What Optimus is
Optimus (also called Tesla Bot) is a general-purpose humanoid robot Tesla has been developing since August 2021. The pitch: reuse the vision-based AI stack from Tesla's cars, the actuator and battery know-how from its EV program, and its manufacturing scale to build a labor robot cheap enough to sell in the millions. Elon Musk has repeatedly claimed Optimus could eventually be worth more than Tesla's entire car business.
The core idea: a robot built like a car
Tesla's thesis is that a humanoid robot is "a car with legs" — the hard problems (real-world computer vision, neural network planning, motors, batteries, thermal management, mass manufacturing) are ones Tesla already solved for vehicles. Optimus runs on the same end-to-end neural network lineage as Full Self-Driving, and its joints borrow design principles from automotive components (its joint design reportedly draws on Model Y suspension concepts; battery thermal management comes from the vehicle side).
This is also why Tesla believes it can hit a $20,000–30,000 price when competitors sell robots for six figures: it plans to reuse its automotive supply chain and build at automotive volumes.
Where the program actually stands (July 2026)
As of mid-2026: Optimus is not for sale. There is no public pre-order, deposit, or waitlist — any "invest in Optimus" offer circulating online is a scam. The robots work only inside Tesla facilities (Fremont and Giga Texas), doing tasks like battery cell sorting, parts handling, and visual quality inspection. Independent estimates put the internal fleet in the low hundreds of units.
The next milestones: a V3 public reveal expected around mid-2026, low-volume production starting at Fremont in late July/August 2026, volume ramp through 2027, and possible external sales in late 2027.
Healthy skepticism: the timeline slippage record
Optimus has a consistent pattern of missed targets. In January 2025 Musk predicted ~10,000 robots built that year; by January 2026 he acknowledged none were doing genuinely useful work yet. The Gen 3/V3 reveal slipped from Q1 2026 to "mid-year." At the October 2024 "We, Robot" event, robots that mingled with guests and tended bar were later reported by Bloomberg to have been largely teleoperated by humans — a recurring criticism of Tesla demos.
Musk's own framing has grown more cautious: he now says initial production will be "extremely slow because everything is new — it's not like building a car," and calls 2026 output "literally impossible to predict" given ~10,000 unique parts on an entirely new line. Treat every date in this guide as a target, not a promise.
Program timeline
AI Day announcement
Musk unveils the "Tesla Bot" concept — memorably represented by a human dancer in a spandex suit. Specs promised: 5'8", designed to do dangerous, repetitive, boring work.
First prototype walks (AI Day 2)
"Bumble C" development platform walks unassisted on stage; a more refined Optimus body is shown but carried out by staff.
Gen 2 revealed
Major redesign: 173 cm, ~57 kg, faster walking, 11-DoF hands with tactile fingertip sensing, improved neck and balance.
"We, Robot" event
Optimus units serve drinks and chat with guests in LA. Bloomberg later reports the robots were largely teleoperated — a credibility setback that still shadows Tesla demos.
Leadership change
Program lead Milan Kovac departs; Ashok Elluswamy (head of Tesla's Autopilot/AI team) takes over Optimus, tightening the FSD–robot integration and shifting training strategy toward first-person video learning.
Supply chain locks in
Reports of a ~$685M actuator order to China's Sanhua Intelligent Control ignite the "T-chain" robot supplier rally on Chinese markets. Tesla audit teams begin intensive factory inspections of Chinese suppliers.
Fremont conversion announced
On the Q4 2025 earnings call, Tesla announces Fremont will convert from Model S/X production to Optimus, targeting an eventual 1 million robots/year. Reports name seven Chinese firms (Lens Technology, Sanhua, Tuopu, Hengli Hydraulic and others) as core V3 suppliers.
Gen 3 demos; Mars plans shelved; Terafab
Tesla publishes Gen 3 demo videos highlighting the dexterous hand and reducer upgrades. Musk says Optimus 3 is "walking around" but delays the reveal for polish. Tesla announces Terafab, an in-house chip fab initiative to reduce dependence on TSMC/Samsung for AI5.
Last Model S/X built
Fremont ends a 14-year Model S run. The line is dismantled and rebuilt for Optimus in roughly four months.
