Family offices are moving capital into Physical AI before mass humanoid adoption has been proven. The thesis is larger than robots that look human: it spans autonomous machines, semiconductors, simulation, industrial automation, defence, power and eventually the financing of machine fleets.
For wealthy investors, the opportunity is compelling precisely because the commercial and valuation risks remain unresolved.
Capital is arriving before certainty. In Q1 2026, a record $16.3 billion was invested across 492 robotics and Physical AI venture deals, according to PitchBook. UBS also found that 65% of 307 surveyed family offices were already invested across the AI value chain. Capital is broadening from software models and data-centre infrastructure towards machines that can perceive, reason and act.
Physical AI demands a higher commercial threshold than generative AI. A flawed paragraph is undesirable. A flawed robot trajectory in a factory, warehouse, hospital or on a public road can cause physical damage. Reliability must therefore survive physics, latency, mechanical wear, power limits and safety constraints.
Family offices investing in Physical AI are underwriting the possibility that artificial intelligence becomes productive physical labour. Many also control operating businesses, property, logistics assets or industrial facilities that can provide real environments for deployment.
The investment question is whether reliability, utilisation, manufacturing yield and labour-substitution economics can improve quickly enough to justify valuations already discounting extraordinary scale.
Executive Summary
- Physical AI is broader than humanoids and already includes commercially deployed autonomy, industrial robotics and machine intelligence.
- Q1 2026 robotics and Physical AI venture funding reached $16.3 billion across 492 deals.
- UBS found 65% of surveyed family offices were invested across the AI value chain, while 77% retained an active operating business.
- Autonomous mobility and industrial robotics currently provide stronger commercial evidence than general-purpose humanoids.
- US frontier AI strength is increasingly confronting China’s manufacturing and robotics scale.
- Private-market valuations have accelerated faster than verified mass deployment, raising material capital-loss risk.
- For UHNW portfolios, Physical AI is a cross-asset exposure spanning private markets, public equities, infrastructure and eventually credit.
What Physical AI Actually Means for Investors
Physical AI combines five economic layers: intelligence, perception, decision-making, embodiment and infrastructure. Foundation models, vision-language-action systems and world models interpret objectives and environments. Cameras, LiDAR, radar and tactile sensors translate surroundings into machine-readable states. Planning systems determine the next action. Motors, actuators, gears and manipulators execute it. GPUs, simulation platforms, digital twins and edge processors train and operate the system.
The distinction from conventional automation is adaptability. Traditional robots perform engineered routines inside known environments. Physical AI seeks to cope with variation.
That raises the commercial bar. Machines face changed lighting, deformed packages, unexpected humans, slippery surfaces and mechanical wear. These long-tail conditions make task success rate, useful operating hours, availability and economic value per task more important than benchmark performance alone.
For investors examining embodied AI investment opportunities, this is the first discipline: separate intelligence from productive machine labour. The former is increasingly visible. The latter still depends on reliability, supervision, maintenance and economics.
Why Physical AI Is Reaching an Investment Inflection Point
Several technologies are converging. Multimodal models allow machines to combine visual information, natural-language instructions and action prediction. That shifts robotics from programming every movement towards training policies that may adapt across related tasks.
Simulation is equally important. Physical training data are expensive because laboratories cannot reproduce every warehouse layout, road condition or object interaction, and broken robots create real costs. Digital twins, synthetic data and reinforcement-learning environments can compress development cycles by allowing machines to practise in virtual environments before deployment.
Nvidia’s Isaac Sim and Isaac Lab illustrate this model. Its Cosmos world models, GR00T robot foundation models and Jetson edge systems extend the loop from simulated training to local inference. Edge AI matters because a physical machine cannot always wait for a cloud round trip. Latency, connectivity, privacy and safety increasingly require intelligence to operate close to the machine.
