hand while keeping the left hand on the real brakes.
It's not a Machiavellian Altman. It's a CEO executing what his board and his investors pay him to execute. The thing is that when a journalist later quotes "Altman asked for regulation," they're omitting half the actual behavior and crediting the engaged citizen without qualifying the corporate lobby. That omission is where Altman's credibility gets artificially inflated.
The AGI Timeline
A second dimension where his credibility deserves scrutiny is the timeline of when artificial general intelligence (AGI) will arrive, a term Altman himself helps keep vague so the promise can be renewed each time the previous deadline draws near.
If you reconstruct his statements over the years —and here I offer a synthesis of my own, not a literal transcription of dated sentences— the deadline has moved. In his earliest interventions he spoke of a horizon of decades. With time the horizon kept shortening. In February 2025 he published an essay titled Three Observations where he writes that, within a decade, "maybe everyone on earth will be capable of accomplishing more than the most impactful person can today," and that the 2025-to-2027 stretch will feel more impressive than the previous two years. What the essay actually argues isn't a firm date for AGI but an investment curve: the cost of using AI is falling fast and the socioeconomic value is growing super-exponentially. The concrete date stays, as almost always, deliberately blurry.
There's a pattern. When there's product success, the timeline comes closer: it feeds capital and expectation. When there's competitive turbulence or when competing models appear in unexpected places —China, Europe, open source— the timeline recedes: it lowers the expectation of an immediate competitor. The pragmatic function of Altman's statements about AGI isn't to predict the future: it's to manage the sector's narrative tension to keep capital flowing toward OpenAI. Which is perfectly legitimate as a CEO's role. But we have to stop reading it as impartial technical prophecy.
Worldcoin, the Awkward Piece
In parallel to OpenAI, Sam Altman co-founded Worldcoin (rebranded World) in 2019. It's a project that scans people's irises in countries of the global south —Kenya, Indonesia, India, Brazil, several others— in exchange for cryptocurrency tokens. The official justification: to create a verified human identity in the era of bots and deepfakes. The criticism: a private corporation is building a global biometric database using a vulnerable population lured by modest economic incentives.
The governments of Kenya, Spain, Argentina, Hong Kong and others have suspended or investigated Worldcoin's operations at various points. MIT Technology Review published an investigation in April 2022 describing deceptive recruitment practices, worker exploitation and questionable handling of biometric data in the initial rollout in countries of the global south. The company has rebutted specific aspects and modified practices. But the underlying question —should a private company build the world's largest biometric database, in countries with less regulatory protection, in exchange for crypto tokens?— still has no comfortable answer.
Here's what's interesting for our discussion of credibility. When Altman speaks publicly about AI ethics, about responsibility, about how to develop technology "for the benefit of humanity," practically nobody asks him how that squares with Worldcoin. Worldcoin's operation exists in a different information window from OpenAI's: the general press that covers AI doesn't usually cover Worldcoin with the same intensity. The result is that a CEO who claims to be committed to ethics in one project actively has another project running that many critics consider ethically problematic, and the public conversation doesn't triangulate between the two.
This isn't an ad hominem attack. It's an observation about coherence. If personal coherence is relevant to assessing the credibility of a leader's public statements, the dissociation between the OpenAI Altman and the Worldcoin Altman deserves to be in the conversation, not buried in a specific interview most readers won't read.
November 2023, the Episode That Defines Him
Altman's firing and reinstatement in November 2023 is already well-known history. OpenAI's board fired him on 17 November, alleging "lack of candor in his communications." Four days of chaos. Microsoft threatened to absorb him. More than 700 employees signed a letter. His return was agreed in the night of 21-22 November. The board changed.
What's worth looking at for our discussion of credibility is what the board never came to explain in detail. The specific reasons for the "lack of candor" stayed buried. The board members who made the decision —Helen Toner, Tasha McCauley, Ilya Sutskever (who later partly recanted)— left without publicly developing the motivation. The company hired the firm WilmerHale to do an independent investigation. The report was published in March 2024 with conclusions that cleared him on the big picture but didn't go into the details that originated the conflict.
