Glossary

Energy out divided by energy in. Q=1 is scientific breakeven; a power plant needs Q well above 1. The single number for how close fusion is.

goal Q ≈ 10 — engineering gain a commercial power plant needs

How hot the plasma gets. Fusion needs roughly ten times the temperature of the sun's core to ignite and sustain.

How long a high-performance plasma is held. Steady-state operation is the bridge from physics demos to a real power plant.

Density × temperature × confinement time — the physics figure-of-merit for how close a plasma is to ignition, and the fairest single axis for comparing different confinement approaches. Breakeven needs ~10²¹; a D-T power plant ~5×10²¹. The magnetic-confinement record (JT-60U) has stood since the 1990s. Inertial-confinement NIF is transient and defined differently — tracked via energy gain Q instead.

goal ≈5×10²¹ — triple product a D-T power plant needs

Cumulative private capital into fusion companies — the clearest signal that investors think commercial fusion is within reach.

Peer-reviewed fusion works published per year (via OpenAlex), cross-checked against arXiv preprints. A proxy for how fast the field is moving — auto-collected, human-verified.

The largest number of physical qubits on a single processor announced to date. Raw count — says nothing about quality.

goal ≈1,000,000 — physical qubits est. for RSA-2048 (with error correction)

Error-corrected qubits that actually do useful work. The number that matters — and the one most often confused with physical qubits.

goal ≈1,000,000 — logical qubits for fault-tolerant cryptanalysis

How often a two-qubit operation succeeds. The threshold for practical error correction sits around 99.9% (“three nines”).

goal 99.99% — “four nines” — comfortable fault tolerance

Investment into quantum-technology start-ups per year (private + public), per the McKinsey Quantum Technology Monitor — the clearest read on how fast capital is entering the field.

Peer-reviewed quantum-computing works published per year (via OpenAlex). A proxy for how fast the field is moving — auto-collected, human-verified.

Grid-connected commercial small modular reactor units worldwide. Only Russia and China operate SMRs today; the West's first units target ~2030.

Peer-reviewed small-modular-reactor works published per year (via OpenAlex). A proxy for momentum — auto-collected, human-verified.

Disclosed cumulative private funding of leading humanoid-robot developers — the clearest signal of how seriously capital is backing the field. Bellwether: Figure AI, valued at $39B in 2025.

Humanoid robots shipped per year by leading developers — the clearest signal of real commercialization. Per Omdia's 2025 ranking, Chinese makers lead by ~30×: AgiBot shipped 5,168 vs ~150 each for Tesla, Figure and Agility.

Peer-reviewed humanoid-robotics works published per year (via OpenAlex). A proxy for momentum — auto-collected, human-verified.

Paid driverless rides per week — the clearest demand signal for robotaxis. Waymo is the volume leader in the US.

Cumulative miles driven fully autonomously with no human behind the wheel (Waymo) — the real-world experience base behind safety claims.

Peer-reviewed autonomous-driving works published per year (via OpenAlex). A proxy for momentum — auto-collected, human-verified.

Orbital launches per year by the leading provider — the clearest signal that reusability has turned spaceflight from rare to routine. SpaceX flew 165 orbital missions in 2025, more than the rest of the world combined.

Publicly cited price to put one kilogram into low Earth orbit on the cheapest available rocket — the headline number reusability is driving down. The Space Shuttle cost roughly $54,500/kg; a reused Falcon 9 is about $2,600/kg; Starship aims far lower.

goal ~$200/kg — Starship's stated target for cheap, frequent heavy-lift access.

The most launches flown by a single first-stage booster — the hardest proof that an orbital rocket is truly reusable, not just recoverable. SpaceX booster B1067 reached 32 flights in December 2025.

Peer-reviewed reusable-launch-vehicle works published per year (via OpenAlex). A proxy for momentum — auto-collected, human-verified.

Cumulative people who have received a company's implanted brain-computer interface — the clearest sign the field has moved from animals to humans. Synchron leads with 10 recipients; Neuralink has implanted 5.

