Google's custom-designed Tensor Processing Unit chips, an accelerator family built to train and run AI models in its data centers and cloud.
Google's Project Suncatcher put a satellite carrying four TPU chips into orbit on Oct. 1 to test AI computing in space. NPR reported the chips run only 15 minutes before cooling down, and that limit decides how much computing each launched kilogram can deliver.
Reality
Arista paired its 7060EX7 switches with three Ethernet rack designs, the largest reaching 1,024 accelerators in one scale-up domain across racks. Teams on non-Nvidia chips such as Microsoft's Maia, the market Arista expects to grow, get a second scale-up fabric to evaluate.
Reality
Google's first orbital TPU satellite can run its chips for only about 15 minutes before its radiators catch up, Ars Technica reports. Power is the easy part in orbit, so radiator area decides how many chips each satellite can carry.
Reality
Google's Project Suncatcher is launching four processors that can run Gemini for only 15 minutes before they must shut down to cool. Heat, launch mass and downlink stand between that payload and the orbital data centers SpaceX pitched when it went public in June.
Reality
Google Cloud plans to double its processing capacity in Brazil between 2026 and 2030, Latin America president Eduardo Lopez said in Sao Paulo. Teams that must keep data in the country stand to get more room at home, on a timetable Google says demand will set.
Reality
Google launches its first Project Suncatcher satellite on October 1 with four TPUs on about 1kW of solar, roughly one data center server's worth of compute. It can show whether the chips stay cool in orbit, but the laser links any capacity case needs wait for a two-satellite test in 2027.
Perspective Coverage
6 publishersReality
Google's MVP satellite takes four TPUs and about 1 kW of solar power into orbit to test whether AI chips survive radiation and shed their own heat. Its results will show whether rebooting away bit flips and cooling through a radiator can carry into the constellations Google has already sketched.
Perspective Coverage
5 publishersReality
Bloomberg reports senior and junior notes of up to $100 billion to fund Anthropic and others. The supplier would now carry financing risk on the buildouts that book its revenue.
Reality
Custom silicon made up 73% of the $16.7bn AI quarter. Hock Tan says that line doubles twice by fiscal 2028. He has already secured wafers and memory through 2028 to serve promises only six buyers have made.
Perspective Coverage
9 publishersReality
Google flies one server's worth of TPUs on 1 kilowatt of solar power on Oct. 1, its first hardware test for orbital AI data centers. The flight tests chip survival and cooling while SpaceX's and Blue Origin's far larger constellations still wait on FCC review.
Perspective Coverage
5 publishersReality
Koray Kavukcuoglu said Gemini 4 is in an early phase of post-training and he wants it released much earlier than the end of the year, with fast iterations to follow. Google's last flagship attempt, Gemini 3.5 Pro, missed three deadlines and never shipped.
Reality
Chief financial officer Jean Hu put high-performance computing at $3 trillion by 2030, up from about $2 trillion in July, with the server-CPU slice alone going from roughly $60 billion to roughly $220 billion.
Reality
Nvidia holds rights to Groq's inference technology and Groq's founder, while Groq remains its own company. Whether that combination needed a premerger filing depends on rights nobody outside the deal has seen.
Reality
Dan Luu's August 30th essay argues engineers automate their own workarounds until the underlying bug stops producing reports, which matters more now that agents can aim a lot of cheap code at the wrong problem.
Reality
Reuters says Anthropic discussed roughly $7bn for MatX and stopped, and MatX is now raising near $4bn. That gap is a useful lesson in how a fabless design team gets priced once the obvious buyer walks.
Reality
SemiAnalysis's AgentX replays recorded coding-agent sessions instead of fixed prompts. On that traffic, the reported Nvidia-AMD cost gap grows as the interactivity target rises.
Reality
Treasury more than doubled its buybacks of long-dated debt on Wednesday and bought less than two sessions of calm. That is the variable to track, not the AI headlines.
Reality
Three weeks after closing on Modular, Qualcomm put Mojo under Apache 2.0 and extended support to six vendors' chips. The bet is that the arbitrage layer beats the lock-in.
Reality
The chipmaker is discounting with dilution rather than price: nearly 1.4 million shares vest in year one, the rest in tranches tied to every $500 million of chips Google orders.
Reality
A Guardian investigation says thousands of advanced accelerators sit in Microsoft inventory while data centers run late. The binding constraint on stated AI capacity is buildings, not silicon.
Reality