How the newsroom publishes
A short note on how this newsroom works — milestone notes are auto-drafted from verified data, explainers and analysis are written here as Markdown, and every figure links to a primary source.
The publishing source of truth behind the Frontier Milestones tracker. Milestone notes report the moment a figure moves; explainers unpack the methods; analysis is our labelled read. Every figure links to a primary source.
RSSA short note on how this newsroom works — milestone notes are auto-drafted from verified data, explainers and analysis are written here as Markdown, and every figure links to a primary source.
Xcimer Energy — Phoenix, a prototype for industrial-scale laser fusion, begins operations — demonstrating end-to-end KrF excimer amplification plus Stimulated Brillouin Scattering (SBS) pulse compression at >1 kJ through a 38 m gas optic (record energy and scale for SBS). First step toward the Vulcan laser and the Athena power plant; the gas-laser path targets lower cost than solid-state drivers.
Xcimer / US DOE — The DOE accepted Xcimer's 724-page preconceptual design and technology roadmap for Athena, its laser-fusion power-plant architecture — billed as the industry's most comprehensive government review of a privately-developed fusion plant, under the DOE Fusion Milestone Development Program.
Germany / EU — Germany joins the EU 'Innovative Core Technologies' IPCEI, focused exclusively on fusion (not fission) — roughly €2.4B this legislative period, part of >€2B pledged for fusion R&D and pilots through 2029. National projects start 2027; the government's stated aim is to host the world's first fusion power plant.
Helion Energy — Thrive Capital-led $465M Series G nearly triples Helion's valuation to $15.5B and brings total raised to ~$1.5B, funding manufacturing scale-up toward its 2028 power deal with Microsoft.
Commonwealth Fusion Systems — CFS published five peer-reviewed papers (58 authors, a Journal of Plasma Physics special issue) laying the physics basis for ARC — its first commercial plant, designed to deliver 400 MW net to the grid in the early 2030s — building on lessons from SPARC.
Avalanche Energy — Avalanche's compact Orbitron device "Jyn" measured apparent ion temperatures above 1 keV (~11M°C, hotter than the Sun's core) — the community's take-notice threshold, hit in a desktop-sized machine.
Pacific Fusion — Pacific Fusion's pulser-module prototype (9 stages, 90 bricks) delivered 440 GW in an 80 ns burst — the highest-power single-step pulsed-power driver ever demonstrated — validating its trigger sync and unlocking a tranche of its >$1B Series A; demo-facility construction starts this summer.
Microsoft / Quantinuum — Peer-reviewed in Nature: error correction (carbon & tesseract codes) cut logical error rates up to 800x below the underlying physical qubits on Quantinuum trapped-ion hardware - the largest physical-to-logical gap yet independently validated, with repeated mid-circuit correction across up to 12 logical qubits.
Atom Computing — Atom Computing demonstrated sustained multi-round quantum error correction with a toric code on neutral atoms — logical error rates falling as the system scales up (sub-threshold), a first for the neutral-atom platform.
Quantinuum — Quantinuum priced an upsized IPO at $60/share, raising $1.68B on Nasdaq (QNT) at a ~$15.7B market value — the quantum industry's first mega-IPO. Honeywell retains ~48% voting power.
Antares / Idaho National Lab — Antares' Mark 0 test reactor at Idaho National Laboratory achieved first criticality — a self-sustaining nuclear reaction and the first new US advanced-reactor design to cross the threshold (it is not yet generating power).
NEURA Robotics — Germany's NEURA announced a Series C of up to $1.4B at a $7B valuation (Nvidia, Amazon, Qualcomm, Tether, Bosch, EIB…) — billed as the largest-ever raise for a full-stack robotics company; the full amount is milestone-contingent. 4NE-1 humanoid (~€98k) ships at scale from late 2026.
SpaceX — SpaceX raised $75B in its IPO (555.6M shares at $135; ~$1.75T valuation), trading on Nasdaq as SPCX from 12 Jun 2026 — the largest IPO in history, dwarfing Saudi Aramco's $29.4B (2019); the book ran >2× oversubscribed (~$150B in orders).
Axiom Space — Axiom closed $525M+ (Feb $350M led by Type One Ventures & QIA, plus a $175M June extension joined by MUFG) to accelerate Axiom Station, spacesuits and human spaceflight — the largest station-sector raise amid NASA's CLD strategy rework.
