The model is cheaper than the deployment
the model is cheaper than the deployment: llm inference eats 70 to 90 percent of enterprise ai operational cost, api prices dropped about 80 percent since early 2025, and deepseek charges 14 cents a million input tokens while a frontier flagship runs hundreds of times more. renting the model can cost 50 times less than renting the gpu.
the math is deciding which models get used.
Context
InfotechLead reported on Sep 3, 2026 that DeepSeek V4 Flash costs about 0.14 dollars per million input tokens against 5 dollars for OpenAI's GPT-5.5. Presenc AI says frontier token pricing fell about 50 times from 2023 to mid-2026.
Presenc AI also says self-hosted open-weight inference can reach about 0.05 dollars per million input tokens at sufficient throughput.
The 14 cents per million input tokens at DeepSeek matches the sources. Inference being 70 to 90 percent of enterprise AI operating cost, API prices falling about 80 percent since early 2025, and renting the model costing 50 times less than renting the GPU were not seen in the sources read, so unsupported here, not refuted. Presenc's 50 times is a longer window, from 2023.
Related work
Watch next
- Find a source for the inference share of AI operating cost.
Sources
- InfotechLead, Sep 3, 2026infotechlead.com
- Presenc AI, May 22, 2026presenc.ai
- BizScoreAI, May 25, 2026bizscoreai.com
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 2 October 2026 at 19:45 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →