News on GPUs, specialized silicon, data center scaling, and the infrastructure powering the AI revolution.
Nvidia just dropped the Rubin CPX, a GPU laser-focused on inference's prefill phase. It's compute-heavy, memory-light — and it widens the moat around their AI empire.
Your next AI-powered app might run cheaper and faster thanks to Amazon's massive Trainium bet with Anthropic. But is this resurgence real, or just datacenter hype?
Uber just handed Amazon a win in the cloud arms race—expanding its AWS deal to run ride-sharing brains on custom ARM chips and a fresh Nvidia rival. But this isn't just about tech; it's a brutal reminder of how tangled Silicon Valley loyalties really are.
Turbines roar across the Mississippi border, fueling xAI's Colossus 2—the world's first gigawatt AI datacenter. In six months flat, they've built what others dream of in years.
AI folks banked on stacking more HBM layers to feed ravenous models. Nope—custom base dies, shoreline squeezes, and vendor drama are flipping the script, for better or worse.
Your next AI breakthrough? It's stuck waiting on reliable hardware. Fresh H100 vs GB200 NVL72 training benchmarks show Hopper's edge in power efficiency and uptime, delaying Blackwell's hype.
Ever wonder if Europe's AI future is just NVIDIA's next sales frontier? Jensen Huang laid it out in Paris — but after 20 years watching this game, I'm asking: where's the profit beyond the chips?
Everyone figured CES would be Blackwell tweaks. Instead, NVIDIA drops Rubin in full production — a six-chip beast that guts AI inference costs. Buckle up; the hardware race just accelerated.
Stressed developers, rejoice—DeepSeek's powerhouse API is back after three weeks offline. But with servers still strained, this signals a seismic shift in AI costs for everyday builders.
Britain just flipped the switch on Isambard-AI, its beefiest AI supercomputer yet. In under two years, it's ready to tackle drug discovery and climate modeling—on UK soil.
A gamer's GPU is appraising Ming vases better than some experts. Twenty years in tech, and this still surprises me.
AI startups scraping by on cheap compute? Tough luck. H100 rentals are climbing fast, turning yesterday's hardware into premium real estate amid exploding demand for reasoning workloads.