Ask who's winning the humanoid robot race and you'll get four different answers — and they're all true, depending on the question you're actually asking. By units shipped, a Chinese company most Americans have never heard of is laps ahead. By the only question with a checkout button — can you buy one? — only a home-robot startup has a real order page. This robotics competition isn't one race, it's four.
Honest framing: "who leads" depends entirely on the metric. By units shipped, Agibot leads by far. By verified productive deployment, only Agility Digit and Figure 02 reach Tier 1 of the Humanoid Reality Index (September 12, 2026) — Optimus sits at Tier 3, "Demoing." By published consumer price, only 1X NEO has a real order page. By strength and durability specs, Atlas leads. Tesla leads only on scale ambition — still targets, not output.
Tesla Optimus specs & timeline
173 cm (5'8"), 57 kg, Gen 2.5/Gen 3 body (original December 2023 Gen 2: 73 kg). Tesla claims 28+ DoF for the body; ~40+ including hands is the honest estimate — no Gen 3 total published. Hand: 11 DoF per hand on Gen 2; Gen 3 reported at 22 per hand with 25 actuators — a secondary-source claim, not a Tesla datasheet (Gen 3 not publicly shown as of early October 2026). Carry 20 kg two-handed (published), max lift ~68 kg (reported), battery 2.3 kWh, "8-hour full work day" a target, not a tested result; compute FSD-derived, vision-only; AI5 targeted ~April 2027 (conditional, per Q3 call). Speed: 8 km/h target, ~1.2 m/s more recent reported figure, older trackers 0.6–0.8 m/s.
Confirmed: Fremont early production began October 2026 (Q3 letter, October 21); first production-quality bots to Optimus Academy training, not sale. Targets: 20,000 a week by end of 2027 ("somewhat bumpy"), outside deliveries around end of 2027, industrial jobs first. H2 2027 is only the earliest window (non-public JPMorgan presentation). Output reports conflict — "several hundred a week" (Ground News, Electrek, The Information) vs "around 1,000 a week by late September, around 15,000 units in 2026" (supply-chain reports via optimusk.blog, September 11, 2026). Tesla disclosed neither.
Boston Dynamics Atlas capabilities
Atlas is the veteran of this innovation race: a product version unveiled in January 2026 at CES and Hyundai's media day, built for factory work. Official spec sheet: 1.9 m, 90 kg (ignore outdated 1.5 m and "~180 cm" figures). 56 degrees of freedom as a company figure, unconfirmed whether that includes the new hand. Launched with 3-finger grippers; new 4-finger, 13-DoF hand unveiled October 1, 2026. Grip-force and tactile specs are not disclosed; speed not disclosed either — a third-party "5.5+ km/h" estimate must never be printed as a Boston Dynamics figure. Where Atlas leads: 50 kg instantaneous lift, 30 kg sustained, 20 kg one-handed, 2.3 m reach, ~4 hours battery (2 hours heavy), autonomous swap in ~3 minutes, IP67, −20°C to 40°C, plus DeepMind partnership and Orbit software.
Fleet scheduled to ship to Hyundai's RMAC in 2026; on September 21, 2026, Atlas was "now in training" there in Savannah, Georgia, on parts logistics/sequencing. New CEO Rohit Prasad took over October 7, 2026, with no new availability date, price, or roadmap. Plan: additional customers early 2027, parts sequencing in Georgia from 2028, component assembly targeted by 2030, Hyundai's 25,000+ units "over the next few years" — all plans, no detailed timeline. Reality check: Atlas is not commercially available, not for sale, not in revenue production deployment.
1X NEO features & price
1X built a home robot: 168 cm (maker-published), ~30 kg, soft fabric, claimed 22 decibels. Dexterity, from 1X's per-part figures: 22 per hand, 7 per arm, 6 per leg, 2 spine, 3 neck — 75 by arithmetic, not a 1X-stated total; don't quote a bare "75" or the "~30" simplification. July 2026 coverage citing 25-DoF hands conflicts with the per-part breakdown; we use the per-part figures. Tendon-drive hand, 1.4 m/s walk (manufacturer claim); 6.2 m/s "max run" on 1X's sheet is flagged by every careful tracker as unverified for sustained running — attribute it, don't present it as demonstrated. Lift 154 lb (~70 kg), carry 55 lb (~25 kg), per-arm 18 lb (~8 kg) — manufacturer claims, no independent measurement. Battery 842 Wh, around 4 hours; brain is NEO Cortex on NVIDIA's Jetson Thor.