What was left? That at some point between the summer and autumn of 2023, members of the board charged with protecting the founding mission concluded that the CEO wasn't being completely sincere with them. That they couldn't sustain their decision for more than four days against the combined pressure of the employees, the investors and Microsoft. And that after the reinstatement, the members who had made the decision left and a board more aligned with executive management came in. The original cause of the conflict was never publicly clarified.
For assessing credibility, this matters because it sets an operational precedent. OpenAI's founding governance, designed precisely to stop a CEO if he acted against the mission, couldn't perform its function. The board signed. The CEO ruled. And the asymmetry proved irreversible. When Altman today speaks publicly about how OpenAI keeps its commitment to the mission, we have to remember that the mission is protected by a structure that has already shown its weakness.
The Growing Debt and the Business Plan
Let's jump to 2026 and look at the numbers. OpenAI took in approximately 3.7 billion dollars in 2024 and some 13.1 billion in 2025, according to figures compiled by sector trackers. It's worth not confusing that recognized annual revenue with the annualized run-rate —what the company would bill if it kept the year-end pace for twelve months— which by the end of 2025 already topped 20 billion. It's spectacular growth in any case: more than triple the revenue in a year. The public narrative underlines these numbers.
What the narrative underlines less is the other side of the balance sheet. The burn rate (how much the company loses annually) was around 8 billion dollars in 2025, with sector projections pointing to some 25 billion in 2026 and some 57 billion in 2027. The gross margin is 33%, limited by the cost of inference: 8.4 billion spent on inference in 2025, projected at 14.1 billion in 2026. And the capex chapter is stratospheric: the company has projected spending of several hundred billion dollars on infrastructure through 2030.
To sustain this plan, OpenAI has had to resort to ever-larger funding rounds. The most recent, closed on 31 March 2026, was 122 billion dollars, with a post-money valuation of 852 billion. Co-led by SoftBank, with participation from Andreessen Horowitz and strategic partners like Amazon and Nvidia.
And an IPO is being prepared. Internally there's talk of a filing in the second half of 2026 and a listing in 2027, with a potential valuation of up to a trillion dollars. According to internal reports, CFO Sarah Friar reportedly suggested delaying the IPO to 2027 against Altman's push to bring it forward, citing the burn trajectory and the pending capex as reasons to wait.
The question for our credibility assessment: when Altman speaks publicly about OpenAI's future, is he describing a technical reality or executing a several-hundred-billion-dollar financial operation that needs to keep the story going until the moment of listing? Both things can be true at once. But the second dimension exists and is almost never mentioned when he's cited as an impartial technical authority.
Various financial outlets have openly discussed whether OpenAI can sustain this model or whether it's on a trajectory that ends with a hard adjustment before the listing. There's no consensus. What there is is the legitimate question of whether the most-watched company in the sector is building a genuine technical story or managing a financial house of cards on a deadline. To assess Altman, this question should always be present. It almost never is.
Sora 2, the Disaster That Also Says Something
On 24 March 2026, OpenAI announced the shutdown of the Sora app —its video-generation platform, built on the Sora 2 model— barely a few months after its launch in late 2025. The reason the company gave pointed to compute cost and an unsustainable economics. The concrete user and daily-spend metrics that have circulated in the press I haven't been able to confirm with a reliable source, so I'm leaving them out. What is documented is the pattern: a generative-video app that consumed an amount of compute hard to sustain and that OpenAI preferred to switch off rather than keep funding.
The case is interesting because it touches two things. One: the public over-promise at launch. Altman had presented Sora as a foundational advance in video generation, and the operational reality was a product the company ended up switching off as unsustainable. Two: behavior with partners. In December 2025, Disney had announced a licensing deal with OpenAI that included a committed investment of 1 billion dollars in the company. When Sora was shut down a few months later, the deal didn't go ahead; according to the parties, no money was ever disbursed because it was never finalized. A billion dollars of announced alliance that evaporates before materializing.
