Key Takeaways

  • Higgsfield spends over $4 million per month on internal AI model inference across its 400-person team, averaging $10,000 per employee each month.
  • Non-technical team members are building custom software with foundation models; one creative spent over $30,000 in a single week on the Astra model to build internal tooling.
  • Mashrabov expects individual 10x engineers and creatives to consume between $50,000 and $100,000 per month in model inference as standard practice.
  • Model adoption has not cut operational headcount; Higgsfield expanded its legal group to over 10 people and customer success to over 40 people to manage release velocity.

The $10,000-Per-Seat Internal Compute Budget

Most startups treat software seats as fixed overhead: $20 for Slack, $40 for GitHub, maybe $200 for specialized tooling. Alex Mashrabov runs Higgsfield on an entirely different math. Across 400 employees, Higgsfield spends more than $4 million every month just on internal model usage.

“On average at Higgsfield, a person on the team spends over $10,000 a month on various models,” Mashrabov says. “So internal usage of models a month is over 4 million.”

This is not passive background compute for customer workloads. It is internal consumption by employees building, experimenting, and automating their own daily tasks. Mashrabov argues that restricting this spend to protect operating margins is backwards. If an employee can replace an entire vendor or build an unreleased product capability in days, capping their API access caps company velocity.

When Non-Coders Spend $30,000 in a Week

The biggest spenders are no longer just traditional backend engineers. Non-technical staff across design and marketing are building their own software tools directly against foundation models.

“What actually started to happen is the creative team started to do vibe coding,” Mashrabov explains. “This month I just caught a guy who spent over $30,000 in a week on the Astra model.” From March to June, the creative team shifted to Claude to build features that did not exist in production, while engineering staff migrated from Claude to Codex.

Mashrabov expects this trend to accelerate as models handle larger context windows and more complex reasoning. “I think it's going to keep growing and I do believe we are going to get to spend close to $50,000 and $100,000 a month for those who can call 10x engineers, 10x creatives.”

Headcount Expands Where Speed Creates Friction

The common narrative says heavy AI adoption eliminates operational staff. Higgsfield shows the opposite effect. When engineering and design output multiplies by ten, operations must expand to absorb the blast radius.

“Our legal team is over 10 people, our customer success team is over 40 people, all of them use AI heavily,” Mashrabov notes. “At these professions, I definitely can say that I don't see any elimination.” Shipping faster creates more contracts to review, more regulatory questions to resolve, and more customer edge cases to support. AI makes individual operators faster, but the surge in company output requires more human oversight, not less.

What to Do With This

Audit your team's API credit caps this week. If your top engineers and product designers are restricted by arbitrary $100 monthly limits on Anthropic or OpenAI, remove the ceiling for your top three performers for 30 days. Track what tools, prototypes, or automated workflows they build when compute is treated as cheap leverage rather than an expense line.