About
SemiAnalysis is an independent research and analysis firm covering the semiconductor and AI supply chain, including capital equipment, foundries, accelerators, datacenters, and AI models. It is read by more than 180,000 subscribers.

SemiAnalysis publishes independent research and data models for semiconductors and AI infrastructure.
semianalysis.comCrawled Oct 15, 2026
SemiAnalysis is an independent research and analysis firm covering the semiconductor and AI supply chain, including capital equipment, foundries, accelerators, datacenters, and AI models. It is read by more than 180,000 subscribers.
Bridging the gap between the world's most important industry, semiconductors, and business.
The company’s privacy policy describes how it collects, uses, and shares information through its website and services.
The terms apply to the website and interactions or transactions conducted on or through it. They also apply to the company, its affiliates, and parties against whom related disputes are asserted.
The global datacenter industry is adding over **60GW of capacity under construction**, with SemiAnalysis projecting **50GW of new capacity by 2027**. Lease contracts totaling **1.3 terawatts (TW) of demand** have already been signed year-to-date. Industry analysts warn of a potential **chip supply shortage** as demand accelerates.
Dario Amodei argues that frontier AI labs should deliberately slow the pace of model capability improvements. The article frames this proposal against Beijing’s speed-first AI safety regime and describes AI safety as an urgent debate.
Beijing is rapidly rolling out AI infrastructure in Inner Mongolia. China has 24 GW of operational compute capacity and another 50 GW planned or under construction, compared with 56 GW operational in the US.
SemiAnalysis measures usage limits across AI subscription plans and introduces a dashboard tracking plans from Anthropic, OpenAI, Meta, SpaceXAI, Cursor, Cognition, Z.ai, MiniMax, and Moonshot. Its rough estimate says Anthropic subscriptions generate 10% of revenue while consuming more than 40% of inference compute, lowering blended revenue per megawatt by about $36 million. Experiments isolate input, cache-write, cache-read, and output tokens to estimate how plans use limits across models.
Powerful AI tools burn through budgets, prompting a rethink of how the technology is priced. The debate comes ahead of frontier lab IPOs.
SemiAnalysis maps more than 300 local datacenter moratoriums and argues they are not stopping the US buildout. Its model finds 20GW of capacity within restricted local boundaries but estimates that only 1,525MW actually slips, while forecasting 38GW of US datacenter IT capacity delivered in 2027—more than double 2026. The forecast includes 22GW under vertical construction and 16GW planned, most already financed and preparing for construction.
Citrini Research founder James van Geelen sells the independent investment-research shop to semiconductor and AI researcher SemiAnalysis for an undisclosed sum. Sources say van Geelen plans to launch a new fund.
Positron AI raises $875 million in a Series C round at a $5 billion valuation to bring its inference silicon to market. Its Titan appliance uses LPDDR5X memory instead of high-bandwidth memory and runs on custom Asimov chips. Positron says simulations show a rack powered by its silicon can process up to 26 times as many tokens per dollar as Nvidia’s Blackwell GB300 NVL72 appliance. The company plans to tape out Asimov at year-end and begin mass production in the second half of 2027.