From Replit’s public split with Anthropic to Nvidia’s European tour and CoreWeave’s distressed bonds, Saturday’s biggest tech stories trace one industry’s loud, public adolescence.
The coding world’s messiest public breakup
Replit, the AI coding platform where a generation of startup founders now build, is apparently done with Anthropic. CEO Amjad Masad said so onstage at TechCrunch Disrupt on Thursday, and he did not mince words: “I think ChatGPT is better now.”
The decision, Masad said, came down to cost, partnership, and performance. OpenAI’s Sol and Astra models score better on Replit’s internal benchmarks for building vibe-coded apps from scratch. That is remarkable given how central Claude has been to Replit’s story: Claude models powered the platform for years and were credited with driving its revenue growth. Replit announced an OpenAI partnership back in August, but nobody realized the Anthropic chapter was closing.
Two stories, one contradiction
The breakup may be messier than the onstage version. A person familiar with the partnership told TechCrunch it is still in place, and a Replit blog post from late September still lists Claude Fable 5 and 5.1 running in Replit Agent’s production environment. So either the split is very fresh, or the CEO got ahead of his own team.
Either way, Masad dropped a number with real weight: Replit was one of Anthropic’s top five customers. That makes this a meaningful revenue question for Anthropic, not just a vibes story. And it signals the new normal in the model wars: performance leads change every few months, costs are brutal at scale, and partnerships are dating, not marriage. The switching costs are low enough that a CEO announces a breakup on a conference stage. Watch whether either side clarifies — right now they’re telling different stories.
Nvidia’s big European week
From breakups to buildups. Nvidia had quite a week in Europe, with Jensen Huang essentially on a continental tour ahead of the company’s GTC event in Berlin. Two deals tell the story.
First, Volvo Cars is expanding its Nvidia partnership on automated driving, pairing Volvo’s safety engineering with Nvidia’s Hyperion compute architecture and autonomous-driving software stack. The first deployment targets supervised driving, but Volvo told Reuters the ambition is hands-free driving in cities and on highways, arriving in cars before 2030. Notably, the features will not be reserved for expensive models: they are meant to launch in multiple regions at once, excluding China.
Second, Reuters reports that Germany will announce a cooperation agreement with Nvidia next week for industrial AI. Handelsblatt, which broke the story, says it covers physical AI tools for training robots and self-driving vehicles, plus industrial AI models.
Step back and the picture is stark: Europe’s automakers and its biggest industrial economy are consolidating around one AI stack, right before Nvidia’s Berlin showcase. Carmakers get a proven stack and a faster path to autonomy — but the continent’s driving future increasingly runs through Santa Clara.
The bond market sounds the alarm on AI infrastructure
Now the less comfortable side of the AI boom. CoreWeave, the AI cloud provider that rents out GPU capacity, just had its bonds flash a distress signal. Bloomberg reported that the company’s unsecured bonds due in 2032 traded below 80 cents on the dollar — the market’s accepted distress threshold. At that price, they yield more than 15 percent.
The peanut gallery showed up. Chamath Palihapitiya flagged the yield, and Elon Musk replied with a single word: “yikes.” The credit-default-swap market is gloomier still: spreads imply roughly a 60 percent chance of default within five years. And CoreWeave carries around $50 billion in debt, with about $4.4 billion due by the end of 2026.
But not everyone is running for the exits. Jensen Huang, speaking in Stockholm, told Bloomberg that CoreWeave’s strength is improving every day and that he does not think the company will need a bailout from Nvidia. Cathie Wood’s ARK bought the dip, picking up roughly 188,665 shares — about $12.4 million.
The bigger question is what this says about the AI infrastructure trade. These companies borrowed enormously against equipment and future demand, and the bond market is now asking whether the cash shows up before the bills do.
A bright spot: Claude designs leaner quantum circuits
Let’s end on something genuinely cool. Anthropic’s Claude designed quantum circuits this week, and the results are wild. Anthropic gave Claude Mythos 5 plain-language descriptions of eight circuit design problems, plus an automated evaluator that scored each design — then let the model iterate in a closed loop, with no human input between rounds.
In seven of eight tasks, Claude beat the best published designs. The headline: a circuit for elliptic-curve point addition that used 17 times fewer of the costliest gates than the previous best, at the cost of 12 percent more qubits. Another design came in 12 times leaner than Google’s prior construction. The hardest design took a single 53-hour run with minimal prompting — Anthropic notes that a comparable improvement in one chemistry estimate previously took many research groups eight years.
Caveats apply: these are designs and estimates, verified by simulation and Lean proofs, not runs on real hardware. Still, it is a nice proof point. Give a model checkable tasks and an honest scorekeeper, and it can do real engineering, not just chat.
Quick hits: shopping carts, robotaxis, and retiring models
- Google launches Universal Cart in the US. A single intelligent cart that follows you across Google Search and the Gemini app, so you can save products from different retailers in one place. It tracks price drops and restocks, surfaces promo codes, and factors in your credit card perks at checkout, which runs on Google’s Universal Commerce Protocol. Launch partners include Nike, Sephora, Target, Ulta Beauty, Walmart, Wayfair, and select Shopify merchants — clearly timed for the holidays.
- Waymo is going international again. The company said Friday it plans to bring its robotaxi service to Toronto in 2028 — its first market in Canada. It is partnering with Toronto-based Element Fleet Management, starts mapping the city now, and promises hundreds of local jobs over time.
- Developers, heads up. GitHub retires three Copilot models on November 16: Claude Opus 4.8, Claude Opus 5, and GPT-5.6 Terra. Annual individual plans keep Opus 4.8; everyone else should switch to Claude Opus 5.5, or GPT-6.1 Sol and GPT-6 Luna, before the cutoff. Nobody wants their pinned model vanishing mid-sprint.
The thread tying it together
One thread ties this week’s stories together. The AI industry is growing up in public. Partnerships are getting renegotiated on stage, governments are placing national bets on compute, and the bond market is grading the infrastructure boom like a credit analyst, not a cheerleader. And underneath all of it, the models keep getting more capable at real work — from designing quantum circuits to building your app.
Sources
- Replit CEO says company is no longer working with Anthropic: TechCrunch
- Volvo expands automated-driving partnership with Nvidia: Reuters
- Germany to unveil Nvidia AI agreement next week: Reuters
- CoreWeave bonds tumble into distress: Bloomberg
- Claude designs leaner quantum circuits (Anthropic research): no verified source available — details as reported by the company
- Google Universal Cart launches in the US: TechCrunch
- Waymo plans first Canada robotaxi service in Toronto in 2028: Reuters
- GitHub Copilot retires three models on November 16: GitHub
Adapted from the October 17, 2026 episode of Daily Tech and AI.
Daily Tech and AI is AI-voiced and produced by Angelo Ruiz.
Leave a Reply