Production team photo
Musk visits the Fremont Optimus assembly line and posts a team photo — widely read as the sign that production start is imminent. He cautions output will be "extremely slow at first." Chinese T-chain supplier stocks (Tuopu, Shuanghuan, Sanhua) surge on the news.
Low-volume production begins
V3 reveal expected near production start. Volume ramp targeted for 2027, when a second, higher-capacity Optimus factory at Giga Texas (for the Gen 4 variant) is due to come online.
Hardware & specifications
Naming gets confusing, and most articles get it wrong. Gen 2 is the 2023 body still used in factories. "Gen 3" usually refers to the upgraded 22-DoF hands retrofitted onto Gen 2 bodies. "V3" is the full new production body due for reveal in mid-2026 — its final specs won't be confirmed until then.
| Spec | Gen 2 (2023, verified) | Gen 3 hands upgrade | V3 (expected, unconfirmed) |
|---|---|---|---|
| Height / weight | 173 cm (5'8") / ~57 kg | Same body | Similar; more integrated, compact structure |
| Hands | 11 DoF, tactile fingertips | 22 DoF, 50 actuators (25 per forearm+hand) — ~4.5× Gen 2 actuation | 22-DoF hands standard; hands ≈20% of total BOM cost |
| Joints / actuators | ~28 structural actuators | — | ~37 joints reported; harmonic + planetary drive mix |
| Locomotion | Walks ~5–8 km/h in demos; balances on one leg | — | ~1.2 m/s walk; stability on 15° slopes (analyst estimates) |
| Payload | ~20 kg carry | — | TBC |
| Compute | FSD-derived SoC | — | Tesla AI5: 2,000+ TOPS, ~5× memory bandwidth of prior chip |
| Vision | 8-camera suite (Autopilot heritage), no LiDAR | — | Same vision-only philosophy |
| Voice / language | — | — | xAI Grok integration for conversation |
| Battery | ~2.3 kWh, full-day target | — | EV-derived thermal management, liquid-cooled joints |
Drill down: actuators — the heart of the robot ~56% of component value
Actuators (joint modules) are the single biggest cost and engineering challenge. Optimus uses two families:
- Rotary actuators — motor + harmonic reducer + sensors + encoder in one unit, used in shoulders, hips, torso. Harmonic (strain-wave) reducers provide precise, backlash-free rotation.
- Linear actuators — motor + planetary roller screw, converting rotation into powerful straight-line motion for knees, hips, and other high-load joints. Roller screws are the most valuable single component inside them.
Gen 3 reportedly uses a mixed reducer strategy: harmonic reducers for upper body and torso, planetary for high-load lower body, with cycloidal/RV types in testing for the highest-load joints. Tesla holds patents on a tendon-driven, mechanically-actuated hand architecture (filed the same day as the We, Robot event).
Component value ranking inside an actuator (per Chinese brokerage research): planetary roller screw > frameless torque motor > harmonic reducer > six-axis force sensor.
Drill down: the 22-DoF hand
The hand is what separates a walking mannequin from a useful worker. Gen 3's hand-forearm system packs 25 actuators per side (50 total), with tendon-driven fingers, micro lead-screws, coreless (hollow-cup) motors for fine control, tactile sensors, and ultra-flexible cabling rated for tens of millions of bend cycles. Musk has said each arm carries 26 actuators in the V3 design and that Tesla had to "create the supply chain from scratch" for hands and forearms — there was simply no existing industry making these parts at scale.
Cost consequence: the dexterous hands alone account for nearly 20% of V3's bill of materials, per Counterpoint Research.
Drill down: compute — AI5, Terafab, and the vision-only bet
Optimus V3 runs Tesla's AI5 chip (2,000+ TOPS, roughly 5× the memory bandwidth of its predecessor), supporting a vision-based end-to-end architecture — camera pixels in, motor commands out, no LiDAR, no modular perception/planning/control stack. The AI chip is estimated at $5,000–6,000 of the robot's cost, roughly 25% of the target BOM.
AI5 is manufactured externally (TSMC in Taiwan; Samsung in Texas). In March 2026 Tesla announced Terafab, an in-house fab program aimed at eventually removing that dependency — a direct response to chip supply concentration risk for both Optimus and Cybercab.
