▼Serving Software Decides Accelerator Economics

Score history

Daily conviction score, 0 to 100. Higher means the thesis is more strongly corroborated.

Now 36 · -2 since Aug 27 · ranged 36 to 38

Showing the last few days. Unlock full score history.

Why the conviction moved

• Aug 26
  Context

  OpenAI's Jalapeño claims are silicon-level efficiency ratios measured on three open models, with no disclosure of scheduling, batching or KV-cache handling in the serving stack used for either side. The comparison leaves the thesis untested: per-watt specs are exactly the axis on which challengers have historically led while still losing on total cost of serving.

• Aug 25
  Strengthened +8

  SemiAnalysis benchmarked real coding-agent traffic and found Nvidia roughly five times cheaper per unit of work than AMD, attributing most of the gap to serving software and concluding Nvidia would remain cheaper per token even if the competing hardware were free. That counterfactual is the thesis stated almost literally: silicon price cannot close a runtime-and-scheduling gap.

Source trail

• Context · August 26, 2026

  OpenAI Publishes First Jalapeño Benchmarks, Claiming Up to 1.9 Times Nvidia's Throughput Per Kilowatt

  OpenAI's Jalapeño claims are silicon-level efficiency ratios measured on three open models, with no disclosure of scheduling, batching or KV-cache handling in the serving stack used for either side. The comparison leaves the thesis untested: per-watt specs are exactly the axis on which challengers have historically led while still losing on total cost of serving.

  OpenAI

• Supporting · August 25, 2026

  SemiAnalysis Benchmark of Real Coding-Agent Traffic Puts Nvidia About Five Times Ahead of AMD on Cost

  SemiAnalysis benchmarked real coding-agent traffic and found Nvidia roughly five times cheaper per unit of work than AMD, attributing most of the gap to serving software and concluding Nvidia would remain cheaper per token even if the competing hardware were free. That counterfactual is the thesis stated almost literally: silicon price cannot close a runtime-and-scheduling gap.

  AI Post (Telegram)