This week, a subject that has been boomeranging round Silicon Valley bounced into the highlight: AI tokens as compensation. The thought is easy sufficient — moderately than giving engineers solely wage, fairness, and bonuses, firms would additionally hand them a price range of AI tokens, the computational models that energy instruments like Claude, ChatGPT, and Gemini. Spend them to run brokers, automate duties, crank via code. The pitch is that entry to extra compute makes engineers extra productive, and that extra productive engineers are price extra. It’s an funding within the particular person holding them, is the concept.
Jensen Huang, the leather-jacket-wearing CEO of Nvidia, appeared to seize everybody’s creativeness when he floated the notion on the firm’s annual GTC occasion earlier this week that engineers ought to obtain roughly half their base wage once more — in tokens. His high folks, by his math, would possibly burn via $250,000 a year in AI compute. He known as it a recruiting software and predicted it will turn out to be customary throughout Silicon Valley.
It isn’t solely clear the place the concept was first, effectively, ideated. Tomasz Tunguz, a famend VC within the Bay Space who runs Idea Ventures and focuses on AI, information, and SaaS startups — and whose writing on all issues information has garnered a loyal following through the years — was speaking about this in mid-February, writing that tech startups have been already including inference prices as a “fourth component to engineering compensation.” Utilizing information from the compensation monitoring website Ranges.fyi, he put a top-quartile software program engineer wage at $375,000. Add $100,000 in tokens and also you’re at $475,000 absolutely loaded — which means roughly one greenback in 5 is now compute.
That’s no coincidence. Agentic AI has been taking off, and the release of OpenClaw in late January accelerated the dialog significantly. OpenClaw is an open-source AI assistant designed to run repeatedly — churning via duties, spawning sub-agents, and dealing via a to-do listing whereas its consumer sleeps. It’s a part of a broader shift towards “agentic” AI, which means methods that don’t simply reply to prompts however take sequences of actions autonomously over time.
The sensible consequence is that token consumption has exploded. The place somebody writing an essay would possibly use 10,000 tokens in a day, an engineer operating a swarm of brokers can blow via thousands and thousands in a day — mechanically, within the background, with out typing a phrase.
By this weekend, the New York Occasions had put collectively a smart look on the so-called tokenmaxxing development, discovering that engineers at firms together with Meta and OpenAI are competing on inner leaderboards that monitor token consumption. Beneficiant token budgets are quietly changing into an ordinary job perk, the paper reported, the best way dental insurance coverage or free lunch as soon as was. One Ericsson engineer in Stockholm instructed the Occasions he in all probability spends extra on Claude than he earns in wage, although his employer picks up the tab.
Perhaps tokens actually will turn out to be the fourth pillar of engineering compensation. However engineers would possibly wish to maintain the road earlier than embracing this as a simple win. Extra tokens might imply extra energy within the brief time period, however given how briskly issues are evolving, it doesn’t essentially imply extra job safety. For one factor, a big token allotment comes with massive expectations. If an organization is successfully funding a second engineer’s price of compute in your behalf, the implicit stress is to supply at twice the speed (or extra).
Techcrunch occasion
San Francisco, CA
|
October 13-15, 2026
And there’s a muddier drawback beneath that: on the level the place an organization’s token spend per worker approaches or exceeds that worker’s wage, the monetary logic of headcount begins to look completely different to its finance staff. If the compute is doing the work, the query of what number of people should be coordinating it turns into tougher to keep away from.
Jamaal Glenn, an East Coast-based Stanford MBA and former VC turned monetary providers CFO, equally points out that what might look like a perk generally is a intelligent manner for firms to inflate the obvious worth of a compensation bundle with out growing money or fairness — the issues that really compound for an worker over time. Your token price range doesn’t vest. It doesn’t recognize. It doesn’t present up in your subsequent supply negotiation the best way a base wage or fairness grant does. If firms efficiently normalize tokens as pay, they might discover it simpler to maintain money comp flat whereas pointing to a rising compute allowance as proof of funding of their folks.
That’s a great deal for the corporate. Whether or not it’s a great deal for the engineer depends upon questions most engineers don’t but have sufficient info to reply.