Anthropic — Anthropic released Claude Fable 5 — a Mythos-class model exceeding any it had made generally available — gated so ~5% of sensitive (e.g. cyber) sessions get a conservatively-tuned model, while the unrestricted Mythos 5 went only to vetted cyberdefenders via Project Glasswing with the US government. Days later the US Commerce Department export-controlled both models, barring all foreign-national access; unable to enforce that selectively in real time, Anthropic shut Fable 5 and Mythos 5 off worldwide (its other models unaffected) — the first time a deployed frontier AI model was export-controlled like a strategic technology.
A $9B valuation for Neuralink prices in a future where brain implants are routine — but today's devices help a few dozen people in trials. Our read on the gap between capital and clinical reality. (Our opinion, not investment advice.)
Demos and valuations are easy to inflate; people living with an implant are not. The count of humans implanted is the hardest signal that a BCI has crossed from lab to clinic — and Synchron, not the loudest name, leads it.
BCIs trade off signal quality against surgical risk. Penetrating arrays (Neuralink) read the most; surface arrays (Precision) sit on the brain; vascular devices (Synchron) avoid open-brain surgery entirely. The tracker compares all three.
165 launches and a 32-flight booster put SpaceX years ahead, but Blue Origin's New Glenn reuse and Rocket Lab's Neutron are finally real. Our read on how durable the lead is. (Our opinion, not investment advice.)
ITER / General Atomics — The 13-tesla, ~1,000-tonne Central Solenoid — the 'beating heart' built by General Atomics to drive ITER's plasma current — is complete and delivered to the site. A key assembly milestone toward first plasma (2034); the full pulsed-magnet system was ~15 years in the making.
Focused Energy — German laser-fusion firm Focused Energy closed an oversubscribed $240M Series A led by utility RWE — the industry's largest Series A — to build its Lighthouse demonstrator at a retired German fission plant (total funding ~$500M incl. grants).
IBM / US Commerce — IBM is spinning off Anderon, a $2B (≈$1B CHIPS Act + $1B IBM) 300mm superconducting-qubit wafer fab in Albany, NY — open to other quantum vendors as a neutral 'TSMC for quantum.'
IonQ — IonQ sold its first 6th-gen, chip-based 256-qubit system (to the University of Cambridge) and posted record Q1 revenue of $64.7M (+755% YoY); its roadmap targets 10,000 networked qubits.
OPG / GE Hitachi — OPG set the ~953-tonne basemat foundation module 35 m below ground for the first of four BWRX-300 units at Darlington — North America's (and the G7's) first SMR build, targeting grid connection by 2030.
1X — 1X began full-scale NEO production at a 58,000 sq-ft, vertically-integrated plant in Hayward, California — ~10,000 units/yr capacity (its first-year run sold out in 5 days), targeting 100,000 units by 2027.
Figure AI — Figure ran its humanoids 200 hours nonstop, sorting 249,558 packages with zero hardware failures and no teleoperation — driven end-to-end by its Helix neural network, a durability/autonomy milestone.
Waymo — Waymo suspended all freeway rides (San Francisco, LA, Phoenix, Miami) to improve construction-zone handling, after recalling 3,791 vehicles over flooded-roadway incidents — a notable safety pullback on its hardest driving domain.
Blue Origin — A New Glenn first stage exploded during a static-fire test at LC-36, Cape Canaveral (28 May 2026) — destroying the booster and damaging Blue Origin's only New Glenn pad. Reported as the most powerful rocket explosion since the Soviet N1 (1969); no injuries. Blue Origin targets return to flight before end-2026, accelerating a vertical-integration redesign.
SpaceX — SpaceX flew Starship V3 for the first time (Flight 12, 22 May 2026) — its most powerful version, built for high-rate Starlink launches and future Moon missions. The suborbital test deployed 20 mock + 2 real Starlinks and was called a success despite engine glitches and a missed booster splashdown; full orbital flight is still to come.
Dollars per kilogram to orbit is the single figure that decides what becomes possible in space — and it has fallen ~20× since the Shuttle. We explain the number and why headline prices need caveats.
Landing a booster is the photo; flying the same booster again — and again — is the economics. We explain why reflights, not landings, are the metric that lowers the cost of space.
eVTOL leaders are pre-revenue and burning cash toward certification. Joby and Archer have raised billions and look funded; many European rivals already went bankrupt. Our read on who has the runway. (Our opinion, not investment advice.)