Number of recording electrodes (channels) on a single implant — the rough ceiling on how much neural signal it can read. Neuralink's N1 and Precision's Layer 7 each carry 1,024, versus the 96 of the original BrainGate array.

Disclosed cumulative private funding of leading BCI developers — how seriously capital backs the field. Neuralink raised a $650M Series E in 2025 at a reported $9B valuation; ~$1.3B cumulative.

Peer-reviewed brain-computer-interface works published per year (via OpenAlex). A proxy for momentum — auto-collected, human-verified.

How far an aircraft has progressed through the FAA's 5-stage type-certification process — the real finish line before US passenger service. Joby completed Stage 4 in 2026; Stage 5 is the certificate itself. (China's EHang is already certified via a separate regulator for its autonomous model.)

goal Stage 5 (certified) — FAA type certificate — clearance for commercial passenger flights.

Cash, equivalents and committed liquidity on hand — the runway that decides whether a pre-revenue eVTOL maker can reach certification and production. Archer reported ~$1.7B liquidity (2025); Toyota committed $894M to Joby.

Maximum stated range per charge — the spec that splits the field's strategies: piloted long-range air taxis (Joby, Archer ~160 km) versus shorter-hop autonomous models (EHang, Volocopter ~30–35 km).

Peer-reviewed eVTOL / electric-aviation works published per year (via OpenAlex). A proxy for momentum — auto-collected, human-verified.

The raw compute (floating-point operations) used to train the largest known model — the clearest single number behind AI's scaling, tracked by Epoch AI. It has grown ~4–5× per year; GPT-4 (2023) was the first at 1e25 FLOP, and frontier models passed 1e26 FLOP in 2025.

Score on ARC-AGI-1 — puzzles easy for humans (~85%) but long resistant to AI. In Dec 2024 OpenAI's o3 reached 76–88%, the first AI to move beyond memorization on this test. ARC-AGI-2 is the harder successor, where frontier models still score low. (A third-party benchmark, not our score.)

goal ~85% (human) — Average human performance on ARC-AGI-1 — the bar AI crossed in late 2024.

Annual private AI investment by region (Stanford HAI AI Index). In 2024 US private AI investment hit $109.1B, nearly 12× China's $9.3B; global corporate AI investment reached $252.3B. The clearest signal of how hard capital is backing the field.

Peer-reviewed large-language-model / frontier-AI works published per year (via OpenAlex). A proxy for momentum — auto-collected, human-verified.

How far a commercial station has progressed toward crewed operations, on a 1–5 ladder: concept/agreement → development → ground qualification → in orbit → crewed continuous operations. Vast leads, targeting a 2026 launch of Haven-1.

goal Stage 5 (crewed ops) — Crewed, continuous commercial operations in orbit.

Pressurized habitable volume of the station (design figures for those not yet flown) — the clearest measure of how much room each contender offers. The ISS has ~916 m³ for reference.

How many crew the station is designed to host. Design ambitions range from compact (Vast Haven-1, 4) to large (Orbital Reef, ~10). The ISS hosts ~7.

Peer-reviewed space-station / microgravity works published per year (via OpenAlex). A proxy for momentum — auto-collected, human-verified.

The smallest process node a foundry is mass-producing. The "nm" name is a marketing label, not a physical dimension, so we read it alongside density — but it still marks the frontier. All three leaders reached 2nm-class volume in 2025 (TSMC N2, Samsung SF2, Intel 18A); TSMC leads on yield and volume. Next target: 1.4nm-class (~2028).

Share of the contract-chipmaking (foundry) market by revenue. TSMC holds roughly two-thirds — a dominance that compounds: more volume funds the next node, widening the lead. Samsung is a distant second; Intel Foundry is a newer external player.

Capital expenditure per year on fabs and equipment — the brutal cost of staying at the frontier. A single leading-edge fab runs ~$20–30B, and only a few firms can sustain it. TSMC alone guides ~$40B/yr.

Peer-reviewed semiconductor-lithography / process works published per year (via OpenAlex). A proxy for momentum — auto-collected, human-verified.

