What MIT's JARVIS Challenge Reveals About AI's Hard Limits in Tough-Tech Engineering

The Day's Thesis

▶

Signal of the Day: IBM stock fell 25% on July 15 — its worst single-day decline on record — after the company issued a second-quarter earnings warning citing client spending shifts away from software toward hardware.

The 30-Second Read:

• IBM shares crater 25% in worst-ever session after Q2 earnings warning tied to software-to-hardware budget rotation
• SK Hynix shares surge 11% in Seoul as Asian semiconductor equities rebound following earlier U.S. selloff
• June CPI prints at 3.5% annually, 30 basis points below the 3.8% consensus, driven by easing energy prices
• Brazil proposes breaking INB's state uranium monopoly, targeting 800 metric tons/year of yellowcake output at Caetite

IBM's collapse and SK Hynix's simultaneous surge on the same day define a capital rotation story inside the hardware cycle: enterprise clients are defunding legacy software platforms to fund AI infrastructure, and the companies supplying the chips are the direct beneficiaries. The day's events across minerals, AI deployment, and capital markets all register as downstream effects of that same reallocation pressure.

---

AI & Research Frontier

MIT's JARVIS Challenge puts a quantitative stress-test on AI engineering copilots — and the results expose where current models fail under real-world constraints.

Professor Devavrat Shah's MIT research targets decision-making under limited compute — a direct counterpoint to the scale-maximalist assumptions embedded in today's frontier model architectures. Separately, MIT students used AI copilots to design, build, and test a functioning jet engine under the JARVIS Challenge framework, producing the first structured benchmark of AI utility in high-stakes aerospace engineering.

The challenge's value is less the output hardware and more the documented failure taxonomy: where AI copilots broke down under thermodynamic constraints, tolerance stacking, and materials tradeoffs that resist probabilistic approximation.