Hardware economics are also moving, although the evidence remains less standardised than software performance data. Goldman Sachs estimated in 2024 that humanoid manufacturing costs had fallen from roughly $50,000 to $250,000 per unit to $30,000 to $150,000, citing component availability, manufacturing improvements and competition. This was an analyst estimate, not an audited industry cost series.
Labour scarcity, ageing populations, manufacturing reshoring and defence modernisation strengthen the adoption case. The technological inflection is increasingly credible, but commercial inflection remains selective and a mass humanoid inflection has not yet been demonstrated.
Why Family Offices Are Moving Into Physical AI
UBS’s 2026 survey covered 307 family offices, with average family-office assets of $1.3 billion, average family net worth of $2.7 billion and aggregate family wealth of $627.4 billion. Some 65% were invested across the AI value chain. The survey also found 37% exposed to power and resources linked to the broader AI theme, 37% to infrastructure and 33% to AI-enabled healthcare. Automation and robotics were cited by 38% of Swiss family offices and 44% of Southeast Asian respondents.
Family capital can theoretically hold through several technology cycles without conventional venture fund-life pressure. That matters where proprietary hardware, manufacturing, field testing, safety validation and maintenance networks require repeated capital before unit economics stabilise.
The more distinctive advantage is operational. Seventy-seven per cent of surveyed families still owned an active operating business. Warehouses, factories, hospitals, hotels, farms or property portfolios can provide both deployment environments and physical training data.
Family offices can therefore combine direct investment, co-investment, listed securities, infrastructure and eventually credit. Yet permanent capital is not informational immunity. Weak governance, prestige-driven allocation, illiquidity, concentration, dilution and mark-to-model valuations remain real risks. Billionaire participation validates interest, not price.
Following the Billionaire Capital Into Robotics
Family-office participation is increasingly visible, although individual cheque sizes are often undisclosed.
Bernard Arnault-backed Aglaé Ventures participated in Humanoid’s $152 million Series A in July 2026 at a $1.35 billion post-money valuation. Bezos Expeditions joined Generalist’s $400 million June 2026 financing. Jeff Bezos was also reported as an investor in Physical Intelligence’s $400 million November 2024 round at an approximately $2 billion valuation, and Figure AI’s $675 million 2024 round at a $2.6 billion valuation. None of those round sizes should be treated as the investor’s own cheque.
The Friedkin Group participated in RobCo’s $100 million Series C, while Exor-owned Lingotto co-led it. Alpha Square Group participated in Atoms’ $1.7 billion debt and equity financing, and MGFO joined Vitestro’s $70 million Series B.
Bloomberg also reported that Azim Premji’s family office was in talks regarding another robotics financing. Those are reported talks, not a completed investment.
The pattern spans humanoids, industrial robotics, foundation models and healthcare automation. Disclosure remains incomplete, and undisclosed participation or cheque sizes should never be inferred.
Physical AI Is Already Bigger Than the Humanoid Robot Story
Industrial robotics provides the clearest installed base. 542,000 industrial robots were installed globally in 2024, taking operational stock to roughly 4.7 million. China alone installed 295,000 units, or 54% of the global total.
Waymo surpassed 14 million fully autonomous rides during 2025, while Aurora launched commercial driverless trucking between Dallas and Houston in 2025. These constrained domains demonstrate that Physical AI can generate commercial activity without taking human form.
Warehouse automation, defence autonomy and healthcare robotics sit at different points on the maturity curve, while Generalist, Physical Intelligence and Skild AI are developing intelligence intended to transfer across machines and tasks.
For ultra high net worth investors, this makes Physical AI broader than a humanoid bet. Structured environments can commercialise earlier because operating domains are narrower, supervision is easier to design and economic value is easier to measure.
Nvidia and the Infrastructure Stack Behind Physical AI
Nvidia is positioning itself across the Physical AI development loop: data, simulation, training, evaluation, edge deployment, real-world feedback and retraining. Omniverse supplies digital environments. Isaac Sim supports robot simulation and synthetic data. Isaac Lab supports reinforcement-learning workflows. Cosmos provides world-model infrastructure. GR00T targets robot foundation models. Jetson extends inference towards autonomous machines.