Drill down: sensors, materials, and battery
- Vision: 8 cameras, Sony CMOS sensors (industry-standard IMX-class machine-vision parts).
- Balance: STMicroelectronics IMUs/MEMS gyros and accelerometers for proprioception.
- Force sensing: six-axis force/torque sensors at major joints (Chinese suppliers Coliy/Keli Sensor and Haozhi in validation/supply); tactile sensors in fingertips.
- Materials: lightweighting via engineering plastics (PEEK for screw nuts and bearings), carbon fiber, and integrated die-casting — "plastic replacing steel" cuts ~30% weight in some parts.
- Battery: ~2.3 kWh pack with EV-derived thermal management; Sanhua supplies miniaturized liquid-cooled joint modules to manage heat in high-load actuators.
- Magnets: NdFeB rare-earth magnets in nearly every motor — a genuine geopolitical chokepoint (see Supply Chain). After China tightened export controls, Tesla reportedly redesigned to reduce rare-earth magnet content per robot.
How Optimus learns
Optimus is not programmed in the traditional sense — no G-code, no teach pendants, no scripted motion paths. It runs a single end-to-end neural network that maps camera input directly to motor commands, descended from Tesla's Full Self-Driving stack. The interesting story is where the training data comes from.
Stage 1 — Learning from first-person human video
When Ashok Elluswamy took over in mid-2025, Tesla pivoted from teleoperation-heavy data collection to a video-first strategy. Workers wear custom camera rigs — helmets plus backpacks with five Tesla-built cameras — and simply perform ordinary tasks (folding laundry, sorting parts, using tools) naturally. The footage is labeled and fed into Tesla's Cortex training cluster (~50,000 NVIDIA H100 GPUs).
The network learns the demonstrated behavior end-to-end, including implicit knowledge like appropriate grip force. Tesla's Optimus AI lead Ashish Kumar called learning directly from first-person human video the key technical breakthrough; the team reports bi-manual dexterous manipulation across many tasks "with barely any data" from teleoperation. Roboticists note the first-person rig captures fine detail — joint and finger positions — that third-person video misses.
Stage 2 — Simulation and "neural physics" world models
Tesla augments real video with synthetic data: video-generation AI models act as neural physics engines, creating simulated worlds where the robot practices without moving a physical servo. Reinforcement learning in these environments hardens skills before they're deployed to hardware. This hybrid (real video for breadth, simulation for controlled repetition) is becoming the industry-standard recipe.
Stage 3 — Fleet learning: the factory flywheel
Every hour an Optimus works in a Tesla factory generates real-world training data that flows back into the shared network — the same fleet-learning loop that powers FSD. This is arguably Tesla's deepest moat: competitors without hundreds of robots doing real work in real factories cannot generate equivalent data. The strategy explains why Tesla is its own first customer.
The endgame — learning from YouTube
Musk has said the goal is for Optimus to learn tasks from ordinary internet video: "If Optimus can watch YouTube videos and learn to do that thing… you really have task extensibility that is dramatic." Cooking demos, repair tutorials, and how-to content would become training data. This remains aspirational — current learning still relies on curated first-person footage — but it's the stated destination.
Reality check: teleoperation vs. autonomy
Two things are true at once. Tesla still uses teleoperation (motion-capture rigs) for some data collection and some public demonstrations — the We, Robot bartenders were largely human-piloted. And the trained tasks shown in Tesla's learning videos genuinely run on the robot's own network. When you watch any Optimus video, the honest question is always: which one is this? Tesla has been criticized for not labeling clearly. By late 2025, observers reported Gen 3 units in Giga Texas doing simple material-handling autonomously on the FSD-derived stack.
Supply chain — the China & Taiwan story
This is where the most revealing Optimus intelligence comes from. Chinese and Taiwanese suppliers file public disclosures, get audited, expand factories, and their stock moves telegraph Tesla's real production intent long before Tesla says anything. Analysts estimate Chinese firms hold roughly 70% of the Optimus component supply chain — Morgan Stanley calculates the Chinese supply chain can cut humanoid manufacturing cost by up to two-thirds, and motion components (motors, gears, screws) alone are ~55% of total cost.