Frontier training compute has grown ~4–5× a year and is the clearest driver of AI's recent leaps. It is a hard, auditable number — but it's an input, not a measure of intelligence.
Air-taxi makers are not building the same vehicle. Joby and Archer bet on piloted, ~100-mile aircraft; EHang and Volocopter fly shorter, often autonomous hops. The range spec reveals the strategy.
There is no agreed test for general intelligence, so a single "AGI %" would be our opinion dressed as data. Instead we track objective, third-party numbers: training compute, public benchmark scores, and investment.
An eVTOL can fly thousands of test flights and still not be allowed to carry a paying passenger. The gate is FAA type certification — a five-stage process — and it, not flashy demos, is the real finish line.
The White House's FY27 request for Fusion Energy Sciences is $755M — $50M below FY26 and well short of the $1.11B authorized by the CHIPS and Science Act — even as public-private partnership funding rises to $135M and a new Office of Fusion is proposed. A contrast with Europe, where Germany just joined a multi-billion-euro fusion IPCEI.
Zap Energy — Adds near-term ~50 MW modular fission reactors alongside its sheared-flow Z-pinch fusion, betting on shared materials, liquid-metal and power-conversion tech. Century platform hit a record 1.6 GPa plasma pressure and FuZE-A came online. Zabrina Johal (ex-General Atomics, US Navy nuclear) named CEO; cofounder Benj Conway → President.
Tokamak Energy / UKIFS — Tokamak Energy named Magnet Systems Partner for the UK government's STEP programme — the first-of-a-kind plant at West Burton (UKAEA / UK Industrial Fusion Solutions). A £70M (~$95M) contract (2026–2029) leading eight HTS magnet work packages via its TE Magnetics division; its ST40 spherical tokamak serves as a high-field testbed.
ARPA-E (US DOE) — ARPA-E committed $135M to fusion over 18 months — more than its entire prior 12 years of fusion funding combined — launching CHADWICK (first-wall materials for a 40-year plant life) and GAMOW (market-aligned fusion) programs.
First Light Fusion — First Light closed a £25m first tranche (Starmaker One + UKAEA) and shipped its first VIPER III velocity amplifiers to Texas A&M — a pivot to selling its amplification tech while pursuing its FLARE inertial-fusion concept (target gain ~1,000).
QuEra / Harvard / MIT — QuEra, Harvard and MIT showed qLDPC codes encoding 1,156 logical qubits into 2,304 physical (≈2:1, >50% rate), simulated into the "teraquop" regime (~1 error per trillion ops) — versus the hundreds-to-one ratio typical today.
Kairos Power — Kairos broke ground on Hermes 2 in Oak Ridge — the first power-producing Gen IV reactor with an NRC construction permit. The molten-salt-cooled plant will feed up to 50 MW to TVA, powering Google data centers.
China (MIIT) / Baidu — After 100+ Baidu Apollo Go robotaxis failed simultaneously in Wuhan (a cloud outage stranded riders up to two hours), China suspended all new autonomous-driving permits on 29 Apr 2026 — the first nationwide robotaxi licensing freeze.
Blue Origin — New Glenn reflew a recovered first stage for the first time, ending Falcon's solo run on orbital reuse.
CU Anschutz · Caltech · USC — Surgeons at CU Anschutz/UCHealth implanted a BCI into a paralyzed patient's higher-functioning cortex — not the usual motor cortex — a world first, aiming for more natural and complete sensory-and-motor control. Part of a Caltech/USC consortium using a Blackrock Neurotech array.
Joby Aviation — Joby flew between JFK and Manhattan's heliport network, demonstrating the urban air-taxi use case.
DeepSeek / Huawei — DeepSeek's 1.6T-parameter V4 runs on Huawei Ascend (950PR), and a Huawei-led team completed full-parameter post-training on ~1,000 Ascend 910Cs — a compute-sovereignty landmark. Pre-training hardware remains undisclosed, so "trained without Nvidia" is NOT established.
Waymo leads on paid driverless rides and miles; Tesla is betting on a camera-only, mass-market path; Cruise exited in 2024. We weigh the very different strategies. (Our opinion, not investment advice.)
A $39B valuation for Figure and billions across the field price in success that hardware and autonomy haven't yet delivered. Our read on whether capital is ahead of capability. (Our opinion, not investment advice.)
