OpenAI and Apple lawyered up. The compute tab hit $50 billion a year and climbing.

article · openai / apple · may 14

Why for you: After the ChatGPT-on-iPhone integration failed to produce the subscriber growth OpenAI expected — Apple controls placement, defaults, and the trust relationship with the user — OpenAI is reportedly considering legal action. The pattern is well-documented: Apple signs distribution deals and then controls the channel in ways partners didn't anticipate. For OpenAI, this is a strategic fight as much as a legal one: distribution via iPhone is uniquely valuable, and losing it to Gemini or a future Apple-native model would be material. The "wouldn't be the first partner" framing isn't just flavor — Epic, Spotify, and others have tried variations of this fight. OpenAI's leverage is different because it has consumer brand pull Apple actually needs right now.

article · openai / legal · may 14

Why for you: Musk's legal team argues OpenAI betrayed its nonprofit founding mission by shifting to profit; OpenAI argues Musk wanted control, not charity — and that Altman's testimony shows Musk demanded 90 percent equity and the CEO role before walking away in 2018. The statute of limitations may matter more than the merits: if Musk's claims are time-barred, the $150B damages ask goes away without touching the governance questions. But the governance questions won't go away either way — this case has surfaced more candid primary-source material about how OpenAI's founding mythology was constructed than any previous reporting, and that material stays public regardless of the verdict.

report · openai / compute · may 15

Why for you: OpenAI CFO Sarah Friar said the company may raise additional capital even after closing a record $122B round — because compute spending alone is projected to hit $50B in 2026. That's not a warning sign; it's a deliberate strategy of spending ahead of capability (buy the GPUs before you know exactly what you'll use them for), which works if revenue compounds and fails expensively if it doesn't. The CFO framing this as a "compute crunch" rather than a spending problem positions the constraint as supply-side (not enough GPUs to buy) rather than demand-side (not enough revenue to justify buying more). Investors appear to be accepting that frame.

essay · pragmatic engineer · may 12

Why for you: Gergely Orosz takes Frederick Brooks's famous 1986 claim — that no single technology would ever produce an order-of-magnitude improvement in software engineering productivity — and asks whether AI finally qualifies. Brooks's argument was that essential complexity (the hard thinking about what to build and why) can't be automated away; only accidental complexity (the mechanical translation of ideas into code) can. Orosz examines whether AI addresses the essential layer or only the accidental one, and whether the gains we're seeing are genuinely order-of-magnitude or a lot of 10–30% improvements strung together. The Brooks framing clarifies a surprising amount of the current debate about AI and engineering work.

essay · sean goedecke · may 12

Why for you: Goedecke digs into Thinking Machines' new scaled-up multimodal system with fully-duplex voice and video integration, and argues the novelty is practical, not technical — the pieces existed before, but they've been assembled at a scale that actually works in real conversation. His frame: interaction model design is about finding the right latency and coherence tradeoffs for a given use case, and Thinking Machines chose different tradeoffs than OpenAI's voice mode. Useful reading for anyone building AI-native features who needs to understand why "just use the voice API" doesn't solve the interaction design problem.

note · simon willison · may 14

Why for you: Willison flags a shift in how teams think about language and platform lock-in: a company rewrote its native mobile apps in React Native with full confidence it could revert if needed, and didn't find that confidence misplaced. He ties it to Mitchell Hashimoto's observation that AI coding tools are making cross-language rewrites feasible that would have been too expensive before. The underlying claim: language and framework monocultures were sustained partly by switching costs, and AI is quietly compressing those costs. For a Rails developer, this cuts both ways — Rails's switching-cost advantage diminishes, but so does the barrier to experimenting with something new.

article · fintech / ai · may 15

Why for you: Sixfold's insurance underwriting AI is now available through Azure Marketplace, which matters less because of Sixfold specifically and more because of the distribution pattern: enterprise AI tools deployed through cloud marketplaces can land in procurement pipelines without a separate sales cycle. Underwriting is one of the better early AI use cases — structured inputs, clear ground truth, measurable accuracy — and Marketplace distribution means global insurers can discover and deploy it within their existing Azure infrastructure. The fintech AI vertical is filling in fast; underwriting and AML are the two clearest near-term applications where the accuracy bar is actually meaningful.

essay · sean goedecke · may 13

Why for you: Goedecke pushes back on the assumption that space-based AI datacenters are a non-starter because there's no coolant in a vacuum. His argument: radiative heat transfer via large shaded radiators works, even for dense AI compute — he estimates roughly 250,000 square meters of radiator area per 100MW of datacenter capacity. The real constraints are cost-to-orbit and latency, not thermodynamics. Worth reading even if you never plan to build a space datacenter: it illustrates how "obvious" engineering intuitions (no atmosphere equals no cooling) can be wrong, and why first-principles thinking matters when evaluating genuinely novel infrastructure proposals.

essay · derek thompson · may 14

Why for you: Thompson maps the dominant forces reshaping the economy and culture this year: AI compression of white-collar work, demographic stagnation in rich countries, the end of globalization's easy phase, an attention-crisis feedback loop, political volatility from economic displacement, and what he calls "agency collapse" — a widening sense that individuals can't affect outcomes. The piece is analysis rather than prediction; Thompson is better at naming what's already happening than forecasting what comes next. The six frames are useful context for reading anything else this week, including the OpenAI legal circus above.