When a CEO presents a launch as a foundational advance and the product is canceled a few months later, dragging a just-announced billion-dollar alliance with it, there's a component worth recording for future assessments of his statements. It doesn't invalidate everything Altman says. It does calibrate. The rhetorical inflation has a recent and documented track record.
The Question of Relative Noise
Another dimension, more qualitative, is worth naming. The AI sector in 2026 has four or five frontier players. It's worth observing the relative media behavior of each.
Sam Altman publishes blog posts, gives interviews, appears at summits, travels to India and Brussels, photographs himself with presidents, proposes "a new New Deal for AI," issues grandiloquent statements about the future of civilization. He generates constant noise.
Dario Amodei, CEO of Anthropic, publishes long essays on his personal site, gives selective interviews, keeps a far lower profile than Altman although his company is competitively equivalent. He makes some noise. Much less.
Sundar Pichai, CEO of Google, appears sporadically for product announcements. Demis Hassabis, CEO of DeepMind, publishes technical papers. Google's media strategy is low personal profile and lots of product. Almost no noise.
Liang Wenfeng, founder of DeepSeek, barely appears in Western media. He gives few interviews. He publishes a devastating technical paper and lets the market react. Zero noise. Deafening results.
The sector that advances in silence overtakes the sector that advances in noise. The sentence may seem naive, but the pattern is real. When a CEO emits as much public noise as Altman, it's reasonable to ask how much of the noise serves the company and how much serves the ego, and how much attention it's consuming on brand management that could be devoted to operations. That the answer isn't unequivocal doesn't mean the question doesn't deserve to be asked. It does.
Why Calibration Matters
When Altman gives an interview, the general media tend to cite him as a source. But Altman is the CEO of a corporation with concrete incentives: raise capital, sustain valuation, manage the regulation, keep the narrative going until the listing. His statements are a commercial act, not a technical report. Reading them as if they came from a neutral expert in the sector is a methodological error. No serious financial journalist cites the CEO of Tesla, Apple or Amazon as an impartial authority on his own company's products. The presumption is of interest. For some reason, that reasonable presumption is relaxed for AI CEOs.
The honest calibration of Sam Altman's credibility isn't "they lie" or "he tells the truth." It's something finer. Altman is an extraordinarily competent strategic actor operating within extreme constraints. Everything he says is optimized for several variables at once: capital, regulation, valuation, team loyalty, social perception. None of his public statements can be read cleanly without discounting those weights.
When he says he's asking for regulation, discount the left hand in Brussels. When he says when AGI arrives, discount the moment in the funding cycle. When he says a product is revolutionary, discount the Sora trajectory. When he speaks of the founding mission, discount the November 2023 episode. When he speaks of ethics, discount Worldcoin.
It's not about disbelieving him. It's about reading him well. And reading him well starts by recognizing that a CEO in his position is not and cannot be a neutral authority on anything related to his own sector. He's an actor with concrete interests, an extraordinary talent for articulating them, and the inescapable responsibility to optimize for his company before the world. That's the implicit contract anyone who holds that post signs. It's not a betrayal of humanity: it's the description of the role.
The question stays open. Is Sam Altman reliable? As CEO of OpenAI, managing capital, talent and narrative: extremely. As a guide to what's happening in AI and where it's going: only if you've discounted all the dimensions we're talking about here. The difference between the two readings is what separates the informed reader from the reader who buys the press release.
Definitions
Burn rate: the rate at which a company loses money per unit of time. In OpenAI's case it was around 8 billion dollars in 2025, with sector projections pointing to some 25 billion in 2026 and some 57 billion in 2027.
Capex (capital expenditure): spending on long-term physical assets, mainly compute infrastructure. OpenAI has projected spending of several hundred billion dollars on infrastructure between now and 2030.
Inference: the operating cost of running the model to answer a user query. At OpenAI, 8.4 billion in 2025, projected at 14.1 billion in 2026. It's the main limiter of the gross margin.
AGI (artificial general intelligence): a term with no agreed operational definition. It usually designates a system with cognitive capabilities comparable to or beyond human ones across a wide range of tasks. Its definition is adjusted over time according to narrative convenience.