Cumulative rider-only miles are the experience base behind safety claims: more real, driverless miles means more rare situations seen and handled. It's the denominator under every crash-rate comparison.
“Deployed at BMW/Amazon/Mercedes” usually means a supervised pilot, not autonomous production work. We explain the ladder from demo to paid pilot to true at-scale deployment.
“Self-driving” spans everything from lane-keeping (Level 2) to no-human-needed robotaxis (Level 4). Waymo runs Level 4 in set areas; most consumer “autopilot” is Level 2. The gap is the whole game.
Not every robot arm is a humanoid. The bet is on a general-purpose, roughly human-shaped, two-legged machine that can slot into spaces and tools built for people — versus cheaper, task-specific automation.
Big tech power deals and listed names (NuScale, Oklo) have lifted SMR sentiment, but Western first-of-a-kind units don't reach the grid until ~2030. We weigh the gap between order books and operating reactors. (Our opinion, not investment advice.)
Cumulative private funding has crossed ~$9.8B across 53 companies (FIA 2025). We look at what the money is — and isn't — telling investors about timelines. (Our opinion, not investment advice.)
TerraPower / NRC — The NRC issued a construction permit for TerraPower's Natrium plant (Kemmerer, Wyoming) — the first US commercial advanced-reactor approval in ~a decade and the first non-light-water reactor approval in over 40 years. 345 MWe sodium-cooled fast reactor with molten-salt storage boosting to 500 MW; completion targeted 2030.
US NRC — After ~6 years, the NRC finalized Part 53 (effective 29 Apr 2026) — a risk-informed, technology-inclusive licensing framework targeting design approvals in 18 months or less at roughly half the cost. A structural unlock for the entire US advanced-reactor field, independent of any single project.
Science Corporation — Science closed an oversubscribed $230M Series C (total ~$490M) to commercialize PRIMA — a retinal BCI that restored form vision in dry-AMD patients (NEJM-published). A European launch later this year would make it the first BCI vision product on the market.
Neuracle / NMPA (China) — China's NMPA cleared Neuracle's coin-sized NEO implant (developed with Tsinghua) for commercial sale to restore hand function after spinal-cord injury — the first invasive-class BCI approved for market anywhere, ahead of Neuralink.
Joby Aviation — Joby cleared the FAA's airworthiness conformity review — the most advanced US eVTOL certification position.
ARC Prize — The first fully interactive ARC benchmark: hand-built game environments with no instructions — agents must discover the rules. At launch every frontier model scored <1% (best 0.37%) while humans solve them all; $2M+ prize pool, results Dec 2026.
Headlines feature Western startups, but the only commercial SMRs operating today are Russia's floating Akademik Lomonosov (2020) and China's HTR-PM (2023). China's Linglong One aims to be the first land-based commercial unit in 2026.
Records are falling fast — gain 4 at NIF, 1,000 s-plus plasmas in China and France — but engineering breakeven and a grid-connected plant are still distinct, harder steps. Our read on the gap. (Our opinion, not investment advice.)
SMRs are smaller (under ~300 MWe) reactors built in factories and shipped as units, betting that repetition and standardization beat the cost overruns of giant one-off plants.
NIF crushes a fuel pellet with lasers; EAST, JET and ITER hold plasma in magnetic fields. They post different numbers (gain vs duration) because they're different machines chasing the same goal.
Logical-qubit records are climbing fast, but breaking RSA-2048 needs thousands of logical (millions of physical) qubits with low error rates held for hours. Our read: real, but not imminent. (Our opinion, not investment advice.)
Hot enough to fuse, long enough to be a power source: EAST's 160M°C and WEST's 1,337 s tackle the two halves of the same problem. Neither alone makes a reactor.
TSMC holds ~two-thirds of the market and the yield lead; Intel is betting its comeback on 18A; Samsung trails on yield. Our read on a race where one player's dominance keeps compounding. (Our opinion, not investment advice.)
NIF's climb to 8.6 MJ and target gain above 4 came less from more laser power than from better fuel capsules. We explain how high-density carbon (diamond) capsules and hohlraum quality turned a one-off ignition into a repeatable record.
Helion Energy — Polaris reaches 150M°C — first privately built machine to do D-T fusion, beating Helion's own 100M°C (Trenta).