The Nvidia Physical AI investment thesis therefore extends beyond chips inside robots. A large machine population could consume compute during training, simulation, synthetic-data generation, inference and retraining.
Financially, the theme remains small relative to data centres. In the referenced quarter, Nvidia’s combined Automotive and Robotics revenue was $586 million versus $46.7 billion of total revenue. Nvidia later reported $215.9 billion of FY2026 revenue, with data centres dominating the mix. Physical AI was not an independent reporting segment.
Robotics is therefore a strategic option on future compute demand, not yet a comparable earnings pillar. Qualcomm, Arm-based processors, custom ASICs, automotive systems-on-chip and lower-cost inference architectures can also compete where battery life, latency and power efficiency matter more than maximum throughput.
The Humanoid Robot Investment Thesis and Its Valuation Problem
Humanoids attract capital because the physical economy is already designed around human geometry. A biped with two arms can theoretically climb stairs, pass through existing doorways, operate human workstations and use tools without requiring every environment to be redesigned around a specialised machine.
Current deployments remain narrower than the narrative. Figure has placed robots in BMW environments and says its Figure 02 programme contributed to production associated with 30,000 BMW vehicles, which does not mean BMW deployed 30,000 humanoids. Figure’s first-generation BotQ target is up to 12,000 humanoids per year, with an ambition to manufacture 100,000 robots over four years.
Hyundai and Boston Dynamics target manufacturing capacity of 30,000 robot units annually by 2028. These are targets, not achieved production. Reuters, by contrast, reported that Unitree had delivered about 18,000 bipedal humanoids across models by July 2026.
Forecast dispersion shows how unresolved the economics remain.
| Institutional view | Time horizon | Humanoid market outlook |
| Goldman Sachs | 2035 | $38 billion market and about 1.4 million shipments |
| Barclays | 2035 | Current market estimated around $2 billion to $3 billion, with an optimistic scenario approaching $200 billion |
| Morgan Stanley | 2050 | More than $5 trillion, including related supply chains and services, with close to 1 billion units operating |
These are not a consensus. They use different definitions, assumptions and horizons.
That matters because private valuations already assume substantial future success. Figure AI moved from a $2.6 billion valuation in 2024 to $39 billion in 2025 after raising more than $1 billion in its Series C. The increase shifts underwriting from whether the technology can work to whether the company can capture a very large share of future machine labour.
For private investors, humanoid robot private market valuations therefore contain two risks at once: technological execution and expectations already capitalised into price.
When Does a Robot Become Cheaper Than Human Labour?
The wrong comparison is robot purchase price versus annual salary. The relevant equation is:
productive hours × task-equivalent labour value × productivity, less maintenance, energy, supervision, downtime, insurance, financing and integration.
Installation, safety systems, useful life and residual value all matter. Reliability determines whether nominal operating hours become productive hours, while the supervision ratio can decide whether labour leverage exists at all. Mechanical wear adds battery, actuator, joint and bearing costs that software does not face.
Robotics-as-a-Service can shift customer capex into opex while moving financing and residual-value risk towards vendors or capital providers. Agility Robotics’ Digit has historically been associated with an indicative $30-per-hour service price in a GXO deployment. That is an example, not an industry-wide benchmark.
The dossier’s Bancara analytical scenario illustrates the sensitivity. At $45 per robot hour with low utilisation, replacing labour may remain unattractive. At $30 with higher utilisation, economics become more competitive. At $22 with high utilisation, the case strengthens. At $15 with high reliability and long operating hours, substitution incentives become materially stronger. These figures are analytical scenarios, not market forecasts.
Physical AI Could Become a Global Productivity Shock
The macroeconomic prize is not robotics revenue. It is higher output per unit of labour and capital.
