OpenAI Digs A Moat, Ethereum Foundation Loses Talent, And Polymarket’s UMA Problem | The Breakdown

OpenAI launched a program offering customers guaranteed compute capacity in exchange for one to three year commitments, with discounts scaling by commitment length and capacity drawable across the full OpenAI product portfolio. Greg Brockman described it as providing discounted tokens and certainty on availability. Sam Altman has said customers are increasingly demanding capacity certainty and expects compute to remain constrained for some time. The underlying strategic logic is that enterprise customers have been switching between OpenAI, Anthropic, and Gemini within roughly a week based on marginal model improvements, making durable switching costs nearly impossible to build without locking customers in contractually. A significant caveat is that committing to one to three years of capacity is difficult to justify if AI models are expected to become substantially more powerful and efficient within that same window.

Ed Zitron has argued OpenAI will likely follow Anthropic's practice of counting upfront commitment revenue as annualized revenue and could soon report 45 billion dollars in annualized revenue. Marc Benioff stated Salesforce is spending 300 million dollars on Anthropic tokens in the current year but argued that most tokens used for coding tasks do not require top-tier models, and predicted a new intermediary layer will emerge routing each task to the most affordable capable model. That prediction, if correct, would undercut the moat OpenAI is trying to build through long-term commitments.

Eight Ethereum Foundation researchers have departed in 2025, including Carl Beat and Alex Stokes, each with seven years at the foundation, Julian Ma with around four years, Trent Van Epp who organized the protocol roadmap, Tim Beiko, Barnabay Mono, Josh the operations lead, and Tomasz Stanczak, the former co-executive director and ex-Nethermind CEO who had been at the foundation less than a year. With roughly 300 total employees, eight to ten departures is relatively modest turnover, though many of those leaving had spent multiple cycles earning below market rates compared to peers elsewhere in crypto.

The departures coincide with Vitalik repositioning the foundation away from top-down roadmap ownership toward a research and grants hub. A 38-page mandate published in March 2026 stated the foundation is not the parent, owner, or ruler of Ethereum and set the ultimate goal as passing the walkaway test, meaning the protocol would continue functioning even if the foundation and core developers disappeared entirely. Network revenue data complicates any narrative that Ethereum is in terminal decline. Ethereum held roughly two thirds of all L1 network revenue as of mid-2024, fell to 11% by January 2025, and recovered to 16% by mid-May 2025, meaning its share was actually higher in May than in January. At that same mid-May point, Hyper Liquid held 28% of L1 network revenue and Tron held 24%. The ETH to BTC ratio declined approximately 19% year to date in 2025, but the SOL to ETH ratio dropped about 40% over the past year and about 5% year to date, meaning Ethereum outperformed Solana over both timeframes. The broader signal is that crypto network revenue is now far more diversified than three years ago, making single-chain supremacy narratives less analytically useful.

A Wall Street Journal investigation published around May 17 detailed structural conflicts of interest in Polymarket's UMA oracle dispute resolution system. Polymarket does not decide many disputed outcomes itself, instead delegating resolution to UMA token holders who vote. Over the past year at least 60% of active UMA voters could be linked to Polymarket accounts, and in more than 300 disputes at least one voter had money riding on the outcome they were voting on. More than 50% of UMA votes in most disputes came from the 10 largest wallets. A committee member known as Stout of Uma Rocks admitted he regularly bet on disputed markets while voting on those same disputes, though he argued conflicted voters may research outcomes more carefully because they have financial skin in the game. A Risk Labs spokesperson said there is no credible evidence of manipulation and characterized complaints as coming from traders who lost money. Polymarket's founder acknowledged in March that the dispute process is messy and indicated improvements are coming. More than 1150 Polymarket bets have triggered disputes so far in 2025, already exceeding all of last year, though a company spokesperson noted that only 0.2% of all Polymarket contracts ever reach a UMA vote.