Proxima Fusion / RWE / Bavaria / Max Planck IPP — Proxima Fusion, RWE, the Free State of Bavaria and Max Planck IPP signed an MOU to build Europe's first commercial stellarator fusion plant. It starts with 'Alpha', a €2B (~$2.3B) demonstration stellarator in Garching aiming to be the first stellarator to reach net energy gain (Q>1) in the 2030s; a commercial 'Stellaris' plant would follow at RWE's decommissioning Gundremmingen nuclear site. IPP leads physics, Proxima engineering, RWE plant construction.
DTT / ENEA — The first 1 MW, 170 GHz gyrotron (THALES TH1509) for Italy's DTT passed acceptance at the Swiss Plasma Centre — 1 MW for 100 s, the first European demonstration of an industrial MW-class 170 GHz tube. A key step for DTT's ECRH heating; 15 more units to follow.
Apptronik — A $520M Series A extension (backers include Google, Mercedes-Benz and the Qatar Investment Authority) values Apptronik at $5B and lifts total funding to ~$1B, to scale Apollo production toward ~$80k/unit at volume from 2027.
Agility Robotics / Toyota — Toyota Motor Manufacturing Canada contracted seven Agility Digit humanoids (starting with three) for its Woodstock RAV4 plant under a robots-as-a-service deal (~$30/hr per robot) — a pilot-to-commercial step in auto manufacturing.
Most station contenders are private — Vast is self-funded, Axiom and Blue Origin unlisted — leaving Voyager Technologies (NYSE: VOYG) as the main public play. Our read on a capital-hungry race with one listed runner. (Our opinion, not investment advice.)
A single advanced fab costs $20–30B and depends on EUV machines only ASML makes. That brutal economics narrows the frontier to a handful of firms — and makes leading-edge capacity one of the world's true chokepoints.
99.9% vs 99.99% looks like a rounding error but means 10× fewer mistakes. Error correction only pays off above a threshold near “three nines”; crossing “four nines” buys real headroom.
NIF passed scientific breakeven in 2022 — but the lasers drew ~300 MJ from the wall to deliver 2 MJ to the target. Engineering breakeven, where the whole plant nets energy, is the bar that matters for the grid.
Station spec sheets range from Vast's 45 m³ Haven-1 to Blue Origin's 830 m³ Orbital Reef. Almost all of these are design figures, not flown hardware — here is how to read them against the ISS's ~916 m³.
No part of a "2nm" chip is 2 nanometers. Node names are marketing labels, not physical measurements — so we read them alongside transistor density and who is actually shipping in volume.
For decades, adding qubits added errors faster than you could correct them. “Below threshold” is the turning point where making the code bigger makes it more reliable — the precondition for a useful machine.
Q is fusion energy out divided by energy in. Q=1 is scientific breakeven; a power plant needs Q well into the tens. We explain why NIF's “gain 4” and a plant's Q aren't the same number.
NASA plans to deorbit the ISS around 2030, and a field of private stations is racing to be ready before the lights go out in low Earth orbit. We track each contender's development stage — not their press releases.
A 6,100-qubit array and 96 logical qubits sound contradictory until you know the difference. Physical qubits are the raw, noisy hardware; logical qubits are many physical ones error-corrected into one reliable unit. The second number is the one that matters.
EAST (ASIPP) / CAS — EAST ran stable plasmas at 1.3–1.65× the Greenwald density limit (vs the usual 0.8–1.0) using ECRH-assisted ohmic start-up — clearing a decades-old barrier in magnetic confinement. Higher density lifts the fusion rate and the triple product toward breakeven. Published in Science Advances (Jan 2026); ASIPP, Huazhong University and Aix-Marseille.
QuEra / Harvard — Fault-tolerant architecture running algorithms on up to 96 logical qubits.
Tesla — Tesla began public Austin robotaxi rides with no human safety monitor in the car (22 Jan 2026); by June it covered the entire Austin metro (~245 sq mi, ~13–20 Model Ys), though remote operators can still intervene.
Waymo — Roughly 1M rides a month across 10+ US cities, widening its lead.
SpaceX — Booster B1067 reached 32 launches and landings, far past Falcon 9's original 10-flight design life.
TSMC / Samsung / Intel — All three leaders entered 2nm-class volume in 2025 (TSMC N2 in Q4); demand outstrips supply, with long foundry waitlists.