Manufacturing, warehousing, logistics, mining, agriculture, healthcare, construction, hospitality and defence all contain repetitive, dangerous or labour-constrained tasks. If machines can extend operating hours, standardise processes and reduce downtime, successful adopters could expand throughput while lowering unit labour intensity.
The distributional effects are less benign. Historical NBER work on industrial robots found that automation reduced employment and wages in affected US local labour markets. Research on French manufacturing found productivity gains alongside a lower labour share of value added and a smaller proportion of production workers.
At the same time, automation creates demand for technicians, engineers, systems integrators and maintenance specialists. US Bureau of Labor Statistics data put median pay for industrial machinery mechanics at $63,510 in May 2024, with employment projected to grow 13% from 2024 to 2034.
The inflation path could therefore be two-stage: capex inflation first as machines require semiconductors, power, factories and components, followed later by productivity disinflation if deployment reduces unit labour costs.
The US-China Race for Embodied Intelligence
Physical AI may develop into a different geopolitical contest from generative AI because model quality is only one part of the system.
The United States has deep advantages in frontier AI models, accelerated computing, venture capital and software. China has extraordinary manufacturing density across industrial robotics, motors, batteries, electronics and low-cost component supply chains.
The installed base already illustrates the gap. China installed 295,000 industrial robots in 2024, representing 54% of global installations, and its operational stock exceeded 2 million robots. Domestic Chinese robot manufacturers captured 57% of their home market. Preliminary IFR figures placed US installations at approximately 38,000 in 2025.
Reuters reported more than $20 billion of Chinese humanoid-related initiatives over the preceding year. Beijing was also establishing a 1 trillion yuan fund, roughly $137 billion in the cited context, spanning AI, robotics and other advanced technologies, not humanoids alone.
By July 2026, US restrictions had expanded towards Chinese humanoid and quadruped robot imports on national-security grounds. Unitree’s August 2026 Shanghai debut then produced a first-day market capitalisation around $50 billion, according to Reuters, highlighting both manufacturing scarcity and valuation risk.
For the US-China competition in humanoid robotics, the key question is increasingly whether superior intelligence can be manufactured at sufficient scale, reliability and cost. A brilliant model attached to expensive or fragile hardware may lose economically to a slightly weaker system embedded in a cheaper, serviceable machine.
The Picks and Shovels Behind Physical AI
The strongest Physical AI supply chain investment opportunities may sit in bottlenecks rather than robot brands.
Compute includes GPUs, AI accelerators, memory, networking and edge processors. Perception depends on cameras, LiDAR, radar, depth sensing, tactile systems, force sensors and encoders. Motion requires servo motors, actuators, harmonic reducers, gears, bearings and dexterous hands. Power introduces batteries, battery-management systems, thermal systems and power semiconductors.
Software adds another layer through simulation, digital twins, fleet orchestration and industrial cybersecurity.
Rent capture is most defensible where technical complexity, scarcity, switching costs, regulatory qualification and recurring revenue overlap. The assembler does not automatically own the highest-margin layer.
Materials such as copper, rare-earth permanent magnets, aluminium and battery inputs can benefit if machine volumes become sufficiently large. Yet present robotics demand remains much smaller than electric vehicles, grids and data-centre infrastructure. Commodity effects should remain conditional on deployment scale, not inferred from venture funding headlines.
Private Markets Are Pricing the Future Before Revenue Arrives
The private-market cycle is moving faster than mass commercial deployment. PitchBook’s $16.3 billion across 492 robotics and Physical AI deals in Q1 2026 was a record quarter.
Several valuations illustrate the intensity. Figure AI reached $39 billion in September 2025 after raising more than $1 billion. Skild AI raised $1.4 billion at a valuation above $14 billion in January 2026. Humanoid raised $152 million at a $1.35 billion post-money valuation in July 2026. Physical Intelligence raised $400 million at an approximately $2 billion valuation in November 2024. FieldAI raised $314 million at approximately $2 billion. Shield AI raised $240 million at $5.3 billion. Anduril raised $2.5 billion at a $30.5 billion valuation.
