Singapore just booted a data center that runs on human neurons
A 20-unit rack of lab-grown brain cells at NUS draws under 1kW. The catch: the neurons die in six months.
Chips, infrastructure and compute. 30 stories, page 2 of 3.
A 20-unit rack of lab-grown brain cells at NUS draws under 1kW. The catch: the neurons die in six months.
AMD is buying Taalas, whose chips hardwire model weights into transistors — its answer to Nvidia's $20B Groq grab.
Starmind's AI1 satellite will run NVIDIA's NVL72 rack in space, powered by solar, selling compute back down to Earth.
Anthropic now has an in-house chip team co-designing hardware and models, with Samsung in talks to actually build it.
SpaceX's Starmind satellites run NVIDIA Rubin GPUs at 250kW peak each — first two prototypes fly in early 2027.
A Shanghai firm started building immersion DUV lithography tools — 5 units this year, ~20 in 2027. Chip equipment stocks tanked.
24GB GPU or Mac gets you a full local AI stack — fast inference, tab completion, speech-to-text, and no censorship.
NVIDIA launched Ising — open AI models that cut quantum calibration from days to hours and decode errors 2.5x faster.
Meta and Broadcom extend their AI chip partnership to co-develop 2nm MTIA accelerators at multi-gigawatt scale.
Meta extends its custom silicon push with Broadcom, targeting multi-gigawatt 2nm AI chip deployment across its apps.
Gemma 4's edge models pack vision, audio, and text AI into 1–3 GB RAM — meaning most phones from 2023+ qualify.
Anthropic secured multiple gigawatts of Google TPU capacity via Broadcom, as its annualized revenue surpasses $30B.