The Dyson Swarm Just Got Its Building Permit

As we covered on January 30th, SpaceX filed with the FCC requesting permission to launch one million satellites as orbital compute infrastructure, what physicist Alex Wissner-Gross described as "filing paperwork for a Dyson Swarm." That filing has now been accepted.

The FCC has greenlit SpaceX's application for one million orbital data centers, the first formal step toward what physicists call a Kardashev Type II civilization. The Kardashev scale measures how advanced a civilization is based on how much energy it can harness. Type I means you can use all the energy hitting your planet. Type II means you can capture all the energy output of your star. For context, humanity hasn't even reached Type I yet. Musk confirmed the scope, saying "anything less than K2 is feeble," and outlined plans to eventually disassemble the Moon to manufacture the satellites.

Not everyone is on board. AWS CEO Matt Garman pushed back at the Cisco AI Summit, saying existing technology is "pretty far" from making data centers in space possible. "I don't know if you've seen a rack of servers recently," he said. "They're heavy. The last I checked humanity has yet to ever build a permanent structure in space, on the moon or anywhere like that." He acknowledged compelling reasons, including infinite power and easy cooling, but called the timeline unrealistic. Amazon founder Jeff Bezos has estimated space data centers could be built within 10 to 20 years.

The space race is heating up on the security front too. European security officials report that Russian spy satellites have intercepted communications by making close approaches to key EU satellites, lingering for weeks in what amounts to a new form of orbital stalking. Meanwhile, China still plans to land astronauts on the Moon by 2030 to establish its own base.

Top physicists at the Institute for Advanced Study held emergency meetings this week after reaching an uncomfortable consensus: AI can now do roughly 90% of their work and will soon push discovery beyond what humans can achieve alone. The Institute for Advanced Study is one of the most elite research centers in the world, the place where Einstein spent his final decades. When those researchers say the machines are catching up, it carries weight.

OpenAI's Chief Research Officer confirmed that the company's goal is now recursive self-improvement, building AI that can improve itself, which then improves itself again, creating a compounding loop of intelligence. The target is an automated scientist that can run its own experiments, form hypotheses, and push the boundaries of knowledge without waiting for human researchers.

Alphabet reported fourth quarter results that beat projections across the board. Revenue grew 18%, driven by 17% growth in search ads and 48% growth in Google Cloud, well above last quarter's 34%. Gemini has now reached 750 million monthly active users.

The real signal was the capex guidance. Alphabet plans to roughly double its capital expenditures this year to between $175 billion and $185 billion. As we covered yesterday, investors have been watching AI infrastructure spending closely after Microsoft lost $400 billion in market value last week partly over the same concerns. Alphabet's bet is that the AI growth justifies the spend.

Alphabet's losses on Other Bets, which include Waymo, more than tripled to $3.6 billion last quarter. Its net loss on company-level activities, which Google said "primarily reflect expenses related to our shared AI research and development," more than doubled to $5.9 billion.

Nvidia is nearing a deal to invest $20 billion in OpenAI. Amazon is also in talks with OpenAI to develop customized models for its products, including Alexa. The deal would give Amazon dedicated OpenAI researchers to fix its voice assistant, which has struggled to keep up with the current generation of AI. Amazon is simultaneously considering an equity investment of tens of billions in OpenAI, making it a customer, investor, and vendor all at once, one of the many circular arrangements defining the AI industry right now.

Elsewhere, Cerebras raised another $1 billion at a $23 billion valuation. Cerebras makes AI chips that compete with Nvidia by using a single wafer-scale chip rather than racks of individual GPUs. Voice AI startup ElevenLabs raised $500 million. And Y Combinator, the most influential startup accelerator in tech, will now fund startups in stablecoins, crypto tokens pegged to traditional currencies designed to avoid volatility.

Compute is becoming the only metric that matters. Epoch AI found that compute costs at top labs now exceed salaries and marketing combined. Internally at OpenAI, researchers are struggling to get compute credits for non-LLM projects, creating friction between the teams building the core models and everyone else.

METR, a nonprofit that evaluates AI capabilities, found that GPT-5.2 with "high" reasoning has a record-breaking autonomy time horizon of 6.6 hours on complex software tasks. That means the model can work independently on difficult engineering problems for nearly seven hours before needing human intervention.

Sam Altman admitted OpenAI has "basically built AGI" and warned that models are about to become "extremely powerful" and fast. A new state-of-the-art submission to ARC-AGI, a benchmark designed to test general intelligence rather than narrow task performance, achieved 94.5% accuracy by ensembling GPT, Gemini, and Claude together.

In China, a new open-weight model called Intern-S1-Pro with one trillion parameters claimed state-of-the-art on scientific reasoning benchmarks. As we covered Tuesday, Moonshot AI's Kimi K2.5 also set a new open-weight record.

As we covered yesterday, Anthropic's Cowork plugins triggered a massive software selloff. The damage has now widened. S&P indices tracking software and financial data have lost roughly $300 billion in value since the plugins launched covering legal, finance, sales, and other knowledge work.

Anthropic also pledged to keep Claude ad-free, a signal about how they see the business model evolving, around subscriptions and enterprise contracts rather than advertising.

On the competitive front, OpenAI's market share on mobile has dropped to 45% as Gemini surpassed 750 million monthly active users. Amazon MGM is using an AI Studio to speed up film production, another example of legacy media automating its pipeline.

Researchers noted that "vibe coding," the practice of using AI to quickly generate code without deeply understanding it, is killing open source engagement. Fewer developers are contributing to open source projects when they can just prompt their way to a solution.

A service called RentAHuman launched to let AI agents hire humans for tasks that require physical presence. One human has already been paid $100 by an AI to hold a sign reading "AN AI PAID ME TO HOLD THIS SIGN" with the subtitle "Pride not included."

Meanwhile, Moltbook, the social network for AI agents we covered on Tuesday, saw usage skyrocket from 30,000 to 1.5 million agents in three days. The agents aren't just messaging each other anymore. They're building their own economy.

Bedrock Robotics raised $270 million to automate multi-ton excavators for constructing data centers. The machines doing the thinking now need machines doing the digging.

Uber is expanding its robotaxi service to Hong Kong and Madrid. As we covered on January 30th, Waymo is now doing over 250,000 trips per week and targeting one million weekly trips by year-end.

Elon Musk declared that Optimus will be the first von Neumann machine. A von Neumann machine is a self-replicating system, a machine that can build copies of itself using raw materials. In theory, you could land one on Mars and it would build an entire factory, then a city, without shipping anything else. Musk is positioning Optimus as the robot capable of building civilization on any planet.

New research shows human lifespan heritability is above 50%, significantly higher than previously estimated. That means more than half of the variation in how long people live comes down to genetics rather than lifestyle. The implication is that the specific mechanisms of aging are discoverable, and once you find the right genetic levers, you can pull them. This shifts aging from an inevitability to an engineering problem.

Congressman Eric Burlison says he has been given the locations of UAP materials, citing what he describes as credible accounts of craft and bodies. The disclosure timeline continues to advance.