An Nvidia H100 has trained a small model in orbit. Between that demonstration and a space data center lie unresolved problems in heat, communications, reliability and economics.
Jensen Huang opened CES 2026 with Rubin, NVIDIA's six-chip platform promising to generate tokens at one-tenth the cost, alongside a family of open models reaching all the way to level 4 autonomous driving.
Nvidia is publishing weights, a model card, data, and tools around a mixture-of-experts model. “Open” is not binary: license, provenance, recipes, evaluations, and hardware dependence must be audited separately.
The new UltraServers arrive alongside services for model customization and interface automation. Choosing an accelerator requires measuring cost per correct task, software portability, utilization, energy, and an exit path.
The record quarter confirms infrastructure sales and heavy data-center concentration. Assessing the cycle requires separating realized revenue, guidance, margins, cash, customers, and the economic returns earned by chip buyers.
The commitments among Anthropic, Microsoft, and Nvidia combine compute purchases, capacity, investment, engineering, and distribution. Separating each flow prevents an announced ceiling from becoming cash paid or value proven.
Anthropic announced future access to as many as one million Google accelerators, but chip count does not reveal performance, use, cost, or energy consumed. The useful signal emerges by separating reservation, deployment, and completed work.
The collaboration sets a scale, divides responsibilities, and gives a schedule, but publishes no cost, chip count, or energy consumption. Separating agreements, term sheet, deployment, and operation makes the ambition testable.
Nvidia will fund AI infrastructure in phases, equivalent to millions of chips. The deal strengthens OpenAI’s capacity while fueling debate over the mutual dependence between the chipmaker and one of its biggest customers.
Nvidia has reached an intraday market capitalization of $4 trillion, an unprecedented milestone for a publicly traded company. The market is rewarding its position as the central supplier of the chips powering the expansion of artificial intelligence.
AMD has begun shipping its Instinct MI350 accelerators, with up to 288 GB of memory per GPU, and previewed Helios, a rack-scale system for MI400 chips planned for 2026. The company aims to establish itself as an open alternative to Nvidia in AI data centers.
Humain and nvidia announced a first phase with 18,000 gb300 gpus and projected capacity of up to 500 megawatts. The source bounds the event: these are deployment targets: the release does not show that equipment, grid power or useful capacity was already operational. The story teaches how to turn an infrastructure announcement into verifiable power, delivery and utilization milestones.
This website uses cookies to improve the browsing experience. Cookie policy.
No results. Try different words.
