The AI Credibility Gap: Who Decides What Counts as Progress
Something remarkable happened in the hours between Nvidia's earnings release on 20 May 2026 and OpenAI's announcement of a mathematical breakthrough. The market treated both pieces of news as evidence of the same phenomenon: the unstoppable ascent of artificial intelligence. Nvidia's shares climbed. OpenAI's profile grew. Commentators reached for their superlatives. And somewhere in the gap between a quarterly balance sheet and a claimed proof, a more unsettling question went unasked.
That question is not whether AI is advancing — it plainly is, at pace that would have seemed implausible a decade ago. The question is who gets to define what that advancement means, and on what basis anyone outside a small circle of insiders should believe them.
A Proof Without Peers
OpenAI announced on 20 May 2026 that an internal model had solved a famous mathematical conjecture that had resisted human mathematicians for nearly eighty years. The announcement arrived via the company's own communications channels, without independent corroboration at the time of publication. No peer-reviewed journal has verified the claim. No external mathematician has replicated the result. OpenAI has not published the full proof, citing competitive sensitivity.
This is not an accusation of bad faith. It is a description of a structure. When a company announces a breakthrough that it alone is positioned to verify, the announcement functions simultaneously as scientific claim and as brand signal. The incentives are not obviously aligned with the incentives of scientific inquiry. A proof that cannot be scrutinised is a proof that cannot be wrong — which is precisely the condition that makes it commercially useful.
The historical record here is instructive without being determinative. IBM's Deep Blue victory over Garry Kasparov in 1997 was real and verifiable. But the pharmaceutical industry's string of self-reported trial successes — many of which did not survive contact with regulatory review — offers a less flattering parallel. The question is not that companies lie. The question is that the verification architecture for frontier AI claims is underdeveloped relative to the capital flows those claims now attract.
The Buyback Economy
Nvidia's first-quarter earnings, reported after market close on 20 May 2026, told a more legible story — and one that is worth reading carefully. Revenue grew 85 percent year-on-year. The company announced an $80 billion increase to its stock buyback programme, bringing the total authorised repurchase to a figure that exceeds the annual GDP of several sovereign states. The headline numbers beat analyst expectations.
The reaction was treated as a vote of confidence in the AI sector. It was, more precisely, a vote of confidence in Nvidia's pricing power. The company sells the picks and shovels of the AI gold rush; the gold itself remains largely theoretical. That the theoretical gold is worth trillions in market capitalisation tells us something about where capital is currently stationed, not where value has actually been created.
Stock buybacks are not inherently problematic. They return capital to shareholders and can signal management confidence. But $80 billion in buybacks — authorised at a moment when the AI sector is pleading for more compute, more energy infrastructure, more data centres — is a statement about priorities. The company that sits at the centre of the AI supply chain is choosing to repurchase its own shares rather than expand capacity further. Either the capacity problem is overstated, or the company expects demand to moderate before new factories come online. Neither reading is reassuring in the way the market chose to read it.
The Concentration Problem
What connects these two stories is not the technology. It is the governance gap. OpenAI's proof, Nvidia's earnings, and the cascade of commentary that treated both as unalloyed good news share a common feature: they were evaluated by a press and investor class with limited ability to verify the underlying claims. OpenAI's mathematicians cannot be cross-examined. Nvidia's guidance is a projection, not a guarantee. The financial media's job, in theory, is to provide exactly that cross-examination. In practice, the commercial relationships and access incentives that bind large outlets to large technology companies make that scrutiny imperfect at best.
This is not a conspiracy. It is a structure. A small number of private companies — Nvidia, OpenAI, and a handful of peers — now exercise decisive influence over the direction of a technology that governments, militaries, and civil societies are being told to treat as transformative. Those companies have strong incentives to frame their progress in the most favourable possible light. They have strong incentives to resist external audit. And they have accumulated enough market power and political capital that questioning their framings is increasingly costly for the institutions that depend on access to them.
What Remains Unsaid
The sources Monexus reviewed for this article consist of the companies' own announcements and market data. We are not claiming the OpenAI proof is fraudulent. We are not claiming Nvidia's earnings are misleading. We are observing that the verification gap between self-reported claims and independently confirmed facts is wider in AI than in almost any other sector of comparable economic weight. That gap serves the incumbents. It should concern everyone else.
The mathematician who spent forty years pursuing that conjecture deserves to know whether the proof is real. The pension fund manager allocating capital to AI infrastructure deserves to know whether the returns are genuine or a function of financial engineering. The regulator tasked with ensuring that concentrated AI power does not produce concentrated social harm deserves to know what that power actually consists of. None of them are currently well-served by the disclosure norms that prevail in the sector.
The AI revolution is real. The question is whether it will be governed by the companies that are currently winning it, or by the institutions that exist to represent everyone else.
Wire provenance
This editorial synthesis draws on the following public wire/social posts:
- https://x.com/polymarket/status/1923472817734320578
- https://x.com/polymarket/status/1923464519879848453
- https://x.com/polymarket/status/1923391824200877108
- https://x.com/polymarket/status/1923276446098796918