The key race: Optimus finally enters production as Unitree's stock hunts for a floor and China's two would-be listings keep waiting
Tesla gave the sector its most concrete Western proof point of the summer. Reporting around August 27-28, 2026 confirmed Optimus has entered production on Fremont's newly converted first-generation line, the same space that built its final Model S and Model X on May 10 and was torn down in roughly 46 days to make way for the robot line. The milestone is narrow but real: initial output feeds only Tesla's internal 'Optimus Academy' training and data-collection program and factory-floor tasks rather than customer sale, Musk has described the ramp as deliberately 'flat and long' before scaling, and the V3 public reveal remains undated. It still lands close to, and slightly ahead of, the 'later this year' timeline Tesla gave on its July 22 earnings call after missing an original late-July/August target, and it is the clearest sign yet that Musk's roughly $1 trillion pay package, which explicitly ties his payout to delivering 1 million Optimus robots, has a production line behind it rather than just a roadmap; the line's long-run design capacity is about 1 million units a year, with a second-generation Giga Texas line planned for up to 10 million.
Unitree Robotics' own trading story stayed unresolved rather than stabilizing. Shares remained roughly 45 percent below their August 19 Shanghai-debut peak through this update, still around four times the IPO price but with more than 200 billion yuan of market value gone, as first-half results showing adjusted net profit down 19.34 percent year over year keep feeding the same bubble debate founder Wang Xingxing opened when he pushed his own commercialization timeline from '1-2 years' out to '2-3 years.' The two listings meant to follow Unitree have still not priced: AgiBot, the Shanghai maker whose shipment volume has overtaken Unitree's, opened its Hong Kong IPO process on July 24 with CICC, CITIC Securities and Morgan Stanley as sponsors, targeting a HK$40-50 billion valuation (about $5.1-6.4 billion), and market chatter had pointed to a possible August listing that, as of this update, still had not happened. Oregon's Agility Robotics remains on the steadier procedural path, its $2.5 billion pre-money SPAC merger with Churchill Capital Corp XI still awaiting shareholder approval and regulatory clearance toward a targeted 2026 close that would make it the first publicly traded pure-play humanoid company in the United States.
Who's ahead: Hyundai puts real numbers on its Atlas order as Figure opens a data-crowdsourcing front and AgiBot's shipment lead holds
Boston Dynamics' demand signal stopped being a roadmap and became a number. At Hyundai's CEO Investor Day on August 26, the company reaffirmed 2028 U.S. robot production of 30,000 units a year across Spot, Stretch and Atlas, said an initial committed volume of 25,000 units, 83 percent of that capacity, is already confirmed, and said the production site will be named by year-end. That reaffirmation came days after Hyundai and the Korean Metal Workers' Union reached a tentative wage agreement on August 24-25 following a full-day walkout, the union's first in a decade, that had pushed cumulative production losses past 42,510 vehicles and roughly $1.3 billion; the deal heads to a ratification vote on August 31, and Hyundai has maintained throughout that the Atlas deployment plan sits outside the wage talks. Figure AI opened a different kind of front the same week: on August 25 it brought Index, a crowdsourced physical-data app, out of stealth, logging 264,000 downloads, 44,000 weekly active users and 16 million videos from 108 countries within days, paying $15 million to contributor 'Creators' so far, and committing more than $1 billion to data and compute over the next 12 months to convert that pipeline into generalization gains for its Helix model.
Independent tracking still puts global first-half 2026 humanoid shipments above 22,000 units, up roughly 300 percent year over year, with AgiBot holding the top spot at about 43 percent share (roughly 15,000 cumulative units shipped as of mid-2026) and Unitree second at about 31 percent, the same Chinese one-two that has led every count taken so far in 2026. XPeng's robotics arm is spending the $900 million-plus it raised on August 24 pushing the IRON humanoid toward mass production by the end of 2026 ahead of a first-quarter-2027 showroom deployment, Li Auto's 'Nexus' program is aiming to ship a first wheeled factory robot within 2026 ahead of a later bipedal humanoid, and BYD's Xiao Di service robot keeps working showroom floors in Zhengzhou, Shenzhen and Shanghai toward EVP Stella Li's goal of two to three units per Di Space location within one to two years. On the software side, Google DeepMind's Gemini Robotics 2 whole-body model remains demonstrated on rival Apptronik's Apollo 2 rather than on Atlas, still the clearest sign that Boston Dynamics' own AI partner is spreading its flagship model across more than one humanoid body.
What decides it: two different bets on closing the same generalization gap, and a labor question that gets its formal answer this week
The engineering gate has not moved: autonomous, long-horizon manipulation remains the hardest unsolved problem, and China's vertically integrated component base still lets Unitree and AgiBot undercut Western prices by an order of magnitude. What is new this cycle is two concrete, opposite bets on closing that gap. Tesla's is real production reps, even slow and internal-only ones: getting an actual robot line running at Fremont, however far from commercial volume, generates the operational data and manufacturing learning a slide deck cannot. Figure's is crowdsourced human data at consumer-app scale, betting that 16 million smartphone videos from 108 countries can feed Helix's generalization faster than any single company's own fleet could. Neither approach has yet produced evidence it actually shortens the timeline Wang Xingxing pushed out to '2-3 years,' and Unitree's own revenue mix, still skewing overwhelmingly toward research and education customers rather than industrial deployment, is the clearest sign the IPO frenzy priced in commercialization the technology has not yet delivered.
