Anthropic Partners With SpaceX and Now Runs on Musk’s Hardware Anthropic signed a deal with SpaceX (now SpaceXAI) to lease the entire Colossus 1 supercluster, putting Musk and Anthropic on the same team for the first time. The deal lets Anthropic double Claude Code's 5-hour usage caps across paid tiers immediately, with more relief coming via API. The hardware: Anthropic gets all of Colossus 1, a 300+ MW Memphis supercluster, with more than 220K Nvidia GPUs coming online within the month. What users get: Claude Code's 5-hour usage caps just doubled across paid tiers, with additional increases via API and no more peak-hour restrictions. Musk's framing: He posted on X that SpaceX will rent compute to "AI companies that are taking the right steps to ensure it is good for humanity," which is a jab at OpenAI dressed up as a mission statement.
The Google angle: The Information reported yesterday that Anthropic is committing to a separate $200B, 5GW deal with Google Cloud over the next five years. Anthropic is sourcing compute everywhere it can find it. Why it matters: This is the enemy-of-my-enemy play in the open. Musk gets to patch the compute hole at OpenAI's biggest rival while SpaceX(AI), apparently the new name, turns into a compute landlord.
Anthropic gets to ship more Claude.
Everyone in this trade is mad at OpenAI. No bot. No awkward intros. Just notes that are ready to use. Supernormal joins your meetings quietly and captures everything from the first second.
When your call ends, your notes are already structured and ready to go: Follow-up email Client recap Task list with owners Strategy doc No rewriting. No cleanup. No wasted time. Most teams spend 30 to 60 minutes after every call.
This cuts it to seconds. Download free and capture your next meeting without a bot. Google Chrome Pushed a 4GB AI Model to Every User Without Asking A new researcher report says Chrome's recent updates have been pushing a roughly 4GB AI model called Gemini Nano onto user devices, with no opt-in, no notification and no easy way to remove it. The same report argues the practice may be illegal in the European Union.
The model: Gemini Nano powers Chrome's on-device AI features like writing assistance and tab grouping. It arrives in the background as part of normal browser updates, and most users never see it land. The EU problem: The researcher argues GDPR and the EU AI Act both require explicit consent before installing AI components that process user data. Google did not ask.
The energy bill: Multiplied across roughly 3 billion Chrome installs worldwide, the cumulative download alone is estimated to consume thousands of kilowatts. Storage and runtime add more on top. Google's position: Documentation describes Gemini Nano as opt-in for developers using Chrome's AI APIs, which is a technically true statement that says nothing about why the model showed up uninvited on user machines. Why it matters: Local AI is sold as the privacy upgrade.
The trade-off nobody mentioned is that local also means nobody notices a 4GB install. If EU regulators decide Google needed consent for this, every browser shipping local models suddenly has a problem, and Chrome got there first. The Pentagon Handed Meta-Owned Scale AI $500M The Department of Defense awarded Scale AI a $500M contract for military AI work, roughly five times the value of Scale's previous Pentagon deal and the largest single AI-services contract the DoD has booked. Scale is now majority owned by Meta, which means a chunk of US military AI dollars is effectively flowing through Mark Zuckerberg's portfolio.
The work: Data preparation, model evaluation and red-teaming for military AI systems. Scale already built much of the labeled data underneath the frontier models everyone uses. The Meta angle: Meta took a majority stake in Scale earlier this year and pulled founder Alexandr Wang into its superintelligence team. The Pentagon contract effectively routes US military AI spend through Meta's portfolio.
Year over year: Scale's prior DoD work was closer to $100M annualized. This contract lands in a year when Pentagon AI spending is reportedly tracking 40% above last year, and one vendor is taking a much bigger slice. Other bidders: Palantir, Anduril and Microsoft all bid for adjacent work. The Pentagon picked the data labeling specialist over the model and platform players.
Why it matters: The data layer is the part nobody talks about, and it just got the biggest defense check of the year. Meta bought Scale nine months ago in a deal that looked like a talent grab. The Pentagon just turned it into a strategic asset, and it cost taxpayers $500M to confirm Zuckerberg called it right. Snap's $400M Perplexity Deal Fell Apart Snap and Perplexity have walked away from a $400M, two-year partnership that would have put Perplexity's AI search inside Snapchat.
Both sides are now talking publicly about why the deal collapsed, which is usually a sign neither one wants to look like the company that backed out. The terms: Perplexity was set to pay Snap $400M over two years for placement inside Snapchat's chat surface, putting its product in front of roughly 800M monthly users. What killed it: Disagreements over data sharing, brand placement and which company owned the user-facing experience. Perplexity's valuation talks reportedly shifted mid-negotiation, which made the economics look different on the way out than on the way in.
What Snap wanted: A revenue line that did not depend on advertisers and a wedge into AI-native traffic. Snap's ad business is healthy but slowing, and a $400M check from Perplexity was a clean number. What Perplexity wanted: Distribution. Perplexity has the product and the brand recognition.
What it does not have is a user base that can compete with ChatGPT, and Snapchat would have changed that overnight. Why it matters: AI distribution deals at this scale are hard to close because the integration cuts deep on both sides and competitive dynamics shift mid-negotiation. Perplexity still needs distribution. Snap still needs AI revenue separate from its ad load.
Tonight, neither got it. Tool of the Day: Wisdom AI Wisdom AI is a conversational business intelligence platform that lets you ask plain-language questions about your company's data and get charts, trends and analysis back in seconds. It connects to your data warehouse, learns how your metrics are structured and replaces a chunk of the work analysts used to do by hand. Think of it as a research analyst that never sleeps, sitting on top of Snowflake or BigQuery.
Try it yourself: Head to wisdom.ai and sign up Connect your data source Ask a plain-language question like "What were our top three regions by revenue last quarter?" Refine the answer with follow-ups until the chart or insight matches what you actually wanted Save the analysis as a dashboard tile your team can come back to This is a fit for analysts, founders and operations leads who run on data warehouses and have more questions than time. Light Bytes Berkeley AI pioneer warns the Musk-Altman jury about the AGI race: Stuart Russell told the Oakland court that the "winner take all" race for AGI is the safety risk, no matter who wins it. Canada formally accuses OpenAI of breaking its privacy law: The federal privacy commissioner ruled OpenAI's ChatGPT training violated PIPEDA and ordered compliance, the polite Canadian way of saying you broke the law. Nvidia and Corning are building two more US optical factories: Plants in North Carolina and Texas will pump out optical interconnects for AI data centers, the infrastructure frontier models actually run on.
New AI surveillance towers go up along the US-Mexico border: Customs and Border Protection deployed computer-vision towers at remote crossings, with civil-liberties groups already pushing back on data storage and sharing. Google's Gemma 4 hits 3x faster inference with speculative decoding: The new open-weights release bakes draft-and-verify into the model itself and runs locally on a laptop with 16GB of RAM.