Core suppliers by component
| Component | Supplier | Origin | What to know |
|---|---|---|---|
| Linear actuators / thermal | Sanhua Intelligent Control (三花智控) | China | Reported ~$685M Tesla actuator order (Oct 2025, officially denied but followed by 320% robotics revenue growth). Also supplies liquid-cooled joint modules. Building Thailand capacity near-shored to Giga Texas. |
| Rotary actuators / joint assemblies | Tuopu Group (拓普集团) | China | Long-time Tesla EV supplier; integrates motor + screw + sensor into plug-and-play joint units. Tesla is ~35–40% of its revenue. Planning Mexico JV capacity. |
| Harmonic reducers | Leader Harmonic / Green Harmonic (绿的谐波) | China | 60%+ Chinese market share; priced 30–60% below Japan's Harmonic Drive Systems. Delivered reducer batches for ~700 robots by Q2 2025; new Suzhou plant targets 500K units/yr; goal of ~60% of Optimus reducer supply. |
| Harmonic reducers (dual source) | Harmonic Drive Systems | Japan | Global incumbent (~35% share). Tesla dual-sources against Green Harmonic for supply security. |
| Planetary roller screws | Wuzhou Xinchun (五洲新春), Beite Tech, Hengli Hydraulic, Xinjian Transmission | China | The high-value linear-motion part, ~50% of German (Rollvis) pricing. Wuzhou's Thailand plant online 2026; Xinjian started IPO guidance Jan 2026; Hengli named among seven core V3 suppliers. |
| RV / cycloidal reducers | Shuanghuan Transmission (双环传动) | China | Enters T-chain indirectly via Tuopu and Sanhua; benefits if high-load joints move to RV-type reducers. |
| Dexterous-hand micro motors | Moons' Electric (鸣志电器), Zhaowei (兆威机电) | China | Moons' reported as coreless-motor supplier for the hand with 10K-set orders; Zhaowei's brushless motor + micro lead-screw module in supply chain audit since May 2025. |
| Structural parts / assembly | Lens Technology (蓝思科技), Changying Precision, Milison | China | Consumer-electronics-grade precision structures and potential assembly; Lens named among the seven core V3 suppliers reported Jan 2026. |
| Force / tactile sensors | Keli Sensor (柯力), Hanwei (汉威), Haozhi (昊志机电) | China | Six-axis force-torque sensors for shoulder/elbow joints; Haozhi passed a Tesla factory audit Q3 2025 per disclosures. |
| AI5 chip fabrication | TSMC (台积电) | Taiwan | Primary fab for Tesla's AI5, alongside Samsung's Texas fab. The single largest non-Chinese external dependency — the driver behind Tesla's Terafab in-house fab announcement (Mar 2026). |
| Vision sensors | Sony | Japan | CMOS image sensors for the 8-camera array. |
| Precision motors | Nidec | Japan | Brushless motors and actuator drive components; long-standing Tesla supplier. |
| IMUs / MEMS | STMicroelectronics | EU | Gyros and accelerometers for balance. |
| Precision actuator assemblies | Moog | USA | Aerospace-grade electromechanical actuation expertise. |
| Rare-earth magnets | Chinese NdFeB producers | China | The chokepoint: China's export controls forced Tesla to cut magnet content per robot. US alternatives (MP Materials) are years from equivalent scale. |
Drill down: why Taiwan matters beyond TSMC Taiwan
TSMC is the headline: it fabricates the AI5 chip that is Optimus's brain, making Taiwan the most concentrated single-point risk in the whole robot. But Taiwan's role is broader:
- Taiwanese tech media as an intelligence source: TrendForce, DigiTimes, and the Taiwanese supply-chain press consistently break Optimus manufacturing news early because they cover the fab and component ecosystem daily. (The July 1 "production team photo" analysis above came via TrendForce.)
- Harmonic drive challengers: Taiwan's Main Drive Corp (spun from ITRI expertise, Miaoli County) is scaling toward 100,000 harmonic drives/year, positioning Taiwan as a third source between Japan and China.
- Precision machinery ecosystem: Taiwan's machine-tool and reducer industry (Hiwin in ballscrews/linear motion, Varitron and others in gearing) sits adjacent to every humanoid BOM, and Taiwanese ODMs are frequently discussed as candidates if Tesla diversifies assembly away from mainland China.
