Rippling disclosed on 7 August that internal AI tooling had reached 40% of its R&D headcount budget — millions of dollars — while growing 80% month over month. On that trajectory the spend would have reached 90% of what the company pays its R&D staff annually.
Concentrated in a few hands
Between 10% and 15% of employees drove 60% of total AI spend. One engineer spent $50,000 in a single month. That distribution is the practical problem: a per-seat licence cannot price a workload where a tenth of the users generate most of the cost.
The saving did not come from using less
This is the finding most summaries drop. Token usage peaked at 605 billion in April. July usage was about 600 billion — essentially flat — but July cost was 37% of April's. Spend fell from 40% of the headcount budget to 15% without the company doing less AI work. The gain came from routing and model selection.
What it sells now
Rippling built an AI Spend Console, generally available to existing HR subscribers with usage-based costs on top, and sold standalone. Chief product officer Matt MacInnis put the pitch bluntly: "the inference providers, like Anthropic and OpenAI, have absolutely no incentives to help you control your spend."
Read the denominator
The 40% is a share of the R&D headcount budget, not of company spend or revenue, and the 90% figure is where the curve pointed rather than money that left the building. The disclosure is also self-interested — Rippling is selling the fix. That does not make the numbers wrong, but it is why they were published at all, and no other company of this size has put comparable figures on the record.