John Carmack

unclaimed

@id-aa-carmack · person

—pending

Subscribe · x402 · soonClaim this profile

win rate—0W · 0L

trades5published & scoreable

resolved0settled outcomes

avg P&L+22.3%per call

unclaimed public trackercorrection path availablescored on resolved calls only

Published claims

John Carmackx/ Jul 7

person

SNDKFlash memory gets a new multi-billion dollar market as AI inference moves to NAND, cutting memory costs 100x. Carmack's architecture case: inference is sequential reads, exactly what NAND does cheaply and well.

directional thesisLONGpublishedxnot a verified trade

SSNSNDKLONG

Flash memory gets a new multi-billion dollar market as AI inference moves to NAND, cutting memory costs 100x. Carmack's architecture case: inference is sequential reads, exactly what NAND does cheaply and well.

$1,648

+5.83%

SourceHBF (endurance and heat are still a problem) https://t.co/PFhJ25Omoi

“NAND flash is over 100 times cheaper per GB than HBM, so there should be opportunity there, even after giving a flash controller a 1024 bit interface with HBM bandwidth.”

evidence attachedpaste author pricesource span

John Carmackx/ Jul 6

person

SNDKNAND flash is taking over AI inference memory as the 100x cost advantage over HBM becomes impossible to ignore. Carmack's case that model weights only need sequential reads aligns exactly with NAND's strengths, and the market is already acting on it: SanDisk, the only pure-play NAND flash company, reported a 251% revenue surge and data centers are projected to surpass mobile as NAND's biggest market in 2026.

directional thesisLONGpublishedxnot a verified trade

SSNSNDKLONG

NAND flash is taking over AI inference memory as the 100x cost advantage over HBM becomes impossible to ignore. Carmack's case that model weights only need sequential reads aligns exactly with NAND's strengths, and the market is already acting on it: SanDisk, the only pure-play NAND flash company, reported a 251% revenue surge and data centers are projected to surpass mobile as NAND's biggest market in 2026.

$1,745

-0.01%

SourceMemory cost and capacity are significant issues for AI accelerators. Unlike game rende…

“NAND flash is over 100 times cheaper per GB than HBM, so there should be opportunity there, even after giving a flash controller a 1024 bit interface with HBM bandwidth.”

evidence attachedpaste author pricesource span

Agent-readable layer

{
  "version": "market-claims.v1",
  "claims": [
    {
      "id": "mcl_f2f0d0b0033ff13ce9",
       ,
       ,
       ,
       ,
       ,
       ,
       {
         ,
         ,
         ,
         ,
         ,
         ,
         
      },
       {
         ,
         ,
         ,
         ,
         ,
         
      },
       {
         ,
         ,
         ,
         ,
         ,
         
      },
       {
         ,
         ,
         ,
         ,
         
      }
    },
    {
       ,
       ,
       ,
       ,
       ,
       ,
       ,
       {
         ,
         ,
         ,
         ,
         ,
         ,
         
      },
       {
         ,
         ,
         ,
         ,
         ,
         
      },
       {
         ,
         ,
         ,
         ,
         ,
         
      },
       {
         ,
         ,
         ,
         ,
         
      }
    },
    {
       ,
       ,
       ,
       ,
       ,
       ,
       ,
       {
         ,
         ,
         ,
         ,
         ,
         ,
         
      },
       {
         ,
         ,
         ,
         ,
         ,
         
      },
       {
         ,
         ,
         ,
         ,
         ,
         
      },
       {
         ,
         ,
         ,
         ,
         
      }
    }
  ],
   
}

AISO tracks public statements, not verified portfolio positions. This is not financial advice.

John Carmack

SNDKFlash memory gets a new multi-billion dollar market as AI inference moves to NAND, cutting memory costs 100x. Carmack's architecture case: inference is sequential reads, exactly what NAND does cheaply and well.

SNDKSanDisk gets a second AI hardware act, shipping the first High Bandwidth Flash samples for AI inference in H2 2026, as a legendary engineer argues NAND's cost advantage over HBM is the biggest untapped opportunity in AI memory.

SKHXSK Hynix's HBM business faces a structural threat as Carmack lays out the case for replacing expensive HBM with NAND flash for AI inference, where SK Hynix holds the most to lose.

SNDKFlash memory gets a massive new use case as AI inference demand soars, because NAND is 100x cheaper per gigabyte than the expensive memory chips currently used in AI accelerators, and inference workloads have predictable enough memory patterns to make it work.

Reach & recent activity