Ex-Google engineer who trained first Gemini Nano sees on-device ML inflection in 2027/28 chips
_arohan_ · x · 2026-10-05
Arohan Dey, who was among the team that trained the first Gemini Nano model shipped on Android, says he's been following on-device ML for a while: it's already interesting, but things get really compelling with the late-2027/2028 generation of mobile chips, when on-device compute should enable a genuine step change.
More from Infra
- Uber details its MCP Gateway: 800+ MCP servers, 5,000+ tools, auto-generated via AutoCrawler — Roger_M_Taylor · 2026-10-05
- Stanford Professor: Sherlock Cluster Has Been Out of Rack Space for Months as Academic Compute Runs Dry — anshulkundaje · 2026-10-05
- Alibaba T-Head unveils Zhenwu V900 with 216GB memory and 1,200GB/s bandwidth, shipping Q1 2027 — teortaxesTex · 2026-10-05
- REFRAG beats prompt caching's exact-prefix limit, but 16x compression still won't fit your database in context — CShorten30 · 2026-10-05
- OpenAI reportedly selling Cerebras-powered Ultrafast inference at ~$200M per megawatt — downingARK · 2026-10-05
- Cross-cloud app-db setup sees 13x latency hit, new networking benchmarks show — DanielLockyer · 2026-10-05