The labor question gets its formal answer within this update's own window: Korean Metal Workers' Union members vote August 31 on the tentative wage deal that ended Hyundai's first full-day walkout in a decade, and ratification would close out the immediate strike risk without resolving the underlying tension the union raised in July, a demand for a labor-management agreement before any robot enters a Hyundai workplace. That precedent matters well beyond one automaker as BYD's Xiao Di and Li Auto's Nexus robots move from labs into retail floors and factories of their own, and Hyundai's own concurrent move, putting a specific 25,000-unit committed order and a 30,000-unit annual capacity target on the record at its Investor Day even as the strike was being resolved, shows the company treating the deployment plan and the labor talks as genuinely separable questions rather than waiting for one to settle before advancing the other.
The money and the rules: Figure's $1 billion data bet adds to a record funding year, and Washington's bill still hasn't moved
Private capital has not slowed down to wait for Unitree's stock to stabilize, and Figure's new commitment is itself a data point: more than $1 billion earmarked for data and compute over just the next 12 months is a bet that the sector's remaining bottleneck is training data, not capital availability. That commitment lands inside a broader wave that shows no sign of cooling: humanoid-specific startups have raised roughly $8.6 billion so far in 2026, already 1.8 times all of 2025, inside a $18.8 billion robotics-funding year against $15 billion for all of 2025. China's own embodied-intelligence sector logged 93.5 billion yuan of financing in the first half of 2026 alone, a fivefold jump year over year, while regulators recorded 116,000 newly registered humanoid-robot enterprises over the same period, evidence that private and state capital are still betting on scale even as the sector's first public shareholders are still sitting on a 45 percent loss from peak.
China's industrial policy did not change direction: the Ministry of Industry and Information Technology's 2026 production target of more than 100,000 humanoid units, a fivefold jump from roughly 20,000 in 2025, stands alongside a national initiative for at least 10,000 units in actual commercial use by year-end, and AgiBot's own disclosed 15,000 cumulative shipments are the clearest evidence any single company is tracking toward that floor. Washington's side stayed just as static: the bipartisan GUARD Act (H.R. 9129), introduced June 3 to place adversary humanoid and quadruped robots on the FCC's Covered List after a one-year national-security review, remains parked in the House Energy and Commerce Committee with no markup scheduled as of this update, so the Pentagon's Section 1260H list and the FCC's July 28 Covered List addition, not new legislation, remain the only binding restrictions on Chinese humanoid hardware reaching the US market.
What to watch through 2027
Watch the Korean Metal Workers' Union's August 31 ratification vote either lock in the tentative Hyundai wage deal or send the dispute back to the table, and separately watch Hyundai name the site for its 30,000-unit-a-year robot factory by year-end. Watch Optimus move past its deliberately 'flat and long' start at Fremont toward real volume, and watch whether Tesla finally sets a date for the Optimus V3 public reveal. Watch whether Figure converts Index's crowdsourced video into disclosed Helix generalization gains, the test of whether $1 billion of data spending actually buys a capability edge. Watch whether Unitree's stock finds a floor above its 150.80 yuan IPO price or keeps sliding as more analysts weigh in on China's humanoid-valuation bubble debate, and watch whether AgiBot finally converts its Hong Kong process into an actual priced listing at its targeted $5.1-6.4 billion range, still unpriced as of this update.
Watch Agility Robotics' SPAC merger clear shareholder and regulatory approval toward its 2026 close and AGLT begin trading, and watch whether Boston Dynamics names a permanent CEO after Chung's larger personal stake fueled Nasdaq IPO speculation. Watch XPeng's newly funded IRON push toward mass production by the end of 2026 ahead of its first-quarter-2027 showroom deployment, watch Li Auto's first Nexus wheeled robot actually ship within 2026, and watch BYD's Xiao Di expand from its initial Di Space cities toward Stella Li's two-to-three-robots-per-store goal. Watch whether full-year 2026 global humanoid shipments close in on the roughly 22,000-units-in-1H26 pace doubling in the second half, whether AgiBot's 15,000-cumulative-unit count keeps widening its lead over Unitree, and whether the GUARD Act finally gets a committee markup after sitting untouched since its June introduction.
Sources
- The Motley Fool - Optimus just entered production at Fremont
- Forbes - Figure launches gig platform to get humans to do work to train robots, will spend $1 billion
- Hyundai Newsroom - Hyundai Motor Group CEO Investor Day 2026
- The Next Web - Unitree shares slump 45% raising robot bubble fears
- The Standard - Global humanoid robot shipments surge 300pc in H1, led by AgiBot
- The Korea Herald - Hyundai Motor's union reaches tentative wage deal, averting more strikes
- CGTN - China's output of humanoid robots set to exceed 100,000 in 2026
