Fortune warns AI compute concentration

Fortune published an analysis on May 16 arguing that artificial-intelligence compute is concentrating in the hands of a small group of companies and countries, deepening the power dynamics that shaped social media. The piece was written by David Liberman and Daniil Liberman, entrepreneurs who said they saw a similar pattern while working at Snap in 2018 as Instagram gained ground with a copy of Stories. They argued that AI concentration sits below the application layer, in cloud infrastructure, chips and model access, making dependence harder to unwind than dependence on a social platform. The article appeared in Fortune’s AI and cybersecurity sections on Saturday. ### What pattern did the Libermans say they recognized from social media? David and Daniil Liberman wrote that Meta’s control of the “layer beneath Snap” — the social graph, feed and audience — helped determine outcomes even when Snap had a strong product. They said that once a company controls that underlying layer, products built above it “bend toward” that platform. (fortune.com) In the Fortune essay, the brothers said AI is repeating that pattern at a deeper level because compute is a prerequisite for training and deploying models. They wrote that a startup without cloud contracts cannot train and that countries without chip access cannot compete, framing compute access as the gate to participation in the AI economy. (tech.yahoo.com) ### Which companies and numbers did the article use to show concentration? Synergy Research Group said Amazon, Microsoft and Google accounted for 63% of enterprise spending on cloud infrastructure services in the third quarter of 2025, a figure the Fortune authors cited to describe concentration in the cloud layer. The same trio has been the core infrastructure base for much of the generative-AI buildout. (tech.yahoo.com) The Fortune article also said Nvidia holds 85% of the data-center GPU market. Web search results show that outside estimates vary, with some analysts placing Nvidia’s AI accelerator share in a range around 80% to the high 80s, but the essay used 85% as its benchmark for chip concentration. Epoch AI and a 2025 Federal Reserve note describe a similar geographic imbalance in high-end AI compute. (srgresearch.com) Epoch AI said the United States hosts about 75% of AI supercomputer performance in its dataset, with China at 15%, while the Fed cited an estimate of 74% for U.S. control of global high-end AI compute and 14% for China. ### Why did the piece say this matters for companies building with AI? (tech.yahoo.com) The Fortune essay said reliance on a single cloud, model or chip stack can make companies brittle because the dependency sits inside the systems needed to run AI products at all. The authors contrasted that with social media, where a new app could still emerge, citing TikTok as proof that application-level disruption remained possible. (epoch.ai) The Libermans argued that AI leaves less room for that kind of reset because compute, data-center access and model distribution are already concentrated before a product reaches users. Their article presented that as a strategic risk for builders whose products depend on one provider’s infrastructure or model layer. ### What kinds of responses did the article point toward? (tech.yahoo.com) Fortune’s analysis said the concentration trend would increase demand for abstraction layers, portable deployment systems and other tools that let companies shift workloads across providers. The article cast those tools as a response to dependence on a single vendor stack rather than as a consumer-facing product category. (tech.yahoo.com) Deloitte said in a 2026 note that enterprises are reassessing compute strategy as inference workloads raise questions about cost, data sovereignty, latency, intellectual-property protection and resilience. That does not mirror Fortune’s argument exactly, but it shows large companies are already treating AI infrastructure choices as an operating issue rather than only a model-selection question. (tech.yahoo.com) ### Who are the authors and where was the article published? Fortune listed David Liberman and Daniil Liberman as the authors of the May 16 article, and Yahoo’s syndicated version identified them as the Liberman brothers. Fortune’s section pages show the piece under both AI and cybersecurity coverage. As of May 17, the article remained available through Fortune’s section pages and through Yahoo’s syndicated posting, which reproduced the essay and its publication date. (deloitte.com) (fortune.com)