Quantinuum — Quantinuum launched Helios: 98 trapped-ion (barium-137) physical qubits with 99.92% two-qubit-gate fidelity and all-to-all (QCCD) connectivity, running up to 48 error-corrected logical qubits — its most accurate commercial system.
Blue Origin — New Glenn's first stage landed at sea — the first reusable orbital rocket besides Falcon.
Vast — Vast finished structural qualification of Haven-1 and flew a pathfinder — clearing the path to launch.
Intel (18A) — Intel's 18A brought PowerVia backside power to a volume node — a structural change rivals are still adopting.
Helion Energy — Chelan County permits Helion's Orion plant in Malaga, WA — up to 50 MW to Microsoft, targeted by 2028.
IonQ — First two-qubit gate fidelity past 99.99% — comfortable headroom over the EC threshold.
1X / Figure — 1X opened NEO home-robot preorders; Figure unveiled the Figure 03 home humanoid.
Caltech — Caltech trapped 6,100 cesium atoms in optical tweezers — roughly 10× prior arrays — holding superposition ~13 s with 99.98% single-qubit accuracy and shuttling atoms without decohering, a key enabler for error correction. Published in Nature.
Oklo / Idaho National Lab — Oklo broke ground (22 Sep 2025) on Aurora-INL — its first Aurora Powerhouse, a 75-MWe liquid-metal-cooled, metal-fueled fast reactor at Idaho National Laboratory (building on EBR-II heritage) under DOE's Reactor Pilot Program, with Kiewit as lead constructor.
Figure AI — $1B+ Series C — a 15× jump from its $2.6B valuation in early 2024.
Joby Aviation — Joby flew a piloted eVTOL air taxi from Marina to Monterey — the first such flight between two public airports.
Waymo — Crossed 100M rider-only miles, doubling from 50M in seven months.
US private AI investment hit $109B in 2024 — then 2025's efficiency shock (DeepSeek) made the bubble question sharper, not simpler. Our read on whether capital is ahead of capability. (Our opinion, not investment advice.)
ITER / Westinghouse — A $180M contract for Westinghouse to complete and weld ITER's vacuum vessel — the hermetic double-walled steel chamber that houses the plasma — joining its nine sectors into a single ring, the most intensive stage of ITER assembly.
Tesla — Tesla began a limited robotaxi pilot in Austin — early and small-scale.
Neuralink — A $650M Series E valued Neuralink at a reported $9B, with five patients implanted.
GE Vernova Hitachi / OPG — BWRX-300 construction approved at OPG's Darlington site in Ontario, Canada.
Synchron / Apple — Synchron users gained native thought-control of iPhone, iPad and Vision Pro via Apple's BCI HID profile.
Paradromics — Paradromics recorded from its Connexus BCI in a human during a 20-minute procedure, widening the race.
Anthropic (Claude Opus 4) — Anthropic's Claude Opus 4 launched with extended thinking and sustained autonomous coding over long tasks — part of a 2025 shift where reasoning/agentic models, not raw scale alone, drove the frontier.
NIF / LLNL — Eighth ignition shot — more than quadruple the laser energy delivered.
Precision Neuroscience — The Layer 7 cortical interface (1,024 electrodes) got 510(k) clearance — a regulatory first for the new wave.
ARC Prize — A harder successor — still easy for humans, hard for AI — resetting the abstraction frontier as v1 saturated.
WEST (CEA) — WEST (CEA, France) held a hydrogen plasma for 1,337 s — about 22 minutes — at 50M°C with a tungsten wall and 2 MW of heating, surpassing EAST's 1,066 s by ~25% for the longest sustained fusion plasma on record.
EAST (ASIPP) — Steady-state operation past 1,000 s — the bridge to a power plant.
Apptronik / Mercedes-Benz — Apollo piloted for parts delivery and inspection at Mercedes-Benz.
Volocopter — After insolvency and a Wanfeng buyout, Volocopter resumed its EASA path with 2,000+ test flights logged.
frontier labs — Largest models crossed 1e26 FLOP — a 10× jump over GPT-4, with compute still growing ~4–5× per year.
DeepSeek (R1) — DeepSeek-R1, an openly released RL-trained reasoning model, matched leading closed models on math and coding — triggering a market reckoning over AI capex.
