Often less than the headline, and sometimes more. A compute commitment is a contract to buy computing capacity over many years; some are binding take-or-pay contracts that must be paid even if the buyer does…
Anthropic's filing shows about $33.2 billion of revenue in the first nine months of 2026, an operating loss of $567 million in the third quarter, and about $895 billion of compute and infrastructure…
According to reports in the Financial Times and The Wall Street Journal, OpenAI lowered its current funding round from $30 billion to $20 billion and has about $11 billion committed so far.
A gigawatt is a measure of power, a billion watts, roughly what a million homes draw, and AI companies now use it as their main unit for the size of a data center.
Muse, dots and Claude Duet are managed services that run on the company's cloud computers, aimed respectively at consumers, paying ChatGPT users and professionals.
Nearly two years after its announcement, Stargate is a set of seven US sites with more than 9 gigawatts planned, and Abilene, Texas, is the one clearly running at scale, at an estimated 824 megawatts of…
Investors reacted less to sales than to cash: Alphabet and Amazon reported fast cloud growth, but their spending on data centres and chips outran the cash their businesses produced, and Intel gave a weak…
Six groups pay for most of it: cloud and chip companies investing in their own customers, venture and growth funds, sovereign wealth funds, public stock markets, bond and private credit investors, and…
The NAACP alleges that xAI ran 27 gas turbines in Southaven, Mississippi, without an air permit to power its Colossus data centers near Memphis, in violation of the Clean Air Act.
Nobody can prove it yet, because a bubble is usually confirmed only after prices fall.
Ottawa, Cohere and the big telecoms say Canada needs AI capacity it controls, while critics ask whether public money buys real control, whether it should go to one firm, and whether the data centre model is…
Capex, short for capital expenditure, is money spent on long-lasting assets such as data centres, servers and chips, rather than on day-to-day costs.
HBM, or high-bandwidth memory, is a stack of memory chips placed right beside an AI processor so data can reach it fast enough.
Residents mainly object to higher power bills, water use, noise, air pollution and secrecy, while supporters point to tax revenue, construction work and national competition in AI.
AI labs earn money mainly by charging for usage through an API, selling monthly subscriptions, and increasingly from ads and from renting out computing power.
On-device AI runs on your own phone or computer, so your data stays there.
Skip the opening story and go straight to the financial tables, the cash flow statement, the contractual obligations, the risk factors, the voting structure and the lock-up section.
Circular financing is when a supplier invests in or guarantees a customer, and the customer then spends that support on the supplier's products.
The argument is about whether handing an AI access to your email, money and accounts is worth the risk, and about whose computer it should run on.
No single rule says you own it. With a hosted assistant, your memories sit on the company's servers under its terms and privacy law, which can give you rights to see, correct and sometimes move your data;…
Canada's AI compute is being built mainly by telecoms such as Bell and Telus, American and Canadian specialist cloud firms, and foreign hyperscalers, with Ottawa adding money, a planned public supercomputer…
Grids can carry the load in most places over time, but not at the speed AI companies want, so connection queues, on-site gas plants and fights over cost are now the main bottlenecks.
Companies like Nvidia, AMD and Broadcom design AI chips but do not make them, and most of the most advanced ones are manufactured and packaged by TSMC, whose leading factories are concentrated in Taiwan.
Compute limits AI in three places at once, how large a model a lab can train, how many experiments it can run, and how many people it can serve.
AI gigafactories are up to seven very large AI computing sites the European Union wants built by private consortia, with public buyers guaranteeing part of their revenue.
The US Commerce Department requires a licence to export chips above set performance limits to China and certain other countries, and it has rewritten those rules several times since October 2022.
India pools tens of thousands of GPUs and subsidizes them so local startups can train models for Indian languages, while Japan is funding a new company, led by SoftBank and others, to build AI for robots and…
Saudi Arabia and the UAE are using oil wealth, cheap land and energy to become AI hubs through state-backed companies, Humain and G42.
The United States offers allies a packaged American AI stack, backed by federal support and tied to export rules, while China offers open-weight models, training and a new international organization.
All three are chips built to do huge numbers of multiplications at once, which is most of what an AI model does.
Public AI compute means supercomputers and shared resources paid for by governments and allocated to researchers, usually through an application rather than a bill.
AI data centers move heat out with liquid running to the chips, then release it through evaporation, which uses water, or through fans and chillers, which use more electricity.
AMD competes head-on by selling its own GPUs in complete racks, while Broadcom helps big customers such as Google and OpenAI build chips of their own.
In Quebec, digital sovereignty means keeping public and personal data out of reach of foreign law while using the province's research strength at Mila and its hydroelectricity to build AI capacity at home.
All four are open models built mainly outside the big commercial labs, but they differ in who controls them and how much they reveal.
Nvidia leads because it sells fast chips together with almost two decades of software, above all CUDA, the toolkit developers use to program its GPUs, plus the networking that joins thousands of chips and the…
The idea is serious enough that Google, SpaceX and Blue Origin are working on or filing for it, because solar panels in the right orbit produce far more than on Earth and need no land or local permits.
An AI data center is a factory that turns electricity into calculations, built around racks of tightly linked chips that draw far more power than ordinary servers.
The public AI movement is a loose network of researchers, foundations and public institutions arguing that some AI should be built and governed like public infrastructure, not only sold by private companies.
A dense model uses all its parameters for every token, while a mixture-of-experts model uses only a small fraction, so it gets more capacity for the same computing per token.
AI for All is the national strategy Ottawa released on June 4, 2026, built on six pillars from protecting Canadians to scaling Canadian companies.
On April 24, 2026, Toronto's Cohere and Germany's Aleph Alpha announced a combination meant to give businesses and governments an alternative to American AI giants.
Apertus is a large language model released on September 2, 2025 by EPFL, ETH Zurich and the Swiss National Supercomputing Centre, in sizes of 8 and 70 billion parameters.
It is a federal plan launched on December 5, 2024 that spends up to $2 billion on the computing power behind AI, split among commercial data centres, public supercomputers and help for smaller companies.