The Physical AI Market in 2026: Players, Capital, and the Labor-Market Question Everyone Asks

Ask ten people what the “physical AI market” is worth in 2026 and you will get ten different numbers, because they are not measuring the same thing. Some forecasts price the humanoid-robot hardware business alone. Others price a much broader category that includes industrial automation, sensors, and AI infrastructure bolted onto physical systems. The honest short answer: humanoid robots are a multi-billion-dollar market today that most credible forecasters expect to become a multi-ten-billion-dollar market by the early 2030s, funded by a small number of very large private rounds and a wave of public listings, while economists remain genuinely split on whether the net effect on jobs will be positive or brutal. Everything below unpacks that answer with the specific numbers, names, and dates that back it up — and flags where the sources disagree rather than picking a winner.

What do people actually mean by “physical AI market”?

This is the first place forecasts go wrong, and it’s worth settling before quoting a single dollar figure. “Physical AI” is a broad label for AI systems that perceive and act in the physical world — a category that spans humanoid robots, but also warehouse automation, autonomous vehicles, industrial arms with vision models bolted on, and the sensor/compute stack underneath all of it. If you want the fuller definition and how it relates to embodied AI more generally, that’s covered in what is physical AI.

Market-size estimates for that broad category diverge enormously depending on scope. One widely cited estimate puts the physical AI market at roughly $1.5 billion in 2026 growing to about $15.24 billion by 2032. A different estimate, using a different methodology, puts 2026 at roughly $7.11 billion growing to $34.89 billion by 2031. A third, broader estimate — pulled from a wider set of adjacent categories — puts the total physical AI market above $430 billion by 2030. These numbers are not typos or rounding errors; they reflect genuinely different definitions of what counts as “physical AI,” and none of them should be treated as interchangeable with each other, let alone with the humanoid-specific forecasts discussed below.

That distinction matters because the two numbers that get quoted most often in 2026 — Goldman Sachs’ $38 billion by 2035 and Morgan Stanley’s $15 billion by 2030 — are both humanoid-robot-only forecasts, not physical-AI-market forecasts. Conflating the two is the single most common error in market commentary right now, and it’s worth watching for when you read another outlet’s numbers.

How big will the humanoid robot market actually get?

Goldman Sachs’ most-cited projection puts the global humanoid robot total addressable market at $38 billion by 2035, with unit shipments reaching 1.4 million. That’s a sixfold increase in market size and a fourfold increase in shipment volume compared to the bank’s own prior forecast, and the revision was driven mainly by two things: AI capability progressing faster than expected, and manufacturing costs falling faster than expected.

China alone is now central enough to this story that Morgan Stanley tracks it as a near-standalone market. In a note dated June 24, 2026, the bank doubled its 2026 shipment forecast for Chinese humanoid robots to 50,000 units — itself already a doubling from a prior estimate of 28,000, which had in turn already doubled from a January 2026 estimate of 14,000 units. That’s three upward revisions within roughly six months, which tells you as much about how fast sentiment is moving as it does about the actual hardware. Morgan Stanley now projects the China market growing to 446,000 annual units and a $15 billion market size by 2030, up from an estimated $2 billion in 2026.

Read those two numbers side by side and the picture is coherent even if the underlying definitions differ slightly: humanoid robotics is being priced by two of the largest banks in the world as a market that roughly doubles or triples in size every year or two through the end of the decade, off a small base. The base is what makes these percentages misleading if taken alone — $2 billion in China in 2026 is a rounding error next to global semiconductor revenue, and the entire premise of these forecasts is that the curve bends upward hard from here, not that it already has.

Who are the players actually shipping hardware in 2026?

The market has consolidated around a handful of companies that are either shipping product, about to ship product, or have just closed the capital needed to try.

Figure AI raised the largest and most closely watched private round in the sector: a Series C exceeding $1 billion, closed in September 2025, at a $39 billion post-money valuation. The round was led by Parkway Venture Capital with participation from Brookfield, NVIDIA, Intel Capital, Salesforce, Qualcomm Ventures, and others — a roster that mixes infrastructure capital, strategic tech investors, and enterprise customers in one cap table. That $39 billion valuation represents a 15x step-up in roughly eighteen months from Figure’s own $2.6 billion Series B, closed in February 2024.

Unitree Robotics took a different path: instead of raising private mega-rounds, it went to the public market. China’s securities regulator, the CSRC, approved Unitree’s IPO registration on Shanghai’s STAR Market on July 3, 2026. Unitree plans to raise roughly 4.2 billion yuan (about $619 million) in the offering. The implied total company valuation is reported inconsistently across outlets — some put it around $6.2 billion, others closer to $5.9 billion (40 billion yuan) — likely reflecting different assumptions about the exact free-float percentage being offered. The raise amount itself is consistent across sources even where the valuation isn’t.

