Why a $200 subscription can burn $100k of tokens: it's all about infra utilization
RexDouglass · x · 2026-11-22
Quoting developer Vlad Mihalcea's confusion — his $200/year Pro plan consumed 16B tokens, worth $100,000 at API rates — a cloud-infrastructure veteran explains the economics:
- Utilization is everything: token cost largely depends on infra utilization rate; higher utilization means lower per-token cost.
- Double-edged sword: running at 80% utilization makes tokens so cheap you could give them away, but one customer's product launch can spike load and crash the infra for everyone — losing your best customer.
- How labs cope: cheap tokens with rate/time limits control the cheap demand; as long as you don't cause spikes, labs profit even on $200 subs.
- KV cache and LLM-specific mechanisms matter, but the underlying hardware-allocation economics resemble traditional cloud servers.
- Practical tip: curious why some compute providers are so cheap? Ask about their concurrency guarantees.
More from Infra
- Qualcomm nears $2B quarterly auto chip revenue, tipped to lead by 2027 — BenBajarin · 2026-11-22
- Gavin Baker: $2T AI CapEx has produced only ~$120B in revenue so far — josh_bickett · 2026-11-22
- Inkling 2 Mini hits 200 tok/s on RTX 5090 via vLLM with native NVFP4 — VoidAsuka · 2026-11-22
- Redditor predicts cheap GPUs will return, 'only a matter of time' — ridhwanss · 2026-11-22
- Designer argues local LLMs are overrated; would buy a 5090 PC instead — AIandDesign · 2026-11-22
- An AI bubble fable: from free cash flow to private credit to annuity blowup — Svwang1 · 2026-11-22