What the people running the AI economy are saying, and what it means for their team’s season.
“what they're really against is this feeling that, man, are the people building AI actually really, truly on Team Human?”
Truly on Team Human?
OpenAI CEO Sam Altman on Semafor, on data centers. Truly on Team Human? Real game, or sideline noise?
Source: @semafor on X
“Three booster catches and a ship”
Musk tallies the tower's catches: three boosters and now a ship
SpaceX's chief summed up Flight 15 in six words: three booster catches and a ship. The post he shared lays out the run, from the first booster catch in October 2024 to the first Starship caught by the tower on 14 Nov 2026. Getting both halves of the rocket back into the tower's arms is the play SpaceX has chased for two years, and now it is on the board.
Source: @elonmusk on X
“It searched both and came back with the blockers.”
Feldman clocks a Cerebras agent finding launch blockers in 36 seconds
Cerebras's captain put a stopwatch on an agent. Feldman says his team pointed an agent running Meta's Muse Spark model on Cerebras at its Slack threads and Linear tickets to find what was blocking a launch. It came back with the blockers in 36 seconds, running at up to 3,000 tokens per second. Speed is the whole Cerebras pitch, and this is an everyday office task, not a lab stunt.
Source: @andrewdfeldman on X
“Future Starship flights will carry over 50 Tb/s of Starlink bandwidth to orbit”
Musk puts a 50 terabit number on future Starship flights
During Starship Flight 15, SpaceX's chief said future Starship flights will carry more than 50 terabits per second of Starlink capacity to orbit. That is nearly double the up to 26 terabits Starlink says this flight's payload adds. Bigger rockets mean bigger satellite drops, and SpaceX wants to set the pace in the bandwidth game.
Source: @elonmusk on X
“What we shipped this week”
Pichai runs the whole week's playbook back in one post
Google's captain listed a full week of releases day by day: two new eighth-generation TPUs, Gemini 4 Carbon and Gemini 4 Hydrogen, Gemini Robotics 2.5, AlphaGenome 2 and pre-orders for smart eyewear. That is chips, models, robots and hardware in five days. Few teams in the league put that many plays on the board in a single week.
Source: @sundarpichai on X
“+58%, on a chip that was supposed to be obsolete by now.”
Dell says the old H100 rents for more than last year
Dell's boss points out that NVIDIA's H100, shipping since 2022, now rents for $3.10 an hour, up 58% from $1.96 a year ago. His read is that a chip many expected to fade is still in demand. That pushes back on the idea that AI hardware goes stale fast, though one rental price is not the whole market.
Source: @MichaelDell on X
Yahoo Finance names Lisa Su its CEO of the Year
Yahoo Finance named AMD's chair and CEO its CEO of the Year at a Yahoo Finance event. Su said she got into tech because she believes it does real good for the economy and the world. For AMD's season, it is a captain's armband moment as the chip race gets tighter.
Source: @YahooFinance on X
Intel's Lip-Bu Tan calls for a healthy timeout in the AI boom
Tan told Yahoo Finance the AI game is here for the long season, the same way the internet was. His warning is that some teams are sprinting too fast, and a pullback would be healthy. Coming from the coach still rebuilding Intel's roster, it reads as a call for patience, not panic.
Source: @YahooFinance on X
“Incredible speed and intelligence for everyday coding, agentic work and all-round use.”
Hassabis sends Gemini 4 Hydrogen onto the field
Google DeepMind's chief rolled out Gemini 4 Hydrogen, a fast model built for everyday coding and agent work. Google pitches it as finishing tasks 2.5 times faster than Gemini 3.8 Flash. This is Google's speed play in the model league, and a direct test for every rival's quick lineup.
Source: @demishassabis on X
“Instead of generating text token by token, it’s built for closed-set choices”
Pichai launches a Gemini model that picks answers instead of writing them
Google just drew up a new play. Pichai launched the Gemini Decisions API, a model built to choose from a fixed set of answers, scoring up to 1,000 questions in one pass. Input runs ten cents per million tokens and output is free. Cheap judgment calls at scale is how Google pulls more developers onto the Gemini bench.
Source: @sundarpichai on X
“Enjoyed the conversation with @drfeifei about the exciting advances in robotics, physical AI, and the opportunities ahead.”
Lisa Su takes AMD's robotics pitch to the robot learning crowd
AMD is scouting the next league. Su spent the day at CoRL 2026 in Austin, the Conference on Robot Learning, in a conversation with Fei-Fei Li on robotics and physical AI. Showing up for the students and researchers who build tomorrow's robots is how a chipmaker earns a spot on their roster before the big contracts get signed.
Source: @LisaSu on X
“Tesla self-driving now approved in Finland”
Musk says Tesla self-driving is now approved in Finland
Tesla picks up another road game in Europe. Musk flagged that Tesla self-driving is now approved in Finland, and Tesla Europe says the rollout begins soon. Every new country opens more ground for the software side of Tesla's season, and Europe has been the slow, rule-heavy stretch of its schedule.
Source: @elonmusk on X
“182 world records.”
Michael Dell posts 182 world records for servers running AMD chips
Dell is stacking the stat sheet. Michael Dell says Dell servers now have 182 world records, 95 from the new PowerEdge R9825 alone, running AMD's 6th Gen EPYC. His numbers: 94% better performance, 80% less cooling power, and 20% more compute at the same rack power. Power per rack is the tightest limit in AI data centers, so that last stat is the one to watch.
Source: @MichaelDell on X
Each clip plays where it was posted, through X’s or YouTube’s own player, with our take beside it. Clips are picked by our desk and credited to the people who said them.