Gravimetric energy density of the best demonstrated cell (watt-hours per kilogram) — the frontier number for the whole field, because it sets range and weight. Conventional Li-ion cells top out near 250–300 Wh/kg; solid-state and lithium-metal designs target 400–500. Watch the fine print: lab/sample cells beat production cells, so each figure is labelled by stage.

goal 500 Wh/kg — automotive solid-state production target (~2030)

Minutes to charge a cell from 10% to 80% — lower is better. A headline promise of solid-state and lithium-metal cells, though the best conventional LFP fast-chargers already approach ~10 minutes, so this is read as a claim/target until verified in production packs.

goal 10 min — solid-state fast-charge target

Volume-weighted average price of a lithium-ion battery pack ($/kWh, BloombergNEF) — the cost floor that any next-gen chemistry must beat to reach mass market. Below ~$100/kWh, electric vehicles approach price parity with combustion. Solid-state cells start far above this line and must ride it down.

goal $100/kWh — price-parity threshold with combustion

Peer-reviewed solid-state-battery works published per year (via OpenAlex). A proxy for momentum — auto-collected, human-verified.

Electrical capacity of the largest single AI training cluster (megawatts) — the frontier number of the buildout era, since power, not chips, is now the binding constraint. Frontier sites jumped from tens of MW (2023) to gigawatt-class (2025). Announced future campuses reach for ~10 GW; figures here are labelled by stage (operational vs build-out).

goal 10 GW — Stargate full build-out (~2029)

Number of accelerators wired into the largest single training cluster. A coarse but vivid measure of scale: frontier clusters went from ~25k (2023) to 200k-class (2025), with million-GPU sites announced. Counts mix GPU generations (H100, GB200), so read alongside power.

goal 1M — first million-GPU cluster (~2027)

Capital expenditure per year on data centers and AI infrastructure by a single hyperscaler — the scale of the bet. The four largest US hyperscalers each now guide tens of billions a year, well over $300B combined, the fastest capex ramp in corporate history.

Peer-reviewed data-center / AI-infrastructure efficiency works published per year (via OpenAlex). A proxy for momentum — auto-collected, human-verified.

Annual CO₂ removal capacity of the largest operational direct-air-capture plant (tonnes per year). The frontier number for the field: plants jumped from Climeworks' Orca (~4,000 t/yr, 2021) to Mammoth (~36,000) to 1PointFive's Stratos (~500,000, 2025), with megatonne sites next. Counts nameplate capacity of operating plants, not announced ones.

goal 1 Mt/yr — first megatonne-scale plant (~2027)

Frontier cost to capture and store one tonne of CO₂ ($/t) — lower is better. Today's plants run roughly $400–1,000/t; the industry target is ~$100/t, the level at which large-scale removal becomes economic. Figures are operator estimates and fall as plants scale.

goal $100/t — economic-scale threshold

Cumulative tonnes of durable carbon removal contracted by a buyer (via CDR.fyi) — the demand signal funding the buildout. Microsoft dominates, having pre-purchased millions of tonnes. A market-size indicator, not a fixed-goal metric.

Peer-reviewed direct-air-capture works published per year (via OpenAlex). A proxy for momentum — auto-collected, human-verified.

Number of operational satellites in the largest broadband constellation. SpaceX's Starlink dwarfs all rivals, growing from 60 (2019) to several thousand — the physical backbone of orbital connectivity. Counts working satellites in orbit, not planned ones.

goal 12,000 — first-generation constellation complete

Active subscribers to satellite broadband, in millions. Starlink crossed from a pilot to several million customers in a few years — the revenue base that funds the constellation. A demand/scale indicator.

Satellites in orbit capable of connecting directly to ordinary phones, no tower — the new frontier. SpaceX has launched hundreds of direct-to-cell-capable Starlink v2 minis; AST SpaceMobile flies a few very large dedicated BlueBird satellites. Read alongside coverage and service status.

Peer-reviewed satellite-constellation / non-terrestrial-network works published per year (via OpenAlex). A proxy for momentum — auto-collected, human-verified.