These businesses span robotics, autonomy and defence, but the capital pattern is clear. Hardware companies require inventory, factories, tooling, warranties, working capital and service networks. Repeated raises can dilute earlier investors, while down rounds may be delayed and headline private valuations need not represent continuously clearing prices.
Exit routes include strategic acquisitions, IPOs, secondary sales and recapitalisations. Unitree’s listing shows that public markets can reward robotics scarcity, but scarcity can also produce extreme multiples. A transformative technology can still destroy capital for investors who enter at excessive valuations.
How Physical AI Could Reprice Public Markets
Public-market transmission begins with enabling infrastructure. Semiconductors can benefit through training, simulation, automotive compute, edge inference, memory and connectivity. Industrial automation companies bring installed customer relationships, safety expertise, service networks and systems-integration capabilities.
Automotive groups occupy several roles at once: manufacturers, early adopters, data sources and high-volume production platforms. Defence autonomy benefits from procurement demand that is already commercial rather than dependent on household adoption. Logistics can gain from autonomous forklifts, mobile robots, manipulation and driverless trucking. Healthcare can extend from established robotic surgery towards repetitive clinical workflows.
The second-order effect is potentially larger. Successful adopters may expand margins or throughput by reducing labour intensity. Companies whose competitive advantage depends primarily on inexpensive labour, manual fulfilment or staffing scale could face pressure if automation compresses that moat.
Power infrastructure, cloud computing, manufacturing equipment and simulation can become indirect beneficiaries of the capex cycle. Mining may benefit both as an adopter of autonomous systems and, eventually, as a supplier of materials into a larger robotics build-out.
For sophisticated investors, this becomes a cross-asset monitoring problem. Bancara’s ecosystem, including BancaraX, MetaTrader 5 and TipRanks, can support monitoring of listed-market transmission across semiconductors, industrials, commodities, currencies and indices. It is not a claim of direct access to private robotics companies.
What Physical AI Means for UHNW and Family-Office Portfolios
Physical AI portfolio implications are easy to underestimate because different assets can look diversified while sharing the same underlying AI factor.
At the highest-risk layer sit early robotics companies, foundation-model developers and experimental autonomy. These are venture-like exposures with technical risk, dilution, illiquidity and uncertain exits.
Growth-stage companies shift the problem from “does the technology work?” towards “can this business manufacture, distribute, maintain and finance the system economically?” Execution, working capital and service infrastructure become as important as model capability.
Public enablers offer liquidity through semiconductors, automation, sensors, manufacturing platforms and selected infrastructure. Yet a family already holding megacap technology, semiconductor funds, private AI, data-centre assets and AI-linked credit may possess substantial hidden concentration despite owning different securities.
Infrastructure-like exposure can include power, electrical equipment, data centres, industrial property and specialist manufacturing. These assets carry less binary technology risk but remain exposed to capex cycles and overbuilding.
Private credit belongs later in the maturity curve. Once robotics fleets produce predictable contracted cash flows, measurable utilisation and dependable residual values, financing structures could begin to resemble equipment finance, vehicle fleets or leasing. Before that point, lenders are effectively underwriting technology and residual-value uncertainty alongside credit risk.
For entrepreneurial families, the most interesting exposure may sit inside businesses they already control. Warehouses, factories, hotels, mines or healthcare assets can benefit through higher throughput, longer operating hours and lower labour intensity. Capturing productivity gains inside an existing cash-generative business may in some cases be more attractive than paying a frontier venture valuation.
The discipline is look-through exposure across venture, growth equity, listed technology, semiconductors, power, industrial assets and operating companies.
What Could Break the Physical AI Investment Thesis?
