One term to define cleanly: "accelerated computing." It's in nearly every AI-hardware headline, usually undefined. The simplest place to pin it down is the filing of the company that popularized it.
The sec.gov filing, surfaced via EdgarBeast, describes "accelerated computing for computationally intensive workloads such as artificial intelligence, or AI, model training and inference, data analytics, scientific computing, robotics, and 3D graphics." Notice the definition is by example: accelerated computing is whatever is too heavy for an ordinary processor to do well.
The mechanism is a division of labor. A CPU is a generalist — it handles the operating system, the logic, the bookkeeping. For tasks that are massively repetitive and parallel, you hand the work to an accelerator (a GPU or similar) that's built to do thousands of identical operations at once. "Accelerated" just means: the hard part runs on the specialized chip, not the generalist.
Why the term caught on: it reframes the product. Selling "GPUs" sounds like selling graphics cards. Selling "accelerated computing" claims a much bigger territory — any heavy workload, not just AI. The 10-K's workload list is that claim, made formally: AI is first, but it sits among data analytics, scientific computing, and the rest.
So next time you read that a company is "investing in accelerated computing," translate it: they're buying specialized hardware to run the heaviest jobs faster, with AI usually the heaviest of all. The definition isn't marketing — it's right there in the workload sentence of the sec.gov filing, which is about as authoritative a glossary entry as you'll get.