Confirmed: order page open since October 28, 2025; 10,000 preorders in 5 days (company claim); Hayward "NEO Factory" (58,000 sq ft), first units to internal R&D/home testing (April 2026 letter). Claimed: US deliveries in 2026, other markets from 2027, CEO says shipping before end of 2026, capacity 100,000+ a year by end of 2027; not verified as of October 2026: delivery to a paying customer's home. One trade outlet reports early-access units reaching homes; no paying-customer delivery is publicly confirmed. The "60–70% autonomous" line has no primary source; a June Forbes "20,000 shipping this year" line exceeds 1X's own 10,000-a-year capacity, unconfirmed by 1X. Privacy trade-off: "Expert Mode" lets 1X staff remotely guide NEO through unfamiliar chores.
Agibot China threat
Flagship: full-size bipedal Yuanzheng A2 Ultra; X2: half-size companion; figures are not interchangeable. A2 Ultra: 169 cm official (older 175 cm press figure superseded), ~69 kg (Lite 63–64 kg). X2: 131 cm, 35 kg. DoF: 40 active on the A2 Ultra (neck 2, arms 7 each, legs 6 each, hand 6 per hand). Hand: 6 DoF standard on the A2 Ultra; 12-DoF SkillHand is an optional accessory, never standard. Walk: 1.2 m/s max official; X2 up to 1.8 m/s, typically 0.8 m/s or less. Payload: A2 Ultra rated around 2 kg per arm officially — never 15 kg (unverified database entry) or 40 kg (unreleased A2 Max). X2: 3 kg max in specific postures, 1 kg or less across its full workspace. Battery: swappable 14.4 Ah, ~3 h standing, 1.5+ h walking, 2-h charge; X2 ~500 Wh (product page) vs ~421 Wh (SDK sheet) — flagged, not resolved. Showpiece: Guinness record 106 km walk, November 2025 (single-source); reported SAIC-GM line deployment (single-source).
H1 2026: ranked number one globally by units — but houses disagree: Counterpoint ~9,700 (over 43%), IDC over 8,600 (~35%), SAG 8,400 (44%). Different scopes — shipments vs sales, wheeled/half-size/consumer units counted or not — explain the spread. Milestones: 15,000th in June 2026, 20,000th by September 24, 2026. Reality check: 300+ robots at Chimelong Spaceship Park, Hengqin (September 2026), in entertainment, guidance, retail, education, hospitality — not factory work. The US import ban on new Chinese humanoid/quadruped robots landed in late July 2026, walling Agibot out of the American market. Hong Kong IPO process started July 2026, pushed to 2027 by the 120-day rule.
Price comparison breakdown
| Robot | What you'd pay | Label |
|---|---|---|
| Tesla Optimus | $20,000–$30,000 (low-volume materials $50,000–$60,000, Reuters) | TARGET — no official price |
| Boston Dynamics Atlas | $140,000–$150,000+ (Spot at ~$75,000 is a floor reference only) | ESTIMATE — no official price |
| 1X NEO | $20,000 or $499/month, $200 refundable deposit | PUBLISHED PRICE — real order page |
| Agibot A2 Lite | $44,560 + shipping/duties (EU distributors list $55,975 Lite / $100,614 Ultra) | STORE-LISTED / CHANNEL VARIANCE |
| Agibot A2 / A2 Ultra | Contact sales | NO PUBLIC PRICE |
| Agibot X2 | $24,240 | STORE-LISTED |
Takeaway: only NEO ($20,000) and X2 ($24,240) are directly buyable; Optimus has a target, Atlas an estimate. NEO's subscription terms are launch-era until 1X republishes.
Capability showdown
| Spec | Optimus | Atlas | NEO | A2 Ultra |
|---|---|---|---|---|
| Height | 173 cm | 1.9 m | 168 cm | 169 cm |
| Weight | 57 kg | 90 kg | ~30 kg | ~69 kg |
| Degrees of freedom | 28+ body (claim) | 56 (company figure) | 75 by arithmetic | 40 active |
| Hand | 11 Gen 2 / 22 Gen 3 (secondary) | 4-finger, 13 DoF (Oct 2026) | 22 per hand, tendon-drive | 6 DoF standard |
| Walk speed | 8 km/h target / ~1.2 m/s reported | Not disclosed | 1.4 m/s claim | 1.2 m/s max |
| Payload | 20 kg carry (published) | 50 kg instant / 30 kg sustained | ~25 kg carry (claim) | ~2 kg per arm |
| Runtime | ~8 h (target) | ~4 h (2 h heavy) | ~4 h | ~3 h standing |
Production scale comparison
Scale is where Tesla's story is biggest and its reality thinnest — and Agibot's quietest and thickest. Tesla's Fremont line is sized at ~1M robots a year; the Texas second-generation line is planned for 10M a year long-term; neither is current output, both are line sizing and plans. Hyundai/Atlas: 25,000+ units "over the next few years" plus a US facility rated at 30,000 a year — plans, no detailed timeline. 1X's Hayward plant: 10,000 units in year one — booked out in five days — claimed target 100,000+ a year by end of 2027. Agibot is the only one of the four with five-figure shipments on the books: ~8,400–9,700 in H1 2026 by house, 20,000 cumulative by September 2026. Disagreement: SAG/Bloomberg put industrial/commercial above 70% of H1 shipments; Counterpoint-via-Cryptopolitan put over 60% of H1 volume in entertainment, performance, research, data — different cuts. Don't claim Agibot's shipments are mostly factory work: named deployments are Chimelong entertainment/service, a reported Longcheer line, a single-source SAIC-GM line.
