what is a $900000 ai job?

what the number is made of

The headline figure is total compensation, not salary. A typical package at that level breaks down roughly like this:

The equity half is the part that moves. It is priced at a private valuation until there is a liquidity event, so two people quoting the same $900k can be holding very different things. That is worth knowing before you treat the number as income.

who is actually paid it

Four clusters, and almost nobody outside them:

• Frontier labs. Openai, anthropic, google deepmind, meta superintelligence, ssi, thinking machines. Senior research scientists and research engineers.
• Big tech ai orgs. Level 6 and above at google, meta, microsoft, nvidia, apple, in roles tied directly to model training or inference at scale.
• Well-funded ai startups buying senior people out of the above, where the cash is lower and the equity share is larger.
• Trading firms and a few quant shops applying the same skills to markets, where the cash share is much higher.

The common thread is not "knows ai". It is having shipped something at a scale very few people have touched: a training run over thousands of accelerators, an inference stack serving millions of requests, or a research result other labs cite.

the honest version of the path

There is no six-month route. Every profile in this band has some mix of a strong systems background, published or shipped work that is externally checkable, and three to ten years of doing it. The scarcest skill is not model architecture. It is distributed systems engineering applied to training and serving, which is why so many of these hires come from infrastructure teams rather than from research programmes.

If you want the realistic ladder rather than the ceiling, how much ai software engineers get paid covers the bands most people actually land in, and how to become an ai software engineer covers the route.

why the number keeps circulating

Because it is the cheapest possible headline. A $900k salary story writes itself, gets shared, and gives every reader the same wrong impression: that ai pay is uniformly enormous. The median ai-adjacent engineering job in the united states pays a fraction of it. Both facts are true at once, and only one of them gets a headline.

The useful takeaway is about supply, not envy. Compensation that extreme means a specific skill is scarce relative to demand, and that gap closes over time. Historically these bands compress within about five years as tooling standardises and the training pipeline widens.