🚀 OpenAI overhauls its structure to chase truly colossal scale
OpenAI is reshaping itself to raise vast sums-think hundreds of billions in revenue and multi-gigawatt compute-as Sam Altman pitches an audacious, trillion-ish vision. Investors love big swings, but the how of funding and building that much infrastructure is still, well, vibes-heavy.
Analysts called it a kings-only sport, which is… a choice of words, but not wrong. The plan leans on mega partnerships and a path to IPO, with plenty of skeptics circling.
🔗 Read more
🧱 Amazon unveils Project Rainier - Anthropic to use 1m Trainium2 chips
Amazon said Anthropic will scale Claude training and deployment on Rainier, a sprawling AWS AI cluster packing nearly half a million Trainium2 chips. Yep-seven zeros’ worth of silicon.
It’s a flex for custom silicon-and a reminder that model roadmaps are essentially power stations with nice APIs.
🔗 Read more
💹 Nvidia hits $5tn valuation, because AI gravity is weirdly real
Nvidia briefly became the world’s first $5tn company, powered by insatiable demand for its AI chips and a feedback loop of compute orders and model dreams. Circular, or just efficient-depends who you ask.
The milestone piles on pressure for rivals and, honestly, raises bubble-ish questions that no one wants to hear at the party.
🔗 Read more
🛡️ OpenAI ships gpt-oss-safeguard - open-weight safety reasoning models
Two open-weight classifiers, 120b and 20b, aim to let devs plug in their own safety policies and get transparent, reasoned decisions. More policy-in, less mystery-out.
It’s part research preview, part olive branch to the open ecosystem-and slightly surprising given the model sizes involved… in a good way, mostly.
🔗 Read more
🛰️ IBM debuts a defense-focused AI model with Janes data baked in
IBM’s Defense Model promises secured, air-gapped deployments and domain-tuned planning, reporting, even wargaming. Sounds intense, because it is.
It rides IBM’s Granite lineage and the watsonx stack-fit-for-purpose, not just a general LLM with camo paint.
🔗 Read more
🏥 Zurich Insurance launches an AI Lab with ETH Zurich and St. Gallen
The insurer is framing AI as more than efficiency-a rethink of the insurance business model, which, cheeky as it sounds, might be true this time.
Research talent plus industry data plus agentic systems… feels like a recipe, or a cauldron. We’ll see which.
🔗 Read more
🧩 Qualcomm teases AI200 and AI250 - datacenter inference at rack scale
Qualcomm is stepping into full racks with near-memory designs and liquid cooling, targeting better tokens-per-dollar-per-watt. The rack is the unit-not the server.
Timelines stretch into next year and beyond, but the direction is clear: inference gets its own playbook, not just hand-me-down training rigs.
🔗 Read more