Drill down: the geopolitical squeeze China risk
A February 2026 Tom's Hardware analysis concluded bluntly that the majority of Optimus's critical materials and suppliers are located in China. The Wall Street Journal reported the same dependency in April 2026, noting China has designated humanoid robots one of six priority industries and aims for a resilient domestic supply chain by 2027 — while US lawmakers proposed a commission on American robotics competitiveness.
Tesla's mitigation playbook, visible in supplier disclosures:
- Dual-sourcing key parts (Japanese HDS + Chinese Green Harmonic for reducers).
- Near-shoring: Sanhua, Tuopu, Xinjian, Beite, and Xusheng approved for Thailand factories serving Giga Texas; Tuopu and Green Harmonic planning Mexico JVs.
- Vertical integration: Terafab for chips; reduced rare-earth magnet content after China's export controls.
The tension is structural: the $20K price target arguably cannot be reached without Chinese component costs, yet a US flagship robot built ~70% from Chinese parts is politically exposed in both directions.
Drill down: how Tesla qualifies suppliers
From late 2025, Tesla audit teams ran intensive factory inspections (审厂) across candidate Chinese suppliers — Hengli Hydraulic, Sanhua, and others — using three criteria repeatedly cited in Chinese industry press: technical exclusivity, mass-production stability, and cost control. Tesla demands calibrated complete modules (a finished rotary joint, not a bag of parts), which pushes system-integration capability down to suppliers and cuts Tesla's own BOM cost. One screw supplier described Tesla's ask: smaller, 25% more durable, and 25% cheaper than the European equivalent.
Production & roadmap
Fremont: the first line
Fremont's Model S/X line (peak capacity ~100K vehicles/yr, recent sales ~30K/yr) ended production May 9, 2026 and is being rebuilt for Optimus in about four months. Low-volume production is targeted for late July/August 2026. The first line's job is not volume — it's process validation, quality calibration, and locking the final technical package. Musk declined to give any 2026 volume target and says early robots will do "simple skills in the factory" first. Long-run Fremont target: 1M robots/year.
Giga Texas: the volume play
A second Optimus factory is under construction on Giga Texas's north campus (~5.2M sq ft reported), with production expected around summer 2027 and designed for the higher-volume Gen 4 variant. Musk's stated ladder: Optimus 4 targeting ~10M/yr, and eventually 50–100M/yr for later generations — numbers best treated as ambition markers rather than plans.
Cost curve and pricing
Counterpoint Research estimates V3 manufacturing cost above $60,000/unit in H2 2026 while volumes stay under 10,000 — with the 22-DoF hands nearly 20% of BOM. Musk's math: at 1M units/year, cost falls to $20–25K, of which the AI chip is $5–6K. The consumer price target remains $20–30K, with limited external sales possibly starting late 2026–2027. Chinese analysts sketch the ramp as: 2026 pilot line (~5K/yr pace), ~50K/yr line by end-2026, 500K in 2027, then a million-unit scale.
What Optimus does today, inside Tesla
- Battery cell sorting at Fremont and Giga Texas
- Parts handling / moving components between stations
- Visual quality inspection using the 8-camera suite
- One consumer-facing cameo: serving at the Tesla Diner in LA (a Gen 3 unit was observed working there before being pulled back within months)
Competition & market context
The rival field
CES 2026 marked the shift from demos to deployments across the industry. Figure AI (Helix vision-language-action model, household demos), Boston Dynamics (electric Atlas entering Hyundai factories), 1X (NEO home robot taking $20K pre-orders), Agility Robotics (Digit in warehouses), and China's Unitree (shipping robots with full SDK access, Shanghai IPO planned) and UBTech (commercial orders in logistics). Tesla's edges: manufacturing scale, custom silicon, and the factory data flywheel. Its gaps: no product on sale, no developer access, and a demo-credibility deficit.
Why the market cares
Musk has said Optimus could be "worth more than the car business and FSD combined," and Tesla's Master Plan 4 (Sep 2025) reframed the company around AI and robotics, with Musk claiming Optimus will represent ~80% of Tesla's future value. Analysts frame the addressable market as "manual labor" itself. Industry forecasts see 500K+ humanoids operating by the mid-2030s. Skeptics counter that demand, reliability, and safety for general-purpose humanoids remain unproven — and Tesla's record of slipped timelines is the pattern to bet on.
