KARDASHEV Index
The leaderboard

Entity dossier · Nº 004

Extropic Corp.

Thermodynamic AI hardware Private extropic.ai ↗ Waltham, Massachusetts, United States

Strong efficiency claims paired with early hardware progress and pro-onshore manufacturing stance. Model synthesis · opinion

Unranked · insufficient data
Not ranked: insufficient data: energy undisclosed and 50% of the remaining weight scored (needs 60%, including compute or growth). We publish what was verified, but an entity is ranked only with at least 60% of the weight scored, more than 40% of the scored weight measured, and confidence above 20%. Ranking rules. Missing data is never filled with a default.

Index composition

65% measured · 35% opinion · method
Energy throughput Measured 30% —   no data
Compute capacity Measured 20% —   no data
Growth gradient Measured 15% —   no data
Frontier acceleration Opinion 15% 8.0 0.0 80%
Builder velocity Opinion 12% 6.0 new 61%
Permission to build Opinion 8% 6.0 new 44%

Δ = change vs run N-1 (2026-10-08). Index = weighted mean of the scored categories (weights renormalized over what was scored) · coverage 35% · confidence 23% · measured share 0% of the scored weight (opinion 100%). No “overall” is asked of the model. Scores are AI-assisted (disclosures).

Run history

Every published measurement, newest first. N is the run on the leaderboard now; failed attempts are not shown. Unranked runs show no Index; Δ and the Index trend compare ranked runs only.

RunPublishedIndexΔK-equivMethodView
N current 2026-10-09 Unranked — — pipeline-v2.4 · weights-v1 · rubrics-v1 viewing
N-1 2026-10-08 Unranked — — pipeline-v2.2 · weights-v1 · rubrics-v1 View run N-1

Measured — computed in code from reported figures

Joules · weight 30%

Energy throughput

—no data

No verified annual energy-consumption figure was found, so this category is unscored.

P = annual energy (J) / seconds per year → average watts. Score = clamp(2.5 × (log₁₀P − 7), 0, 10). Kardashev-equivalent K = (log₁₀P − 6) / 10.

Watts of thought · weight 20%

Compute capacity

—no data

No verified operating data-center capacity was found, so this category is unscored.

Score = clamp(2.5 × log₁₀(MW), 0, 10).

Slope · weight 15%

Growth gradient

—no data

Fewer than two comparable annual figures were found, so growth is unscored.

Each available sub-metric's compound annual growth rate (up to 3 years) is mapped piecewise-linearly through the anchors, then combined with weights capex 50%, revenue 25%, energy 25% (renormalized over what is available).

Opinion — model-judged against written rubrics, from verified quotes only

Opinion · Capability · weight 15%

Frontier acceleration

8.0/ 10

Multiple claims of 10,000x energy efficiency gains vs GPUs on key workloads [E189][E190][E192] indicate major cost-per-capability reductions.

Opinion Rubric · confidence 80%

Opinion · Shipping · weight 12%

Builder velocity

6.0/ 10

Announced first scaled Z1 thermodynamic chip and began bringing Z1 clusters online with performance demos [E194][E196].

Opinion Rubric · confidence 61%

Opinion · Policy · weight 8%

Permission to build

6.0/ 10

Supports CHIPS R&D funding to scale and onshore TSU fabrication at US foundries [E198].

  • Verified evidence
  • Planned CHIPS R&D funding to scale Extropic’s Thermodynamic Sampling Units for AI workloads and fabricate them at an American foundry. extropic.ai · fetched 2026-10-09 ↗
Opinion Rubric · confidence 44%

Sources we fetched and verified

  1. From One to One Billion: Torx, Thermalizers, and Z1 - Extropic extropic.ai/writing/from-one-to-one-billion · fetched 2026-10-09 · sha256 2e8a3554346f · primary
  2. Thermodynamic Computing: From Zero to One | Extropic extropic.ai/writing/thermodynamic-computing-from-zero-to-one · fetched 2026-10-09 · sha256 848aa06500da · primary
  3. Extropic Signs $75 Million Letter of Intent with U.S. Department of Commerce to Scale and Onshore... extropic.ai/writing/thermodynamic-computing-chips-in-america · fetched 2026-10-09 · sha256 ff26a4d4f9ce · primary
  4. Hardware | Extropic extropic.ai/hardware · fetched 2026-10-09 · sha256 13488af38856 · primary
  5. Z1T | Transformer-like models for Z1 extropic.ai/writing/z1t · fetched 2026-10-09 · sha256 0b83813e58e4 · primary
  6. First Sparks of Thermodynamic Recursive Intelligence | Extropic extropic.ai/writing/baby-thermo-rsi · fetched 2026-10-09 · sha256 0620d779bda8 · primary
  7. [2510.23972] An efficient probabilistic hardware architecture for diffusion-like models ar5iv.labs.arxiv.org/html/2510.23972 · fetched 2026-10-09 · sha256 28cdf9f69039
  8. Energy-efficient Codon Optimization on Thermodynamic Hardware arxiv.org/html/2606.17327 · fetched 2026-10-09 · sha256 9bd369330dbe
  9. What happens if AI suddenly needs 100x less energy? jamiegull.substack.com/p/what-happens-if-ai-suddenly-needs · fetched 2026-10-09 · sha256 83832bd815cc
  10. An efficient probabilistic hardware architecture for diffusion-like models | alphaXiv alphaxiv.org/abs/2510.23972 · fetched 2026-10-09 · sha256 8b7cda712bf6
  11. Extropic Aims to Disrupt the Data Center Bonanza | WIRED wired.com/story/extropic-aims-to-disrupt-the-data-center-bonanza · fetched 2026-10-09 · sha256 9899542c7c0b

1 other candidate link was rejected (dead, soft-404, blocked or unreadable) and are not shown.

Run N · 2026-10-09 11:10 PDT · 43 s · 8 stages · grok-4.3
StageStatusTimeWeb searches
resolvesucceeded0.0 s—
researchsucceeded9.6 s3
edgarskipped0.0 s—
fetchsucceeded6.6 s—
extractsucceeded11 s—
computesucceeded0.0 s—
judgesucceeded15 s—
aggregatesucceeded0.0 s—

pipeline-v2.4 · prompts-v2.3 · rubrics-v1 · weights-v1 · bundle 24911655c12e · code ce1fc2f

Last updated 2026-10-09 How the Index is computed