What Reinhart actually said
Speaking at the RetailClub AI Festival in Huntington Beach in September, ThredUp co-founder and CEO James Reinhart described the cost of scaling AI use across his company in blunt terms: token usage grew faster than he had expected, and his reaction was that he was both surprised and did not care at all. That is an unusual thing for a public company CEO to say about an expense line that grew beyond his own forecast, and it is worth taking at face value rather than reading it as spin, because he paired it with an explicit acknowledgment of the forecasting problem it creates.
To illustrate his point, Reinhart presented a slide deck at the event that he said he had never seen before walking on stage. It had been built using Claude and checked over by his assistant beforehand. He described the exercise as something he thought would be a really interesting experiment, using his own presentation as a live demonstration of the AI-first posture he was describing to the room rather than just talking about it abstractly.
The bet behind not caring
Reinhart's underlying argument is that employee AI use should not be throttled because of near-term cost uncertainty, since the productivity benefit should eventually outweigh the upfront spend. That is a defensible long-run bet, and it lines up with what he told investors on ThredUp's August second-quarter earnings call, where he said advancements in AI would provide significant cost savings for the business by slowing headcount growth and boosting productivity. The logic connects: if AI substitutes for headcount growth over time, a volatile token bill today is a reasonable price for a structurally lower cost base later.
The harder question is how a management team defends that bet to a board or a public market when the cost side of the equation keeps outrunning its own projections. Reinhart's answer leans toward improving the forecasting itself rather than pulling back on spending. He said explicitly that as a public company, ThredUp needs to get better at projecting these costs, which is a tacit admission that the current approach amounts to committing to the strategy first and building the budgeting discipline to match it after the fact.
Why this is a CFO problem disguised as a CEO quote
Most conversations about enterprise AI adoption focus on capability: what the model can do, which workflows it can touch, how fast it ships. Reinhart's comments point at a different and less discussed failure mode, which is that the unit economics of AI usage inside a company do not behave like a traditional software license. Token consumption scales with employee enthusiasm and workflow complexity in ways that are genuinely hard to model in advance, especially in the first year or two of broad internal adoption, and that unpredictability lands squarely on finance teams that are used to forecasting software costs as a near-fixed line item.
That is a governance gap more than a technology gap. Most companies currently rely on department-level anecdotes and after-the-fact invoice review to understand their AI spend, rather than the kind of usage forecasting discipline applied to cloud infrastructure or headcount. ThredUp is a relatively small public company admitting this gap exists in public, at an industry event, rather than quietly fixing it internally first. That is itself useful information for any CTO or CFO who assumed their own forecasting processes were further along than ThredUp's turned out to be.
The financial backdrop this is happening against
This is not a cash-rich company treating AI spend as a rounding error. ThredUp's second-quarter 2026 results showed revenue up about 17 percent year over year to 90.8 million dollars, alongside a net loss that grew 15 percent to 5.9 million dollars and operating expenses up more than 17 percent to about 78.6 million dollars. Active buyers and order volume both grew in the quarter, which supports the top-line growth story, but the company is still operating at a loss while absorbing an AI cost line that its own CEO admits outran his forecast.
That combination, growth alongside a widening loss and an unpredictable new expense category, is exactly the setup where a board typically asks for tighter controls rather than more experimentation. Reinhart's public comfort with the uncertainty suggests either strong internal conviction that the productivity payoff is already showing up in ways the earnings release does not isolate, or a willingness to accept short-term investor scrutiny in exchange for not slowing down adoption while competitors are still deciding how aggressive to be.
What this means for your own AI budget
The lesson here is not that cost discipline does not matter. It is that cost forecasting for AI usage deserves the same seriousness companies already apply to cloud spend, and most organizations are behind on building that muscle even as adoption accelerates ahead of it. If your finance team is still treating AI spend as a single line item reviewed quarterly rather than a usage-driven cost center tracked the way AWS spend is tracked, you have the same gap Reinhart described, you simply have not said so publicly yet.
The more durable takeaway is that willingness to tolerate short-term cost unpredictability, in exchange for not slowing adoption, is becoming a real strategic choice that executive teams are making consciously rather than by default. Reinhart's version of that choice happens to come with a public admission that his own projections missed. Whether your organization makes the same bet or not, it is worth making the decision explicitly rather than discovering your own token bill's trajectory the same way he apparently did, after the fact.