Quick answers
Can I buy one?
No. As of July 2026 there is no sale, pre-order, deposit, or waitlist. Anything claiming otherwise — especially "Optimus investment" or crypto-bonus offers using Musk deepfakes — is a scam. External sales are tentatively targeted for late 2026–2027.
How much will it cost?
Target: $20,000–30,000 at scale. Actual build cost today is far higher — $60K+ per unit at low volume, with early internal builds estimated at $50–100K+.
Is it autonomous or remote-controlled?
Both, depending on context. Factory tasks shown in Tesla's learning videos run on the robot's own neural network; some data collection and past public demos used human teleoperators. Always ask which mode a given video shows.
What's the difference between Gen 3 and V3?
"Gen 3" mostly refers to the 22-DoF hand upgrade on the Gen 2 body; "V3" is the full new production-intent robot expected at the mid-2026 reveal. Media use the terms interchangeably, which causes endless confusion.
What should I watch next?
Four signals, in order of reliability: (1) the V3 reveal and Fremont production start (Jul–Aug 2026); (2) Tesla earnings calls (next one covers Q2 2026); (3) Chinese supplier disclosures and T-chain stock moves — real orders show up in supplier revenue before Tesla says anything; (4) the Q4 2026 milestone of whether the first ~100K/yr-capacity line actually ramps.
Latest news
This list is rebuilt automatically every morning at 9:00 AM ET by a scheduled web search across US, Chinese, and Taiwanese sources. It loads the latest saved run when you open the page; hit refresh to pull the newest saved batch. Below that, a set of daily-reading links you can bookmark.
Musk shares Optimus production team photo at Fremont
Musk visited the Fremont assembly line and posted a team photo, dismissing rumors that reduced demos meant production was ahead of schedule: "Optimus production will be extremely slow at first because everything is new. It's not like building a car."
T-chain stocks surge as production nears
Chinese Optimus suppliers rallied hard on the production-line photo: Tuopu Group and Shuanghuan Transmission hit daily limit-up; Sanhua, Zhejiang Rongtai, Wuzhou Xinchun and Moons' Electric followed. Analysts frame it as the shift from concept cycle to engineering-delivery cycle.
V3 cost estimated above $60K/unit in H2 2026
Counterpoint says Tesla has built a strong supply chain for V3 mass production, with the 22-DoF hands ~20% of BOM; expects long-term supplier agreements and capacity expansion in China and overseas, with AI5 delivering 2,000+ TOPS.
Last Model S/X rolls off Fremont line
A 14-year production run ends; the line begins its ~4-month conversion into Tesla's first dedicated humanoid robot factory, with Optimus output targeted for late July/August.
Q1 earnings: production confirmed for late July, no volume target
Musk confirms Fremont start, calls 2026 output "impossible to predict" across 10,000 unique parts, confirms a second Optimus factory at Giga Texas for ~summer 2027, and pushes the V3 reveal to mid-year — citing competitors copying frame-by-frame.
WSJ: Optimus depends on the Chinese supply chain
Report details Tesla reducing rare-earth magnet use after China's export controls, Chinese suppliers pre-building Southeast Asia capacity to dodge tariffs, and Morgan Stanley's estimate that Chinese supply can cut humanoid costs by up to two-thirds.
Optimus 3 "walking around," reveal delayed for finishing touches
Musk says Gen 3 is operational but needs polish before public unveiling; sightings at the Tesla Diner in LA showed the robot serving until sunset.
Gen 3 demo video published; supplier audits confirm redesign
Tesla's official Gen 3 demo highlights the dexterous hand and a mixed reducer strategy (harmonic upper body, planetary lower body, RV in testing). Chinese suppliers report late-2025 factory audits specifically for robot mass production.
Tesla announces Terafab in-house chip fab initiative
A direct response to AI5 supply concentration at TSMC (Taiwan) and Samsung (Texas) — if it works, it removes the largest non-Chinese external dependency for Optimus and Cybercab.
Seven Chinese firms reported as core V3 suppliers
Lens Technology, Sanhua, Tuopu, Hengli Hydraulic and others named (unconfirmed by Tesla) as core suppliers covering joint actuators, liquid-cooled modules, roller screws, and precision structures/assembly.