AI compute and chip race
The market is shifting from model quality to control over compute and chips, with Anthropic reportedly exploring its own AI chips as it scales revenue and big players trying to lock supply. (cnbc.com) OpenAI told investors it had roughly 1.9 gigawatts of capacity in 2025 versus Anthropic’s ~1.4 gigawatts, while Google still hosts a majority of global AI compute — a dynamic that strengthens the case for portable data and governance layers. (qz.com) (networkworld.com)
Anthropic is reportedly considering building its own artificial intelligence chips instead of relying only on outside suppliers, even though the plan is still early and could be dropped. Reuters reported the move on April 10, 2026, after months of tight supply for the processors that train and run large language models. (cnbc.com) That sounds like a hardware story, but it starts as a power story. Training and serving artificial intelligence now depends on giant clusters of chips inside data centers, and the companies with the biggest clusters can launch larger models, answer more user requests, and cut waiting times. (networkworld.com) Investors have started measuring those clusters in gigawatts, which is an electricity unit usually used for power plants. OpenAI told investors it had about 1.9 gigawatts of capacity in 2025, while Anthropic had about 1.4 gigawatts, turning compute into a scoreboard instead of a back-office detail. (cnbc.com) OpenAI used that gap to argue that Anthropic was “compute constrained,” and the same memo said OpenAI wants 30 gigawatts by 2030. That is the language of utilities and heavy industry, not the language the artificial intelligence industry used when the fight was mostly about model benchmarks. (cnbc.com) Google sits above both of them because it already owns much of the plumbing. Network World reported on April 10 that more than 60% of global artificial intelligence compute capacity sits with hyperscalers, with Google in the lead after years of building custom chips, networking, and data centers as one stack. (networkworld.com) That is why custom chips matter. If a company designs its own processor, it can tune the chip for its own workloads, lower dependence on Nvidia’s supply, and tie software more tightly to the machines underneath it. (cnbc.com) Anthropic already has one foot in that world through partners. Broadcom said on April 6 that it expanded its deal with Anthropic, giving the startup access to about 3.5 gigawatts of computing capacity built on Google’s artificial intelligence processors. (cnbc.com) So Anthropic’s chip exploration is not a clean break from Google. It looks more like a hedge against a future where every major lab wants guaranteed access to silicon, and renting capacity from rivals becomes too expensive or too strategically risky. (cnbc.com) OpenAI is making the same bet from the other direction by locking up supply wherever it can. Quartz reported that OpenAI struck giant chip and cloud deals in 2025, including a six-gigawatt agreement tied to Advanced Micro Devices and a broader push to diversify beyond Nvidia and Microsoft. (qz.com 1) (qz.com 2) The result is that artificial intelligence competition is starting to look less like a software race and more like an airline race where gates, fuel, and aircraft matter as much as the ticket app. The labs still talk about safety, reasoning, and product quality, but underneath that they are fighting over transformers, substations, chip packaging, and who gets first call on the next shipment. (networkworld.com) (cnbc.com) That shift changes what customers should worry about too. If the winning companies are the ones that control the hardware stack, then businesses may care more about whether their data, workflows, and governance rules can move between clouds than about which model won a benchmark in one quarter. (networkworld.com)
