Gurley: Anthropic midwifing a deity, not writing software
Bill Gurley argues Anthropic's leadership believes they are creating a superior species/deity (Dr. Frankenstein theory), evidenced by their Constitutional AI framework and Dario A…
Gurley: Cursor building own model, forced to reckon with frontier labs moving up stack
Bill Gurley notes application-layer companies like Cursor are being forced to develop proprietary models as frontier labs (Anthropic, OpenAI) move up the stack into applications,…
Chamath: Abacus sold out of on-prem AI appliances for regulated enterprises
Chamath reveals their incubator company Abacus builds integrated hardware/software stacks letting regulated industries (insurance, healthcare) run AI on-prem with data sovereignty…
Chamath: Apple is the dark horse in AI via intelligence sovereignty
Chamath argues Apple's combination of local hardware (M5 chips, Mac Studio with up to 1TB unified memory) and historical privacy stance positions it uniquely for 'intelligence sov…
Chamath: Rogo evals show frontier model convergence/commoditization
Rogo's financial analyst benchmark shows top models (Opus 4, GPT-5, Sonnet 4) separated by <0.3%, suggesting rapid commoditization. This raises ROI questions on massive training s…
Chamath: Elon rewrote training stack in C for 10x speedup
Elon's team rewrote the entire training complex in C, achieving order-of-magnitude speedups on 220k GPUs. This suggests training cost curves are collapsing, enabling $10M runs vs…
Andy Jassy's directive to prove AI-first before hiring, plus elimination of 600,000 planned positions, shows large enterprises operationalizing AI for structural cost reduction, n…
Sax/JCal: Kirkland Ellis spending $500M to build proprietary frontier model
Top law firm investing half a billion in custom frontier model signals enterprise trend of regulated industries building sovereign AI rather than relying on third-party APIs.