Agility Robotics chose a third route: a SPAC merger. The company is going public via a $2.5 billion combination with Churchill Capital Corp XI, announced in the first days of July 2026. Unlike some of its peers, Agility already has a commercial deployment record to point to: its Digit humanoid is running at Schaeffler, GXO, Toyota Motor Manufacturing Canada, and Mercado Libre facilities, and has logged more than 65,000 operating hours across nine customer sites. At a GXO warehouse in 2025, Digit moved over 100,000 totes — work that earned it an RBR50 Robot of the Year award.

Tesla is the wildcard because Optimus is not a standalone robotics company raising its own capital — it’s a program inside a much larger balance sheet, and Elon Musk has treated its production timeline as a moving target for several years running. The current plan is to begin Optimus V3 production in summer 2026 by converting part of the Fremont Model S/X assembly line. Musk’s stated target is a per-unit price below $20,000 at scale, but the current estimated manufacturing cost is still in the $50,000–$100,000 range, and high-volume output — tens of thousands of units — isn’t expected until 2027. Some outlets have floated 2026 unit-volume figures in the 50,000–100,000 range as aspirational targets, but Musk’s own March 2026 comments describe only a summer production start with initial output he called “quite slow.” Treat any specific 2026 Optimus volume number you see as unconfirmed rather than a committed figure.

1X Technologies is targeting a different buyer entirely: the home, not the warehouse. Its NEO humanoid is available for Early Access preorder at $20,000 (with a $200 deposit) or via a $499/month subscription, with first U.S. consumer deliveries beginning in 2026. NEO is a lighter-duty machine than its industrial peers — it weighs 66 lb (29.9 kg), can lift 154 lb (69.8 kg), and carry 55 lb (24.9 kg) — built around household chores rather than factory throughput. For a broader look at how these different robot bodies compare and where the category is heading, see humanoid robots explained.

Player-by-player snapshot

Company	Latest move	Valuation / raise	What it ships
Figure AI	Series C, Sept. 2025	$39B post-money, >$1B raised	Humanoid, industrial focus, not yet at consumer scale
Unitree Robotics	STAR Market IPO approved, July 3, 2026	~$619M raise; ~$5.9–6.2B implied valuation	G1 ($16,000–$73,900), R1 (from $4,290–$5,900)
Agility Robotics	SPAC merger w/ Churchill Capital Corp XI, July 2026	$2.5B deal value	Digit — deployed at Schaeffler, GXO, Toyota, Mercado Libre
Tesla	Optimus V3 production start, summer 2026	Internal program, no standalone raise	Target <$20,000; current cost ~$50,000–$100,000
1X Technologies	NEO preorders open, deliveries 2026	Private, preorder revenue	NEO — $20,000 early access or $499/month

The spread in the table is the story in miniature: one company chose IPO, one chose SPAC, one chose a mega private round, one is self-funding inside a public company’s existing balance sheet, and one is trying to sell direct-to-consumer before it has industrial-scale proof. There is no single playbook yet for how a humanoid robotics company gets to scale — which is itself a signal that the market is still early.

How much capital is actually moving into robotics right now?

Strip away the headline valuations and look at the raise mechanics, and 2025–2026 reads as the period when robotics capital diversified beyond venture rounds into public markets. Figure’s round was private and venture-led. Unitree’s is a regulator-approved public listing on a Chinese exchange. Agility’s is a SPAC merger, a structure that fell out of favor after the 2021 SPAC boom cooled but that clearly still works for capital-intensive hardware companies that want public-market liquidity without a traditional IPO roadshow.

That diversification matters for anyone trying to read the sector, because each funding path implies different disclosure, different investor bases, and different pressure to hit near-term revenue numbers. A SPAC merger and an IPO both mean public financial filings going forward — Agility’s Digit operating-hours and customer-deployment numbers, for instance, are already public precisely because of the SPAC disclosure process. Private mega-rounds like Figure’s, by contrast, disclose only what the company chooses to announce. If you’re tracking the sector closely, the shift toward public listings is worth watching specifically because it should mean more verifiable operating data becomes available over the next year, not less. For a fuller rundown of the specific rounds, dates, and investors across the sector, see robotics funding rounds 2025-2026.

For anyone thinking about exposure to this capital wave without picking single-company winners, the public-market listings — and the ETFs that are starting to bundle robotics and automation names together — are one way in. That comes with its own tradeoffs around concentration and timing, covered in robotics ETFs and investing.

What does a humanoid robot actually cost to buy in 2026?

Pricing is where the market stops being abstract and starts being a purchase order. The spread across current offerings is wide, and it roughly tracks payload capacity and intended duty cycle.

