Elon Musk touts NVIDIA GB300 compute
- Elon Musk used X to hype NVIDIA’s GB300 systems while the bigger news landed elsewhere: SpaceX agreed to lease Colossus 1 compute to Anthropic. - The scale number is the hook — more than 220,000 Nvidia GPUs and 300 megawatts of capacity are slated for Colossus 1 by May. - That matters because AI competition is shifting from model bragging to compute control — who owns the clusters, power, and customer access.
AI news keeps pretending to be about models. But a lot of it is really about datacenters, power, and whoever can line up the most GPUs. That is why Elon Musk’s latest praise for NVIDIA’s GB300 matters less as a product endorsement and more as a signal about where his AI empire is trying to go next. On May 6, the concrete development was not just a post on X — it was a deal for Anthropic to use the compute capacity at Colossus 1, the giant Memphis cluster tied to xAI and now folded into SpaceX’s infrastructure story. (wsj.com) ### What is GB300, exactly? GB300 is NVIDIA’s new rack-scale Blackwell Ultra system. The flagship setup, GB300 NVL72, links 72 Blackwell Ultra GPUs with 36 Grace CPUs in one liquid-cooled platform. NVIDIA is pitching it for reasoning workloads — basically the kind of inference that gets expensive when models spe(wsj.com)her attention performance than earlier Blackwell systems. (nvidia.com) ### Why would Musk brag about that now? Because the industry argument has changed. A year ago, companies mostly bragged about model quality. Now they brag about compute stacks — chips, racks, networks, cooling, and power delivery. If Musk is touting GB300 as the best AI computer, he is also telling customers and investors that xAI wants to stay on the front edg(nvidia.com)us being marketed as a training supercomputer and with xAI’s own site highlighting 150,000-plus GPUs already running jobs at scale. (x.ai) ### What actually happened with Colossus 1? The immediate news is the Anthropic deal. Multiple outlets reported on May 6 that SpaceX would provide Anthropic access to Colossus 1, with the Wall Street Journal saying the buildout reaches 300 megawatts and more than 220,000 Nvidia GPUs by the end of the month. That turns Colossus from an internal Musk compute weapon into a commercial capacity business too — at least in practice. (wsj.com) ### Why is 220,000 GPUs such a big deal? Because at that size, the cluster stops being “a lot of servers” and starts acting like industrial infrastructure. The bottleneck is no longer just buying chips. It is finding land, substations, transformers, cooling loops, networking gear, and enough engineering discipli(wsj.com)9% uptime. Add another 70,000 or so GPUs and you are talking about one of the world’s most concentrated pools of AI compute. (x.ai) ### Why would Anthropic use a rival’s cluster? Because the frontier AI market is weird now. Anthropic competes with xAI on models, but compute shortages can matter more than rivalry. If Claude demand is outrunning Anthropic’s existing capacity, renting a giant ready-made cluster can be faster than waiting for new datacenters to come online. Wired framed the deal as a strange partnership between riv(x.ai)ut turns out scarcity makes strange bedfellows. (wired.com) ### So is this really a Musk story or an NVIDIA story? Both, but in different ways. NVIDIA still supplies the picks and shovels. Musk is trying to own the mine, the power line, and now maybe the toll booth. The catch is that compute concentration cuts both ways — it can speed up model development, but it also gives a tiny number of companies outsized control over who gets frontier-scale AI access. (nvidia.com) ### Bottom line? Musk’s GB300 hype is the surface-level post. The deeper story is that giant GPU clusters are becoming businesses in their own right. Whoever controls those clusters will shape the next phase of AI more than any single boast about a model.
