PHYSICAL AI SIGNALS - March: The Rise of the Humanoid Economy

Physical AI stack underwent a massive re-rating. The biggest takeaway? The intelligence layer is beating the hardware layer: From the White House to the Factory Floor

Physical AI Builders  |  50builders.ai

1.  EXECUTIVE SUMMARY

The intelligence layer is beating the hardware layer. Three software-first robotics companies are now collectively valued at over $25B — before deploying a single commercial unit at scale.

March 2026 was the month Physical AI stopped being a futures trade and started demanding a present-tense investment thesis. NVIDIA codified its full-stack platform at GTC, Tesla crossed irreversible production thresholds for Optimus, foundation model valuations doubled in four months, and a humanoid robot walked the White House red carpet. The market is no longer asking whether physical AI is real — it is asking who captures the value.

KEY SIGNALS THIS MONTH

• Physical Intelligence is in talks to raise ~$1B at an $11B+ valuation — doubling from $5.6B just four months ago. With Founders Fund and Lightspeed circling, this marks the fastest institutional re-rating in robotics history.
• NVIDIA GTC declared 'every industrial company will become a robotics company.' GR00T N1.7 is now commercially licensed; GR00T N2 previewed with 2× task success vs. leading VLA models. NVIDIA projects $1T in AI compute demand through 2027.
• Tesla Optimus Gen 3 mass production officially began at Fremont in January, with 50,000–100,000 unit targets for 2026. Fremont's former Model S/X lines have been fully repurposed. A Giga Texas factory targeting 10M units/year is under construction.
• Skild AI deployed its robot brain on Foxconn lines building NVIDIA Blackwell GPU servers in Houston — the first large-scale commercial deployment of a foundation model robot intelligence layer. This is the proof-of-concept the entire sector needed.
• Figure 03 became the first humanoid robot to enter the White House, debuting at Melania Trump's 'Fostering the Future Together' summit. The geopolitical framing — 'a humanoid built for the United States of America' — signals the start of state-level robotics brand competition.
• Power infrastructure hit a regulatory flashpoint: NERC issued a formal grid-risk warning on AI data center demand; PJM projects a 6 GW reliability shortfall by 2027; electricity costs are up 42% since 2019. The energy constraint is no longer theoretical.

NON-OBVIOUS / CONTRARIAN SIGNALS

• The 'no commercialization timeline' posture of Physical Intelligence ($1B raised, 80 employees, no revenue guidance) is increasingly being rewarded, not penalized. Investors are betting that the winner in robot foundation models will be whoever compiles the most diverse real-world task data — and that any premature revenue pressure will compromise data diversity. This mirrors early LLM dynamics.
• Tesla's actual moat may not be Optimus hardware. It's the AI5 chip + FSD training infrastructure being dual-purposed for robotics. No other humanoid company has a proprietary inference chip at this scale. The hardware race may resolve faster than expected — and Tesla's true advantage surfaces in the software inference layer.

2.  CAPITAL FLOWS: WHERE CONVICTION IS FORMING

2.1  Venture Capital: Foundation Model Layer Commanding the Premium

Physical Intelligence — Series C (In Discussion)

Stage Late Series C (rumored) ~$1B (in talks as of March 27) $11B+ | up from $5.6B four months prior — 2× re-rate

Lead Founders Fund (set to participate); Lightspeed in talks; Thrive Capital + Lux Capital returning

Technical Thesis General-purpose AI models for any robot — 'ChatGPT for robots.' 80 employees, zero commercialization timeline.

Claimed Moat Foundation model generalizes across embodiments without hardware-specific retraining. Compute-first approach to robot policy.

Investor Read Capital is pricing the data flywheel, not current revenue. PI is buying time and compute to compile the world's most diverse robot task dataset before locking in an architecture.

