Project
SemiAnalysis chip benchmarking project running popular open-source models on current accelerators, published free at inferencex.com.
OpenAI took its Jalapeno chip from RTL to tapeout in nine months with its own AI models, against a norm its hardware head puts at 18 to 24 months. The count stops at first tapeout, before a B0 stepping reported to be up to 25% more efficient per watt.
Reality
Jalapeno's lead is measured against last generation and the volumes are tiny, but a first-pass ASIC clearing Nvidia, AMD and Google parts reprices the design barrier, not the supply chain.
Perspective Coverage
7 publishersReality
SemiAnalysis says its AgentX benchmark clocked Nvidia's Vera Rubin NVL72 at up to seven times Blackwell's token throughput per megawatt on pre-release software, and puts the profit gain at over two times per gigawatt.
Reality
The 1.7x to 3.6x latency range is set by the baseline systems, not the chip, and the report's own publication date is unsettled. Read it as direction, not evidence.
Perspective Coverage
9 publishersReality
OpenAI published a benchmark win on three large models, but the figure a GPU vendor has to price is the 10-gigawatt Broadcom program running through 2029, with racks targeted from the second half of 2026.
Reality
SemiAnalysis's AgentX replays recorded coding-agent sessions instead of fixed prompts. On that traffic, the reported Nvidia-AMD cost gap grows as the interactivity target rises.
Reality
SemiAnalysis has pushed fixed-length serving into maintenance mode. On replayed Claude Code traffic, NVIDIA's GB300 is credited with 15x Hopper, against 40x over H200 on the retired static test.
Reality
A presentation transcript published by InfoQ argues you cannot reason about inference economics without the chips, the power and the measured behaviour of the provider you rented from.
Reality