The AI Bubble Won't Pop From One Bad Earnings Call, It Would Take Six Dominoes
A true AI bubble crash would need six dominoes to fall in sequence: ROI doubt, hyperscaler capex cuts, supply chain contagion, debt defaults, market selloff, and government backstops, not just one bad NVIDIA earnings call.
What it is
A framework for thinking about what an actual AI bubble crash would require. Not a single stock dropping, but six dependent failures happening in order, from investor doubt about AI's return on investment all the way to government intervention in what the US and China now both treat as strategic infrastructure, not just an industry.
What it does
It traces six links: ROI disappointment first (companies concluding AI helps but not enough to justify the spend), then hyperscaler capex pullbacks (Microsoft, Google, Amazon, and Meta together are already committing close to $700 billion for 2026, with Meta's free cash flow collapsing to $784 million from $8.55 billion a year earlier even as it raises its own capex floor, and NVIDIA still posting $81.6 billion in quarterly revenue on the back of it), then a slowdown moving down the chip and construction supply chain, then debt stress at operators who financed data centers assuming years of high utilization, then a Wall Street selloff, and finally government backstops. Unlike the 2000 dot-com bust, this chain has a cheap-AI escape valve too: China's push toward aggressively priced, open-weight models like Alibaba's Qwen3.8-Max could make compute a commodity before the debt dominoes ever get the chance to fall, a bet Michael Burry is already making directly against infrastructure names like Oracle and Nebius while going long Alibaba on the other side.
Why it matters
For anyone building product or engineering teams right now, the second-order effect matters more than predicting the exact crash date. Whether AI infrastructure gets cheaper through a bust or through pure competition, the same shift is already reaching software teams. Anthropic's own 2026 Agentic Coding Trends Report documents developers moving from writing code to orchestrating agent teams, and the research on what that costs is uneven: METR found experienced developers ran 19% slower with AI tools while believing they were 20% faster, and both a meta-analysis of 23 studies and a controlled study on coding agents found the same pattern: agents boost task completion but measurably hurt code comprehension, worst for juniors. The scarce skill stops being who can write the code and becomes who can decide what to build and catch what the agent got wrong.
How to use it
Read in this order for the full picture in under an hour: Burry's own framing of why he thinks the market is mispricing this, then Meta's Q2 2026 capex guidance, then NVIDIA's Q1 FY2027 results, then the Anthropic and METR research linked above. That order covers the bear case, the actual spend, the chip demand it is buying, and the developer impact it is already having.
One Endpoint, Many Backends: OpenRouter's Fallback Feature Has a Catch
OpenRouter's automatic fallback routing can send identical requests to different backend providers running different serving software, producing inconsistent outputs for the same model name.
re.match() Isn't Gone, Just Officially Discouraged in Python 3.15
Python 3.15 soft-deprecates re.match(), marking it as discouraged for new code without any plan to remove it, in favor of a clearer alternative.
OpenAI Claims a Navier-Stokes Millennium Problem Resolution, Attribution Fight Follows
OpenAI says an unreleased model produced a resolution to the Navier-Stokes existence and smoothness problem, one of seven Millennium Prize Problems carrying a $1,000,000 prize since May 2000.