CAIS · Scale AI — As models saturated existing tests, a 2,500-question expert exam launched on which frontier models initially scored in the single digits — a fresh yardstick for the distance to general capability.
Google — Errors fell as the code grew — the long-sought sign that scaling can work.
Rigetti Computing — Rigetti launched its 84-qubit Ankaa-3 superconducting system with a 99.5% median two-qubit gate fidelity — a major reliability jump from a redesigned qubit layout and Alternating-Bias Assisted Annealing, available on Rigetti's cloud and later AWS Braket / Azure.
OpenAI (o3) — o3 scored 76–88% on ARC-AGI-1 (human ~85%) — the first AI to move beyond memorization on it.
Figure AI / BMW — Figure spent ~11 months at BMW Spartanburg, supporting 30,000+ vehicles.
Boston Dynamics — Boston Dynamics retired its decade-old hydraulic Atlas and revealed a fully electric Atlas with super-human range of motion (360° hip, waist and neck rotation) — its pivot from research robot to industrial humanoid, with pilots beginning at Hyundai.
KSTAR / KFE — In its 2023–24 campaign, after a tungsten divertor upgrade, KSTAR sustained high-confinement (H-mode) plasma for 102 s and held ion temperatures of 100M°C for 48 s — both device records. The team's next goal is 300 s above 100M°C by the end of 2026.
Agility Robotics — Digit began moving totes in paid pilots at GXO and Amazon.
Neuralink — Noland Arbaugh received the 1,024-electrode N1; weeks later he moved a cursor and played chess by thought.
SK hynix — High-bandwidth memory became the other scarce frontier part, with SK hynix leading supply for AI GPUs.
JET (UKAEA) — JET's final deuterium-tritium campaign, from just 0.2 mg of fuel.
Harvard / QuEra — First programmable logical processor running algorithms on dozens of error-corrected qubits.
China (Huaneng / Tsinghua) — World's first commercial pebble-bed high-temperature gas-cooled SMR.
Atom Computing / IBM — Atom Computing (1,180) and IBM Condor (1,121) both crossed 1,000 physical qubits.
EHang — China's CAAC certified the autonomous EH216-S — the first passenger eVTOL type certificate anywhere.
Waymo — Regulators allowed 24/7 paid driverless service in a major dense city.
OpenAI (GPT-4) — GPT-4 was the first model at the 1e25 FLOP scale; over 30 models from 12 developers have since crossed it.
NuScale Power — NuScale's design became the only SMR certified by the US regulator.
NIF / LLNL — 3.15 MJ out from 2.05 MJ of laser energy in — the first lab fusion gain.
China — China finished its three-module Tiangong station — a crewed government station, and a competitive backdrop for commercial LEO.
OpenAI — ChatGPT reached 100M users in two months — the fastest-adopted app to date and AI's consumer inflection point.
Samsung — Samsung's 3nm was the first to ship GAA transistors, the architecture now succeeding FinFET at the frontier.
NASA (CLD) — NASA's Commercial LEO Destinations program funded Orbital Reef, Starlab and a Northrop concept to seed ISS successors.
Hyundai Motor Group — Hyundai Motor Group completed its acquisition of a controlling 80% stake in Boston Dynamics from SoftBank (valuing it at $1.1B) — bringing the Atlas humanoid maker in-house and positioning the group as a humanoid frontrunner.
EAST (ASIPP) — Peak ion temperature ~10× the sun's core — the heat fusion needs.
Waymo — Waymo One opened fully driverless rides to the public in Phoenix.
Rosatom — Akademik Lomonosov began commercial operation in Pevek with two 35 MWe KLT-40S reactors.
Google — 53-qubit chip did in minutes a task framed as intractable for classical supercomputers.
ASML / TSMC — TSMC's 7nm+ became the first high-volume node using ASML's extreme-ultraviolet machines — the tool that unlocked everything below 7nm.
SpaceX — Booster B1021 flew a second time (SES-10) — proving an orbital rocket can be reused, not just recovered.
SpaceX — Falcon 9 became the first rocket to land its first stage after an orbital launch (Dec 21, 2015).
BrainGate — A paralyzed person controlled a cursor via a 96-channel Utah array — the field's proof of concept.
ISS partners — Humans have lived aboard the ISS without a break since November 2000 — the legacy these stations must carry on.
JET (UKAEA) — 16 MW peak fusion power, Q ≈ 0.67 — the benchmark for 25 years.