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

Eight Ethereum Foundation researchers have quit so far in 2025, exceeding the initially reported six. Departures include Tim Biko, who ran All Core Devs calls as the public-facing protocol coordinator, Carl Beat after seven years on consensus layer research, Julian Ma after four years on crypto economics and mechanism design, Alex Stokes after seven years as an operations lead, Barnaby Mono from Proposer Builder Separation research, Trent Van Epp who organized the Protocol Guild, and Tomas Stanzak, a former co-executive director and ex-Nethermind CEO who lasted less than a year. David Kanellis noted the foundation has around 300 employees, making eight to ten departures a relatively small turnover percentage, though below-market compensation has been a recurring grievance for staff who in some cases witnessed three bull and three bear cycles.

The Ethereum Foundation published a 38-page mandate document in March describing itself as part constitution, part manifesto, and part guide. The document states the foundation is not the parent, owner, or ruler of Ethereum and introduces a walkway test, meaning the protocol and core application layers should function and evolve even if the foundation and today's core developers disappeared tomorrow. Vitalik repositioned the foundation away from top-down roadmap ownership and toward a focused research and grants hub.

Ethereum's competitive position is more nuanced than decline narratives suggest. Ethereum held 11% of L1 network revenue in January 2025, rising to 16% by mid-May 2025, while Hyperliquid held 28% and Tron held 24% at that same point. The ETH-BTC ratio has declined approximately 19% year to date, but the SOL-ETH ratio has dropped about 40% over the past year and about 5% year to date, meaning Ethereum has outperformed Solana over both periods.

The Wall Street Journal reported that Polymarket's UMA Oracle dispute process has a serious conflict-of-interest problem. When a Polymarket outcome is disputed, UMA token holders vote on the result and are supposed to be financially disinterested. The Journal found that 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 UMA voter had money riding on the outcome they were voting on. In most disputes more than 50% of votes came from the 10 largest wallets. A UMA Rocks committee member known as Scout admitted he often bet on disputed markets while voting on those same disputes, arguing that conflicted traders may research outcomes more seriously than uninvolved voters. More than 1,150 Polymarket bets have triggered disputes so far in 2025, already exceeding all disputes from the prior year. Polymarket's founder Shane Copland acknowledged in March that the dispute process is messy and teased improvements, while a Risk Labs spokesperson said there is no credible evidence of manipulation and attributed complaints to traders who lost money. Polymarket has also overridden UMA Oracle dispute calls in some instances. Separately, Minnesota became the first US state to outright ban prediction markets, and the CFTC is suing Minnesota, claiming only it has jurisdiction over prediction market regulation.

OpenAI launched a guaranteed compute capacity offering for customers willing to commit to one-to-three-year terms, with discounts scaling by commitment length. Sam Altman stated customers are increasingly asking for certainty on compute capacity and expects the world will remain capacity constrained for some time. The offering is partly a strategy to reduce churn, as users frequently switch between Anthropic, Gemini, and OpenAI when new models release. Ed Zitron is skeptical and believes OpenAI is counting upfront payments as annualized revenue, similar to what Anthropic does. The core tension is whether a genuine moat exists in AI given that a marginally better model can cause a mass customer exodus within roughly a week, and locking into one platform for years creates real price and quality volatility risk for buyers.

Mark Benioff stated Salesforce is spending approximately 300 million dollars on Anthropic tokens annually and argued that the vast majority of tokens used for coding do not need to go to high-end models and could be routed to smaller ones. Benioff predicted a future company will emerge to sit between enterprises and AI providers to route tokens efficiently and reduce costs, and that compute costs will decrease over time, making current token pricing a temporary condition.