OpenAI is Not Dead

Noah Smith’s essay What if AI Succeeds, But OpenAI Fails argued that OpenAI’s infrastructure costs were enormous, competition was intensifying, first movers often lose technology races, and the company could become the Netscape of AI. Business Insider, Axios, and VentureBeat amplified the story about how Anthropic had surpassed OpenAI in business adoption in May. On the All-In Podcast last month David Sacks suggested Anthropic had “won” and might become a dominant monopoly. Articles about OpenAI missing internal targets and the Elon Musk lawsuit against Sam Altman generated a narrative of a lost company.

However, CFO Sarah Friar took the stand at the Liquidity Summit 2026 and presented a highly cogent case that not all was lost. OpenAI’s heavy compute investment, once criticized, now looks strategic given memory and data centre shortages. This will help their Codex product become more useful in the enterprise. Anthropic is currently buying compute from Elon Musk, but it’s questionable whether the $1.25 billion they pay xAI per month is a good deal.

On the consumer side, data from Arnal Dayaratna, an IDC market intelligence researcher and old high school friend, shows that OpenAI has 900 million users and generates $10-$13 billions from consumer subscriptions versus $2 billion in consumer subscriptions for Anthropic. Claude has 50 million users, but many of them are from the coding and enterprise community, making Anthropic’s revenues more concentrated. OpenAI’s investments into consumer don’t pay as lucratively at the start, but over time they may eat into Google’s search revenue. I’ve tested buying products through OpenAI and Google’s AI search and OpenAI serves up much higher quality often grounded with good insights from our conversations in the past. Google’s feels much more like a sponsor fed it to them. As more people experiment, OpenAI’s consumer revenue could mushroom.

Their partnership with Jony Ive could also yield interesting results in hardware. Apple felt it was enough of a threat that it buried its ChatGPT integration.

OpenAI also has interesting ‘GPTs’, which are verticals in areas where AI might be able to displace a legacy app or profession. One example is DocuGPT, which could displace DocuSign, a SaaS company that generates $1 billion of free cash flow per year. OpenAI’s legal and accounting modules also have the potential to divert big markets.

Defensively, OpenAI is less reliant on the API/platform layer. About 46% of revenues comes from this API layer. Anthropic is 64%. The issue with this layer is that customers have more capability to be price sensitive particularly as Chinese models catch up. DeepSeek tokens are only 10% of the cost of Anthropic. In the Oddlots podcast, Hudson Bay Financial, who sports a significant development team and utilizes tokens, noted that Claude Code gave their team a 50% productivity boost, very significant, but not the 400% that Anthropic has cited in its essays. If Composer 2.5, DeepSeek, Codex and others can design effective coding tools, then there might be more aggressive pricing in this layer. Focusing on the consumer layer may yield a much higher pay-off in the end.

Trouble With the Marks in Switzerland and Private Credit Continues to Grind Lower. Partners Group AG’s share price plummeted 17% as short-seller Grizzly Research alleged that 40% of Partners Group investments may be mismarked. Partners announced that it was gating $9.7 billion in its PE evergreen funds. These redemptions are mainly from institutional investors. In his 2026 shareholder letter, Jamie Dimon suggested that private equity is holding companies much longer than historically and increasingly relying on continuation funds.

At the Bloomberg Forum, Jay Clayton, attorney general of the Southern District of New York said his office was looking at “divergent marks of the same loan” and is “committed to rooting out any shenanigans in private assets.” Barclays’ Na Wei reported clients are saying private credit funds are bidding to lose. Another market participant said the $230 billions of cable and broadband debt face a “day of reckoning” from cord-cutting and competition. Diameter Capital predicted that BDCs would sell their highest-quality software loans at a discount to par leaving the remaining loans marked at high prices that they can’t sell for redemptions. This is likely to trigger more redemptions.

Bitcoin Having Some Challenges. Last year’s momentum trade was Bitcoin. Strategy, a listed company, would sell preferred shares and then buy Bitcoin. The premium to its Bitcoin holdings would attract more capital, and Strategy would repeat the process. Last year, Mike Saylor, Strategy’s CEO said he would never sell Bitcoin. Now he has reversed course and sold 32 BTC. While this is a tiny amount, the $1.7 billion annual dividend becomes harder to service if Bitcoin prices weaken further.

It also shows manias can come and go.

More Hedging Coming. Barclays estimates $2 trillion of mortgage-backed securities (MBS) now carry coupons of 5% or higher, roughly four times the level of three years ago. The higher coupon securities sit close to par value. In 2022, these bonds sold off to well below par due to interest rates hikes. Hedging activity with Treasuries dissipated. Now if rates were to increase, Barclays believes it could trigger mass convexity re-hedging, which could force repo-driven deleveraging and an MBS sell-off. Barclays believes MBS convexity has the potential to destabilize rates and is under-appreciated by the market.

This is a bit technical, so I don’t have a strong opinion. But systemic risk feels elevated as AI valuations, the fiscal deficit stimulus, higher inflation, private credit issues and technical dynamics in the bond market all interconnect.

The Fed Independence Test. Kevin Warsh, the new Fed Chair, is preparing the most significant overhaul of the Federal Reserve communications in decades. Former officials expect him to roll back “forward guidance” as soon as the mid-June FOMC meeting. He may not submit his own quarterly dot plot, he may remove easing/tightening language in FOMC statements, and he may curtail forward guidance more broadly.

Hawkish policymakers at the Fed support the change particularly as the Iran War has reignited inflation concerns and prompted three April dissents. They don’t want to be strapped into an easing dot plot when conditions change.

However, reducing transparency at the exact moment the Fed faces unprecedented White House pressure creates risks to Fed independence. Without the dot plot, it becomes harder for the market to detect when policy is drifting from what economic conditions warrant. The markets lose a tool to hold the Fed accountable to its own reaction function.

If Warsh pairs reduced guidance with a hold and hawkish-leaning statement, the change will be perceived as positive to continued Fed independence. Otherwise, the markets could read the opacity as covering for political accommodation.

Expect higher-rate volatility with the change. The front-end and belly of the curve will become more reactive to each data print. The term premium will likely rise. There might be a wider dispersion on SOFR futures.

China Joins the AGI Party. Former OpenAI researcher, Yao Shunyu now Tencent’s Chief AI Scientist, publicly declared his goal of building AGI in China, a significant shift from China’s traditional focus on AI applications rather than frontier capability. He framed AI as a “long-term game” suggesting that the “second half is just starting.” He compared AI to personal computers in the 1970s implying most of the value creation still lies ahead. “I don’t think there are any secrets. What we need to do is get the infrastructure and data right, and the algorithm will be the easy part.”

While Yao has identified the problem correctly: chips and data pipeline, fixing those elements won’t be easy. But I’d watch what he does. China tends to surprise markets with what they can do.

End Note

I randomly connected with a venture capitalist recently, and their firm is funding a start-up whose aim is to replace primary physicians with an agentic AI system. You would visit an AI doctor, it would ask you questions and diagnose, AI agents would then send you to a specialist doctor or prescribe treatment (not sure how exactly). And then the primary physician can be eliminated.

This seems like such a waste to me. AI assistance in diagnosis and simplification of billing creates an opportunity for doctors to meet with patients and provide better explanations, more empathy, notice depression or stress. Medicine might become more like the television shows. Instead, it’s let’s try to use AI to fire primary care physicians. Maybe someone should fire the venture capitalists.

Omar Sayed