Robot	Configuration	Price
Unitree R1 (dual-arm modular)	Entry	from $4,290
Unitree R1	Base	$5,900
1X NEO	Early Access preorder	$20,000 (or $499/month)
Unitree G1	Base	$16,000
Unitree G1	EDU Ultimate	$73,900
Tesla Optimus V3	Target price at scale (unconfirmed)	<$20,000
Tesla Optimus	Current estimated manufacturing cost	$50,000–$100,000

Two things stand out. First, Unitree has effectively built a price ladder rather than a single product — from a sub-$5,000 lightweight unit up to a nearly $74,000 fully-equipped research configuration — which lets it serve hobbyists, researchers, and light-commercial buyers off a single platform. Second, Tesla’s situation illustrates the gap between an aspirational target price and the actual cost of production today: Musk’s sub-$20,000 figure is a scale target, not a current sticker price, and the real manufacturing cost right now sits two to five times higher. Anyone modeling near-term unit economics for humanoid robots should be using the $50,000–$100,000 range for anything shipping in 2026, not the long-run target.

Will robots actually take people’s jobs?

This is the question that shows up in every comment section under every humanoid-robot headline, and the honest answer is that credible institutions disagree by an order of magnitude depending on what exactly they’re measuring — so it’s worth being precise about which number you’re citing.

The World Economic Forum’s Future of Jobs Report 2025, published January 8, 2025, and based on data from over 1,000 companies across 22 industries and 55 economies, forecasts 170 million new jobs created and 92 million jobs displaced globally by 2030 — a net gain of 78 million jobs. That is a net employment forecast: it explicitly nets creation against displacement and lands positive.

The McKinsey Global Institute’s frequently cited analysis tells a very different-sounding story: automation could displace 400 million to 800 million workers worldwide by 2030, depending on the pace of adoption, with 75 million to 375 million of those workers potentially needing to switch occupational categories entirely and learn new skills. That figure is not a net employment forecast — it’s a gross displacement/exposure estimate under an aggressive-adoption scenario, and it says nothing on its own about how many new jobs get created alongside that displacement. It also originates from a 2017 report, though it remains the most-quoted number in labor-market commentary about automation nearly a decade later.

Put the two side by side honestly: WEF says more jobs get created than destroyed net, by 2030, using company-survey data on job creation. McKinsey says gross displacement exposure — how many workers could be affected in some way — could reach the hundreds of millions under fast adoption, without weighing in on how many replacement jobs appear. These are two different questions wearing the same “will robots take jobs” headline, and treating either one as a complete answer to the other’s question is where most bad reporting on this topic comes from. If you want the plain-language grounding for terms like “displacement,” “reskilling,” or “task automation” that get thrown around in this debate, the glossary is a good place to check definitions before repeating a claim.

What both reports agree on, implicitly, is the mechanism: physical AI and robotics are expected to change which tasks a human does, not simply remove humans from a workplace wholesale, at least on the timeline through 2030. The scale of that change — net positive employment story or net painful disruption story — depends heavily on adoption speed, retraining investment, and which industries move first, none of which either report can fully predict from today’s vantage point.

What’s actually holding the market back — or accelerating it?

Two forces are doing most of the work in either direction right now, and neither is a technology story.

Regulation is one. As humanoid and mobile robots move from research labs and controlled warehouses into spaces shared with untrained members of the public, machinery safety and AI-system regulation both apply simultaneously, and in the EU that overlap is not yet fully settled in practice. How the EU AI Act interacts with existing machinery regulation for robots is a live compliance question for any company selling into that market, and it’s covered in more depth in EU AI Act and machinery regulation for robots. Companies planning multi-market rollouts in 2026 and beyond are building regulatory review into their timelines earlier than they would have three years ago, precisely because the rules are still moving.

Capital structure is the other. As the funding table above shows, 2026 is the year robotics financing genuinely diversified — venture mega-rounds, SPAC mergers, and direct exchange IPOs are now all live simultaneously in the same sector, sometimes for direct competitors. That diversification is generally a sign of a maturing market: it means more than one type of capital believes the sector is investable, and it means more disclosure is coming as more of these companies become public reporting entities. The full landscape of that market, investment, and regulation cluster — including the pieces above — sits under the Market, Investment & Regulation hub.

So what should you actually take away from all this?

If you’re evaluating the physical AI market in 2026, three things hold up no matter which headline number you’ve seen. First, be precise about which market you’re citing — humanoid-robot-specific forecasts like Goldman’s $38 billion by 2035 or Morgan Stanley’s China-specific $15 billion by 2030 are not the same category as broader “physical AI market” estimates that range from roughly $15 billion to over $430 billion by the early 2030s depending on scope. Second, the capital story is now genuinely diverse rather than concentrated in one financing mechanism — private mega-rounds, SPAC mergers, and direct IPOs are all live in the same year, for companies with very different commercialization strategies, from Figure’s industrial-focused private raise to Unitree’s public listing to Agility’s SPAC merger backed by real deployment hours. Third, on labor market impact, resist any single number presented as the final word — the WEF’s net-positive jobs forecast and McKinsey’s gross-displacement exposure figure are answering different questions, and a defensible view of 2026 has to hold both alongside the reminder that even the most aggressive hardware production plans, including Tesla’s, are running behind their own announced timelines. The market is real, the capital is real, and the uncertainty about who wins and what it means for work is real too — which is exactly why it’s worth tracking closely rather than settling on whichever headline number is easiest to repeat.