Skild AI — Series C (Closed January 2026)

$1.4B, $14B+ | up from $4.5B in July 2025 — 3× in 7 months

Lead SoftBank Group; NVentures (NVIDIA), Bezos Expeditions, Macquarie Capital

Strategic Backers, Samsung, LG, Schneider Electric, Salesforce Ventures, CommonSpirit Health

Technical Thesis Omni-bodied intelligence: a single foundation model that runs any robot without hardware-specific retraining.

Claimed Moat In-context learning across embodiments (quadruped, humanoid, tabletop arm, mobile manipulator). Revenue ~$30M in 2025 from zero. Already profitable.

Investor Read The intelligence OS thesis. SoftBank is betting Skild captures the ABB/FANUC installed base through software, similar to how Android commoditized handset OEMs. Strategic investors (Samsung, LG, Schneider) are paying for option value on their own automation roadmaps.

RoboForce — Series A

$52M (oversubscribed); $67M total raised

Lead YZi Labs ($10B fund); Jerry Yang, Myron Scholes (Nobel Laureate), Carnegie Mellon University

Technical Thesis Physical AI for industrial 'robo-labor': solar farms, data centers, mining, manufacturing, logistics.

NVIDIA Stack Full NVIDIA integration: Jetson Thor (edge), Isaac Sim/Lab (simulation), Cosmos (synthetic data), OSMO (orchestration).

Investor Read Narrow-vertical industrial deployment with clear labor economics. YZi's check signals China-adjacent capital still flowing to US-founded robotics. CMU backing adds data/research access. The NVIDIA integration is a capability accelerant but also a dependency.

TREND SIGNAL: The gap between foundation-model valuations and industrial-deployment valuations is widening fast. PI and Skild are valued at 10–30× revenue multiples with minimal deployments. RoboForce is raising on labor-cost displacement math. Capital is bifurcating: 'platform bets' vs. 'deployment execution bets.' Both are valid but they are very different risk profiles.

2.2  Corporate / Strategic Investment: The Platform Control Play

NVIDIA's GTC was structurally a capital deployment signal disguised as a product launch. Every partnership announced — ABB, FANUC, KUKA, Universal Robots, Hexagon Robotics — was NVIDIA buying an integration point into industrial robot fleets. NVIDIA is not building robots. It is inserting its compute, simulation, and model stack between every robot OEM and their customers' operational data.

• ABB is integrating NVIDIA Omniverse into its RobotStudio platform (HyperReality release expected 2026). This means ABB's customers will train robot behaviors in NVIDIA simulation — locking in NVIDIA's data pipeline at the industrial operator level.
• Skild AI + NVIDIA + Foxconn: Skild's Brain is now running on Foxconn's Blackwell GPU assembly lines in Houston. This is the first real-world proof that robot foundation models can operate in a high-precision, high-throughput manufacturing environment. NVIDIA funded it; Foxconn deployed it; Skild trains on it. All three benefit from the data flywheel.
• Figure AI's BotQ manufacturing target (50,000 units/year by end 2026) combined with its 'Hark' AI voice integration and shift away from OpenAI voice systems signals an emerging vertical integration strategy. Figure is trying to own hardware + intelligence + voice interaction simultaneously.

2.3  M&A and Joint Ventures: Supply Chain & Vertical Integration

No major closed acquisitions in March 2026, but two structural patterns are accelerating:

Tesla: Vertical Integration at Maximum Scope

Tesla's repurposing of Fremont lines (previously Model S/X) to Optimus Gen 3 production is one of the largest reconfigurations of automotive manufacturing capacity in a decade. This is not a prototype lab — it is a bet that humanoid robot manufacturing can be treated like automotive manufacturing. The Giga Texas Optimus facility targets 10M units/year, meaning Tesla is pre-building supply chain infrastructure at a scale that presupposes mass consumer adoption.

• Strategic implication: Tesla's Gen 3 hands (22 DoF, 50 actuators) plus the AI5 chip are NVIDIA-independent. Tesla does not run GR00T. This is the most consequential supply chain differentiation in the sector — Tesla is building a closed vertical stack that bypasses the NVIDIA Physical AI platform entirely.

