Cloud compute is renting processing power, storage and AI accelerators from a provider's data centre over the internet, paying per second, per gigabyte or per token instead of buying and running the machines yourself.
No jargon, no math
You send your job (a training run, a batch of inference calls, a database) to a provider's data centre. Their building, their power bill, their cooling, their broken-hardware swaps at 3am. You get virtual machines, often with GPUs or TPUs attached, plus a control panel that scales them. Elasticity is the whole point: run one machine for an hour, or a thousand for a weekend, then give them back. Pricing comes in two flavours: reserved capacity you commit to for a discount, and on-demand capacity billed per second at a premium. Two things travel with every job: your data goes to their building, and latency is a round trip to wherever that building is. Where it breaks: a forgotten training job can burn thousands overnight, moving everything to another provider is real work, and if your data is not allowed to leave your premises, cloud compute is simply not an option.
A GPU is a processor built to run thousands of small calculations at the same time. Designed for graphics, it turned out to be the perfect engine for AI, because running a neural network is mostly one enormous pile of matrix multiplications.
ReadA TPU is Google's custom chip built for a single job: the matrix multiplications at the heart of neural networks. You do not buy it; you rent it by the hour inside Google Cloud, where it often beats general-purpose GPUs on training cost.
ReadTraining is the process that turns an untrained model into a working AI: you feed it thousands or millions of examples, compare its guesses with the correct answers, and nudge its internal settings until the guesses get good.
Read§ 03
We design and ship AI systems with permissions, logs and human approval. It starts with a 30-minute scoping call.