Worldcoin / World: a project co-founded by Sam Altman to create a global biometric identity verified by iris scan in exchange for cryptocurrency. Active and controversial since 2019.
References
Time (June 2023). Exclusive: OpenAI Lobbied the EU to Water Down AI Regulation. The basis for the section on the Brussels lobbying.
Transcript of Sam Altman's testimony before the Senate Judiciary Subcommittee, 16 May 2023. The source of the testimony quotes.
Altman, S. (February 2025). Three Observations. https://blog.samaltman.com/three-observations. The basis for the section on the AGI timeline.
The New York Times (November 2023 to March 2024). Coverage of Altman's firing and reinstatement and of the WilmerHale report. Cross-checked chronology in Removal of Sam Altman from OpenAI, Wikipedia. https://en.wikipedia.org/wiki/Removal_of_Sam_Altman_from_OpenAI
MIT Technology Review (April 2022). Deception, exploited workers, and cash handouts: How Worldcoin recruited its first half a million testers. https://www.technologyreview.com/2022/04/06/1048981/worldcoin-cryptocurrency-biometrics-web3/. The basis for the Worldcoin section.
Sacra. OpenAI revenue, valuation & funding tracker. https://sacra.com/c/openai/. Revenue and valuation figures.
CNBC (November 2025). Sam Altman says OpenAI will top $20 billion in annualized revenue this year, hundreds of billions by 2030. https://www.cnbc.com/2025/11/06/sam-altman-says-openai-will-top-20-billion-annual-revenue-this-year.html. Annualized run-rate and capex projection.
Fortune (November 2025). Coverage of OpenAI's burn rate and loss projections. https://fortune.com/2025/11/12/openai-cash-burn-rate-annual-losses-2028-profitable-2030-financial-documents/. Burn rate, gross margin and inference figures.
Bloomberg (31 March 2026). OpenAI Valued at $852 Billion After Completing $122 Billion Round. https://www.bloomberg.com/news/articles/2026-03-31/openai-valued-at-852-billion-after-completing-122-billion-round. The funding-round data.
Gizmodo (2026). OpenAI's CFO reportedly wants to delay the IPO from 2026 to 2027. https://gizmodo.com/openais-cfo-reportedly-wants-to-delay-the-ipo-from-2026-to-2027-2000753760. Sarah Friar's internal recommendation on the IPO.
Variety (March 2026). OpenAI Shutting Down Sora Video App, Disney Investment Dead. https://variety.com/2026/digital/news/openai-shutting-down-sora-video-disney-1236698277/. The Sora shutdown and the collapse of the Disney deal.
Deadline (March 2026). Sora Shut Down, Disney Investment Dead. https://deadline.com/2026/03/sora-shut-down-disney-investment-1236764689/. Confirmation of the Sora shutdown and the Disney deal.
Fortune (6 April 2026). Sam Altman says AI superintelligence is so big that we need a 'New Deal'. Critics say OpenAI's policy ideas are a cover for 'regulatory nihilism'. https://fortune.com/2026/04/06/sam-altman-says-ai-superintelligence-is-so-big-that-we-need-a-new-deal-critics-say-openais-policy-ideas-are-a-cover-for-regulatory-nihilism/. The "New Deal" for AI proposal.
CNN Business (26 February 2026). Anthropic rejects latest Pentagon offer. https://www.cnn.com/2026/02/26/tech/anthropic-rejects-pentagon-offer. NPR (27 February 2026). OpenAI announces Pentagon deal after Trump bans Anthropic. https://www.npr.org/2026/02/27/nx-s1-5729118/trump-anthropic-pentagon-openai-ai-weapons-ban. Anthropic's rejection and OpenAI's subsequent signing with the Pentagon.
Hao, K. (2025). Empire of AI. Penguin.
Frankfurt, H. (2005). On Bullshit. Princeton University Press.
También te interesa
- Who OpenAI Really Is
- Anthropic and OpenAI, Two Ways of Understanding AI
- How Did the Public Conversation About AI Break?
No comments yet
No comments yet. Be the first.
Leave a comment