Provenance Record

Well-structured News piece (689 words, within the 400-1200 News range). Clean Overview / What We Know / What We Don't Know / Analysis arc. Every specific traces to a cited source, and the article is careful to attribute the headline 10x figure to the company and to disclose that the underlying benchmark is a single third-party test by Gimlet Labs.

Read all five gzipped snapshots in sources/2026-06/d-matrix-moves-its-corsair-inference-accelerator-into-full-production-betting-on-chip-sram-beats-the-gpu-at-decode/. source-0 (PR Newswire, 200): CONFIRMS full production, the prefill/decode split, the verbatim 'GPUs dominate the compute-intensive prefill portion... while Corsair excels at the decode phase' quote, 'speed up AI model response time by more than 10x', the Gimlet Labs 24-second-to-under-two-seconds benchmark, Sid Sheth's 'Age of AI Inference' and 'Corsair takes off from where the GPU leaves off' quotes verbatim, the June 9 2026 date, TSMC + Alchip manufacturing, the Arista/Broadcom/Supermicro SquadRack ecosystem, and the hyperscalers/neoclouds/frontier-labs customer categories. source-2 (SiliconANGLE, 200): CONFIRMS 'embeds processing components into the memory', 'repurposed some of Corsair's SRAM circuits to carry out vector-matrix multiplications', the $275M Series C at a $2B valuation led by Bullhound/Triatomic/Temasek, '9,600 trillion calculations per second' on MXINT4, and the SquadRack 100-billion-parameter-in-SRAM claim. source-4 (The Register, 200): CONFIRMS the per-card 2GB SRAM @ 150 TB/s, 256GB LPDDR5 @ 400 GB/s, the 'blisteringly fast SRAM... much slower, but higher capacity LPDDR5' framing, and 2.4 PFLOPS (MXINT8) / 9.6 PFLOPS (MXINT4). source-1 and source-3 (both finance.yahoo.com) bot-blocked: their snapshots captured the consent.yahoo.com GDPR/privacy consent wall (Italian-locale 'Le tue scelte relative alla privacy', ~2.7KB, no article body), not the article. source-1's facts (TSMC, Alchip, Arista, Broadcom, Supermicro) are independently confirmed verbatim in the PR Newswire release (source-0). For source-3, as a documented last resort I performed a single live WebFetch of the cited allowlisted Yahoo URL, which confirmed verbatim: Microsoft participates via its M12 venture arm; Qatar Investment Authority and Singapore's Economic Development Board Investments joined as new participants; and 'tenfold better performance, triple the cost efficiency, and up to five times superior energy efficiency compared to GPU-based systems.'

No hallucinations detected. Every direct quote appears verbatim in its cited snapshot. No fabricated dual-card aggregate specs (no 6,400 mm2 / 4 GB / 300 TB/s / 4,800 TFLOPs) — the article correctly uses The Register's per-card figures. No CNBC-only specifics leaked in (no TSMC 6nm node detail, no ~$500M figure, no 'DRAM chokepoint' phrasing); the article instead uses TSMC N6 only where PR Newswire supports it. Entity names ('d-Matrix', 'Corsair', 'M12') reproduced verbatim.

APPROVE. Technically valid (signature and hash verified, all 18 checks pass). All five sources personally read; three load-bearing snapshots (PR Newswire, SiliconANGLE, The Register) fully corroborate every specific and quote. The two Yahoo consent-wall snapshots are bot-blocks, not content failures, and their facts are corroborated independently (source-1 via PR Newswire; source-3 via a documented last-resort live WebFetch of the cited allowlisted URL). Ready to publish as-is.