The bear case begins with generalisation. A robot can succeed in a polished demonstration and still fail when lighting changes, flooring shifts, an object is damaged or a person interrupts the sequence. Long-tail physical environments may remain materially harder than digital tasks.
Reliability is equally unforgiving. A 99% task success rate still means one failure every 100 actions. In operations involving thousands of actions per day, intervention can remain frequent and expensive.
Human supervision can erase much of the labour thesis if machines require continuous teleoperation. Hardware costs may fall slowly. Batteries can constrain runtime. Mechanical wear can lift maintenance and warranty expense. Manufacturing yields may disappoint.
Safety, product liability and regulation become particularly demanding in transport, healthcare and defence. Cybersecurity is also physical safety when a compromised machine can move, lift, drive or manipulate.
Commercial risks compound the technical ones. Consumers and workers can resist adoption. Labour backlash can influence regulation. Chinese manufacturers may compress Western hardware margins through faster scale and lower component costs. Trade restrictions and semiconductor controls can fragment supply chains.
Finally, valuation can break the thesis even if the technology succeeds. Companies dependent on repeated capital injections can face down rounds or insolvency if funding conditions tighten. Technological progress does not eliminate financing risk.
Internet, Smartphones, EVs or Dot-Com? What History Says About Physical AI
Historical analogues all point to the same distinction: technological inevitability and investor returns are separate questions.
The internet transformed the economy, but many infrastructure companies still destroyed capital. The dot-com cycle showed that correctly forecasting adoption did not justify every valuation. Cloud computing demonstrated that infrastructure platforms can capture value before application economics are settled. Smartphones created enormous value across chips, operating systems, connectivity, manufacturing and app ecosystems, not only handset assembly.
Electric vehicles offer an especially relevant warning for humanoids. They combine enormous capital requirements, manufacturing execution, Chinese cost competition, price wars, startup valuations and eventual consolidation.
Railroads provide an older version of the same lesson. They transformed commerce while producing overcapacity, leverage and substantial investor losses.
Physical AI may follow a similarly uneven path. The largest economic value can emerge in enabling layers, while the most visible brands compete away returns. Social utility does not guarantee capital discipline.
Physical AI Through 2030 and 2035
The following ranges are Bancara analytical scenarios, not third-party consensus forecasts.
| Scenario | Humanoid shipments in 2030 | Humanoid shipments in 2035 | Investment character |
| Bear | 50,000 to 150,000 annually | 300,000 to 1 million annually | Specialised adoption, valuation compression, limited fleet finance |
| Base | 200,000 to 750,000 annually | 1 million to 4 million annually | Broad structured-environment adoption, consolidation around leaders |
| Bull | 1 million to 3 million annually | 5 million to 15 million annually | Accelerated labour substitution, major capex and edge-compute cycle |
In the bear case, conventional automation and specialised machines continue to improve, but general-purpose reliability stalls outside controlled settings. Hardware costs remain high, supervision requirements persist and venture valuations compress.
The base case assumes meaningful adoption in warehouses, manufacturing, transport, defence, mining and selected healthcare applications. Humanoids gain share where human-form flexibility carries genuine economic value, but they remain far from universal. Private-market consolidation becomes part of the maturation process.
The bull case requires several breakthroughs to arrive together: falling hardware costs, higher uptime, lower supervision, manufacturing scale and stronger real-world generalisation. Physical AI then becomes a broader labour-substitution cycle, increasing semiconductor, factory, power, sensor and machine-fleet capex.
These scenarios should not be blended with external forecasts. Goldman Sachs, Barclays and Morgan Stanley use different horizons and definitions. Scenario discipline is more useful than false precision.
The Indicators Sophisticated Investors Should Watch
The Physical AI market outlook through 2030 and 2035 will become clearer through operating evidence rather than fundraising headlines. The most useful dashboard includes:
- Commercial deployments and actual deliveries versus company production targets.