Funding & investment analysis
Tesla closed at $372.22 on Friday, October 23, 2026, down 0.63%, market cap ~$1.47 trillion — a dated snapshot, not investment advice. Q3 revenue $28.05B (flat), net income $787M (down 43%), R&D $2.76B (up 69%). Tesla discloses no Optimus revenue at all; Optimus is a valuation narrative — Musk's September 2025 forecast of "around 80% of Tesla's value" one day — not an earnings driver, no timeframe attached. Boston Dynamics, majority-owned by Hyundai since 2021, discloses no Atlas price or standalone revenue.
1X funding: $23.5M Series A2, March 2023 (OpenAI Startup Fund); $100M Series B, January 2024 (EQT Ventures) — $125M+ confirmed. September 2025 talks to raise up to $1B at $10B+ reportedly raised less than half the target; August 2026 SoftBank majority-stake talks at ~$6B (talks, not a deal); Agibot, founded February 2023 by ex-Huawei engineers, 10+ rounds; Tencent led its Series B (March 2025) alongside HongShan, Hillhouse, BYD, LG, Mirae Asset. Funding total conflicts: over $725M per one tracker vs $200M+ disclosed per another — a range, with that caveat. Valuation ~RMB 15B in 2025 to ~RMB 50B in the latest round (Caixin via RobotBelt, October 2026).
Market predictions 2027–2030
IDC (October 2026): H1 2026 nearly 25,000 units (up 432%); 2030 outlook over 750,000. Smart Analytics Global (August 2026): ~60,000 full-year 2026, 500,000 by 2030. Counterpoint: full-year 2026 above 50,000 after H1 above 22,000 (up ~300%). Goldman Sachs (revised 2026 outlook): 75,000 in 2026, 890,000 in 2030, ~6.5M a year by 2035, ~$138B market — older base case (1.4M units, $38B by 2035) superseded, don't quote it as current. Morgan Stanley: ~1 billion humanoids by 2050, $5T+ incl. supply chain/services.
Tech innovation is real but the forecasts are scenarios, not promises, and analysts agree: tens of thousands in 2026, hundreds of thousands by 2030, millions a year by 2035 — but the range between houses is ~9x even for 2035, and every forecast assumes reliability, safety, and cost get solved on schedule. Conflict: Morgan Stanley's 2030 deployment figure of roughly 40,000 versus Goldman's 2030 shipments of 890,000. Different definitions — deployed vs shipped — so don't average them.
| Next milestone | The catch | |
|---|---|---|
| Optimus | 20,000/week + outside sales, end of 2027 (target) | No output, price, or customers disclosed |
| Atlas | Parts sequencing, Georgia, 2028 (plan) | Not for sale; no price; "in development" |
| NEO | US deliveries in 2026 (claim) | No paying-customer delivery confirmed |
| Agibot | Hong Kong IPO, 2027 (plan) | US import ban; volume isn't factory work |
Which humanoid robot can you actually buy? 1X NEO at $20,000 (or $499 a month) and Agibot's X2 at $24,240. Optimus and Atlas have no price.
Will Optimus hit 20,000 a week by end of 2027? That's Musk's target, "somewhat bumpy" ramp and all — an ambition, not a schedule.
The bottom line
There is no single winner in the humanoid robot race, and anyone selling you one is selling you their metric. If you count units shipped, Agibot already won 2026; if you count verified productive work, Agility's Digit and Figure's pilots are ahead of every robot here. If you count "can I buy one," NEO is the only answer with a checkout page. If you count industrial muscle, Atlas is the machine you'd put on a factory floor today. If you count ambition, Tesla's million-unit line and 20,000-a-week target are in a league of their own — as targets. What to watch: 1X delivery, Atlas sequencing in 2028, Tesla's end-of-2027 targets, Agibot's 2027 IPO.
