Earth sits at K ≈ 0.73 — roughly 19 TW of harnessed power. Type I needs ~500× more. This index scores companies on one thing: how much they bend a civilizational cost curve, times the scale they do it at. Not vibes — a live formula you can trace and tune.
Twenty companies, scored across five pillars with current 2026 data — launch $/kg, fusion milestones, GPU cost-per-generation, sequencing curves. Tap any row for its radar, the note, and the full formula trace.
Only credible path to full launch reuse. Single-use Starship already ~$820–980/kg; sub-$200/kg target redraws the frontier curve. Near-unique → highest counterfactual.
Every frontier chip on Earth depends on it. The most advanced manufacturing humanity has. An effectively irreplaceable single point — counterfactual maxed.
SPARC ~75% built, Q>1 net-gain targeted 2027. If it lands, it's the Kitty Hawk of energy. High variance / pre-revenue — score reflects potential, not delivery.
Defines the compute curve — Rubin claims ~10× lower inference cost per generation, ~80% datacenter GPU share. Mild drag from allocation gatekeeping and pricing power.
Hit 150M°C with D-T fuel in Polaris (Feb 2026); Microsoft PPA signed. Direct-electricity approach. Pre-revenue, high uniqueness.
Rare three-pillar play: grid-scale storage (Megapack), Optimus + FSD automation, Dojo compute. Scaled the EV/storage curve more than invented it.
Gemini at the frontier; AlphaFold rewrote structural biology's cost curve. Drag from big-co inertia and an ad business that isn't accelerationist.
Frontier models, but high substitutability — several labs at the same frontier means low counterfactual. Drag from closed posture and moat-seeking.
Frontier models with a safety-weighted posture. Same substitutability discount as peers; lower drag. (Scored neutrally — this index is built by Claude.)
General-purpose humanoid = a labor multiplier across all of Atoms. Several humanoid players now (Tesla, 1X, Unitree) so not unique. Unproven at scale.
Natrium sodium fast reactor + integrated storage. Fission is proven physics; multiple SMR competitors temper the counterfactual.
Wafer-scale compute bending the inference curve from a different angle than NVIDIA. Smaller scale, more substitutable.
Drove $/genome down faster than Moore's law for two decades. Drag from historical near-monopoly pricing; newer rivals now bend the curve harder.
Iron-air long-duration storage — the missing piece that turns intermittent renewables into baseload. Storage is the quiet enabler of the energy curve.
Neutron reusable + high Electron cadence + Varda manufacturing partner. Real frontier work, but in SpaceX's shadow on the cost curve.
Robotics pioneer (Atlas), but slow commercialization keeps realized acceleration below the technical frontier.
mRNA as a programmable-biology platform — design-to-candidate in weeks. Broad therapeutic surface well beyond vaccines.
Huge US-based CdTe solar manufacturing, but solar is commoditized and China-dominated — First Solar rides the curve more than bends it.
Cellular reprogramming for longevity — extends the productive human substrate. Frontier, few serious players, entirely pre-revenue.
Synbio cell-programming platform. Strong primitive, weak commercial execution and SPAC overhang create drag.
The whole point of an index is that the judgment is transparent. Load a company, drag the pillars, the counterfactual and the drag — watch its KA score and its rank against the field recompute live. This is the formula, running.
Acceleration is a derivative. A company riding a cost curve scores low; one bending its slope downward at scale scores high. Energy and Intelligence carry the most weight — both are master multipliers on everything else.
The peak term rewards specialists who max one curve at order-of-magnitude scale, so a pure fusion or frontier bet isn't buried under a broad-but-shallow conglomerate.