AI Mornings with Andreas Vig
Your daily AI news briefing in under 10 minutes. New models, product launches, research breakthroughs, and industry shifts, explained clearly, no hype.
AI Mornings with Andreas Vig
Anthropic's SpaceX Compute Deal & DeepSeek's $45B Valuation
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Hey, welcome to AI Mornings with Andreas Vig. It's the 7th of May 2026, and we've got a wild one today. The biggest story is about two rivals becoming partners. Anthropic just signed a deal with SpaceX to use all of the compute capacity at their Colossus 1 data center in Memphis. That's over 220,000 NVIDIA GPUs and 300 megawatts of power available within the month. What makes this remarkable is that SpaceX houses XAI, Elon Musk's AI company, and a direct competitor to Anthropic. So you have Anthropic essentially renting XAI's infrastructure. The deal will directly improve capacity for Claude Pro and Max subscribers, and Anthropic also said they're interested in partnering with SpaceX to develop orbital AI compute capacity data centers in space. Along with this, Anthropic doubled Claude Code's rate limits, removed peak hour restrictions, and raised API limits for Claude Opus models. This adds to their already massive compute pipeline: 5 GW with Amazon, 5 Gigawatts with Google and Broadcom, $30 billion in Azure capacity with Microsoft, and a $50 billion infrastructure investment with FluidStack. Speaking of SpaceX, they're planning something even bigger. The company is considering spending up to $119 billion on a semiconductor factory in Texas dubbed Terrafab. The initial investment would be at least $55 billion. This is a joint effort with Tesla and Intel to produce chips for AI servers, satellites, SpaceX's proposed orbital data center, and Tesla's autonomous vehicles and robots. Musk has argued that semiconductor manufacturers simply aren't making chips fast enough for his companies. The combined SpaceX XAI entity is reportedly valued at $1.25 trillion with an IPO expected in June. Now to China, where DeepSeak is raising its first ever venture round, and the numbers are striking. The valuation has jumped from $20 billion to $45 billion in just weeks. The round is led by China's state investment fund, with Tencent and Alibaba reportedly participating. DeepSeq came to prominence in early 2025 for training competitive models at a fraction of the cost of U.S. labs. The founder, a Chinese hedge fund billionaire, controls nearly 90% of the company and decided to raise funds primarily to offer employees' shares and compete for talent. DeepSeak has optimized its models to run on Huawei chips, which is strategically important for China as it seeks to develop homegrown AI that bypasses US technology restrictions. OpenAI is getting into phones. Supply chain analyst Ming Chi Kuo reports that OpenAI is fast-tracking its first AI phone for mass production in the first half of 2027, a full year earlier than previously planned. MediaTech will be the sole chip supplier, and the device will use two AI processors to handle vision and language tasks simultaneously. The standout feature is an enhanced image signal processor designed to improve AI agents' real-world visual sensing. GUO estimates potential shipments of 30 million units across 2027 and 2028. What's unclear is how this relates to OpenAI's device work with Joni Ives I.O. which the company acquired last year promising products beyond screens. Alright, a few more things worth knowing about today. Samsung just hit a trillion dollar market cap, becoming only the second Asian company after TSMC to reach that milestone. Shares jumped more than 10% after an earnings report showing profits eight times higher than a year ago. The driver is high bandwidth memory chips, essential for AI systems, where demand is crushing supply. Apple is also reportedly in talks with Samsung and Intel to manufacture chips on US soil, which would be a significant shift from their heavy reliance on TSMC in Taiwan. A Miami startup called SubQuratic came out of stealth with 29 million in seed funding and a bold claim. They've built the first fully subquadratic frontier language model. Called SubQ, it supposedly uses about a thousand times less compute than standard models at a 12 million token context window. That's the kind of context that could let agents work for weeks without degrading. But researchers are asking for independent verification, noting that previous subquadratic architectures have struggled to match traditional attention at frontier scale. The company says the model is 50 times cheaper than leading competitors. Coinbase is cutting 14% of its workforce, about 700 people. CEO Brian Armstrong cited crypto market conditions and a strategic shift toward AI native teams. The company is flattening its org chart and focusing on employees who can drive outsized impact using AI tools. It's part of a broader trend we're seeing of companies reallocating headcount budgets toward AI capabilities. Finally, a California startup called Span is partnering with Nvidia to put mini AI data centers on the exterior walls of residential homes. The XFRA boxes use Nvidia's liquid-cooled GPUs and can be installed six times faster than building a traditional data center at one fifth the cost. They're already testing with a major home builder. It's a creative approach to grid strain from data center demand, though I suspect not everyone will want a data center box mounted on their house. That's it for today. See you tomorrow.