- Productive operating hours, task success rates and human-supervision ratios in real environments.
- Robot unit costs, maintenance expense, battery economics, actuator costs and sensor costs.
- Robotics-as-a-Service renewals, which reveal whether customers expand beyond pilots.
- Private funding volumes, valuation trends, IPO activity and secondary-market liquidity.
- Edge AI semiconductor revenue and industrial automation capex, indicating whether robotics is becoming financially material to suppliers.
- Labour costs and persistent shortages, which determine the substitution hurdle.
- US-China policy changes, including trade and semiconductor restrictions.
- Physical AI revenue disclosed by public companies rather than buried in broader categories.
The most important transition will be from press-release metrics to financial metrics: revenue, utilisation, gross margin, retention, service costs and productive fleet hours. When those disclosures become routine, the sector will have crossed a significant maturity threshold.
The Real Bet Is on Machine Labour
Family offices investing in Physical AI are underwriting a proposition larger than humanoid robotics. Autonomous vehicles, industrial robots, defence systems, warehouse automation and specialised healthcare machines already show that intelligence can create value when it moves from digital output into physical action.
Family-office interest is strategically rational. Permanent capital can absorb long development cycles. Flexible mandates can move across venture, growth equity, public markets, infrastructure and eventually credit. Operating businesses can provide deployment environments and may capture productivity gains directly.
None of that validates current prices.
The central tension is that capital formation is running ahead of mass-scale commercial proof in parts of the market. Figure AI’s rise from a $2.6 billion valuation to $39 billion, the scale of recent robotics funding and Unitree’s public-market scarcity premium show how aggressively investors are discounting future machine labour.
The eventual winners may sit across compute, simulation, edge processors, precision motion, industrial software, sensors, power infrastructure and financing rather than only among robot brands.
For UHNW investors and family offices, the durable framework is to separate technological direction from capital discipline. Physical AI can transform industrial economics while still disappointing investors who overpay or back the wrong layer.
The defining question is not whether machines become more intelligent. It is where economically productive machine labour ultimately generates durable returns on capital.
This material is provided for informational purposes only and does not constitute investment advice.
Works Cited
- https://www.bloomberg.com/news/newsletters/2026-08-05/why-family-offices-are-investing-their-billions-in-physical-ai-startups
- https://www.ubs.com/global/en/media/display-page-ndp/en-20260528-global-family-office-report-2026.html
- https://pitchbook.com/news/reports/q1-2026-robotics-physical-ai-vc-trends
- https://www.svb.com/trends-insights/reports/physical-ai-robotics/
- https://ifr.org/ifr-press-releases/news/global-robot-demand-in-factories-doubles-over-10-years
- https://ifr.org/ifr-press-releases/news/us-robot-industry-returns-to-double-digit-growth
- https://www.goldmansachs.com/insights/articles/the-global-market-for-robots-could-reach-38-billion-by-2035
- https://www.morganstanley.com/insights/articles/humanoid-robot-market-5-trillion-by-2050
- https://home.barclays/news/press-releases/20260/01/barclays-research-finds-humanoid-robotics-on-track-to-become-a–/
- https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Releases-New-Physical-AI-Models-as-Global-Partners-Unveil-Next-Generation-Robots/default.aspx
- https://www.reuters.com/business/robotics-startup-figure-valued-39-billion-latest-funding-round-2025-09-16/
- https://www.reuters.com/business/robotics-startup-humanoid-raises-152-million-series-round-135-billion-valuation-2026-07-21/
- https://www.reuters.com/technology/artificial-intelligence/robot-ai-startup-physical-intelligence-raises-400-mln-bezos-openai-2024-11-04/
- https://waymo.com/blog/2025/12/2025-year-in-review
- https://www.reuters.com/world/asia-pacific/chinese-humanoid-robot-maker-unitree-set-jump-over-600-shanghai-debut-2026-08-19/