AI Compute Budgets Rising
Box's CEO warned that companies will need to budget for steadily rising AI 'token' and compute costs as teams deploy more generative tools across workflows — a growing line-item for operations and procurement. That forecast signals higher IT and infrastructure spend for distributors adopting AI in sourcing, quoting, or support. (businessinsider.com)
Nvidia CEO Jensen Huang suggested on the All‑In Podcast that he would be “deeply alarmed” if a $500,000‑a‑year engineer did not consume roughly $250,000 in AI tokens, a comment companies and execs have used as the context for recent budget debates. (cnbc.com (cnbc.com)) Box CEO Aaron Levie responded on X that token and compute use will “monotonically” increase as workers who properly leverage AI consume more, explicitly warning this becomes a recurring line‑item for businesses. (africa.businessinsider.com (africa.businessinsider.com)) Levie told Fast Company he synthesized views from ~20 enterprise AI and IT leaders and flagged that adoption is moving beyond engineering into legal, sales and continuous agent workflows that will drive steady inference demand. (fastcompany.com (fastcompany.com)) Third‑party market research shows enterprise generative‑AI spend scaled rapidly to roughly $37 billion in 2025, up from about $11.5 billion in 2024, underscoring why CFOs and procurement teams are seeing new recurring compute line‑items. (menlovc.com (menlovc.com)) Academic tracking from Stanford’s AI Index highlights that inference (per‑use compute) costs have become a major ongoing budget pressure for firms running models in production, shifting spend profiles from one‑time training bills to steady per‑token expenses. (hai.stanford.edu (hai.stanford.edu)) Practically, LLM pricing studies show output (response) tokens can cost 3–10× more than input tokens depending on provider and model, meaning quoting, search, or support features that require long outputs will disproportionately drive bills. (cloudidr.com (cloudidr.com)) Enterprise software vendors are already adapting commercial models — many moving to hybrid seat-plus‑usage pricing — while Box has been rolling agent and inference features (Box Automate, AI Studio) into its product set to capture and bill for those workloads. (data-mania.com (data-mania.com); techcrunch.com (techcrunch.com))