ABB + VoltaGrid (Data Center Power Infrastructure — March 25, 2026)

Signed at CERAWeek in Houston, ABB extended its partnership with VoltaGrid to supply 35 synchronous condensers and associated prefabricated eHouse units for global hyperscale AI data center power stabilization. Financial terms undisclosed; orders to be booked Q2 2026.

• Strategic implication: ABB is simultaneously playing the robot manufacturing layer (NVIDIA integration) and the power infrastructure layer (VoltaGrid). This is a systems bet that industrial electrification and robotics are one and the same market at the infrastructure level.

2.4  Capital Allocation Synthesis

Where is conviction concentrated:

• Foundation model layer (PI, Skild): ~$3.4B deployed in 90 days. Highest multiples, lowest near-term revenue. 'AGI for robots' thesis.
• Hardware humanoids (Tesla Optimus, Figure AI): Production infrastructure investment running into tens of billions in total capex commitments. Revenue runway 12–18 months out.
• Industrial deployment (RoboForce, RobCo): Series A–C range, quantifiable labor economics, narrower markets. Execution risk over thesis risk.
• Infrastructure (power, edge compute): ABB, VoltaGrid, Vertiv moving into 800V DC architectures. This is the unglamorous constraint layer that determines how fast robots can actually scale.

UNDERFUNDED LAYER: Human-robot safety perception — the 'visual cortex' for robot situational awareness in unstructured environments. Multiple executives flagged this at GTC and the White House summit. No foundation model startup is explicitly raising around this. It is a prerequisite for commercial deployment at scale.

OVERFUNDED RISK: General-purpose foundation model companies with no commercialization timeline. The 'ChatGPT moment' analogy is seductive, but robot policies require physical hardware, data collection infrastructure, and safety validation that software LLMs did not. The data flywheel assumption may not scale as cleanly as the language model analogy implies.

3.  THE BIG TECH PHYSICAL PIVOT

NVIDIA — Platform Declared, Stack Closed

GTC 2026 (March 16–19, San Jose) was NVIDIA's most consequential Physical AI announcement in the company's history. Jensen Huang's declaration — 'every industrial company will become a robotics company' — is not hyperbole. It is a product roadmap statement.

Infrastructure & Platform

• Physical AI Data Factory Blueprint — Open architecture for generating, augmenting, and evaluating robot training data. Reduces the cost of rare-scenario data generation. Published to GitHub in April.
• OSMO Orchestration Framework — Integrates with Claude Code, OpenAI Codex, and Cursor so AI coding agents can autonomously manage robotics data pipeline bottlenecks. Azure and Nebius incorporating into cloud services.
• IGX Thor (GA) — NVIDIA's AI edge computing platform is now generally available for safety-critical use cases. Johnson & Johnson (surgical), Karl Storz (endoscopy), Medtronic (evaluation) are first adopters.
• 5G Edge AI — Partnership with T-Mobile and Nokia to deploy RTX PRO 6000 Blackwell servers at network edge sites. Pilot 'City Operations Agent' in San Jose reduces incident response time 5×.
• $1 Trillion Compute Demand — Jensen raised AI infrastructure demand projections from $500B (through 2026) to $1T (through 2027). Inference has overtaken training as the dominant workload. Vera Rubin (Blackwell successor, 10× performance/watt) shipping later in 2026.

STRATEGIC INTENT: NVIDIA is converting the robotics data problem into a compute problem. The bottleneck for smarter robots is no longer real-world data collection — it is compute for simulation training. This means NVIDIA's revenue accelerates as robots get smarter. Every GR00T improvement is a sell-through event for Blackwell and Vera Rubin.

Tesla — Production Infrastructure Commitment Is Now Irreversible

Tesla crossed a line in Q1 2026 that cannot be uncrossed: it shut down Model S and Model X production and converted those Fremont lines to Optimus Gen 3 manufacturing. This is not a skunkworks robotics program. This is Tesla's primary manufacturing bet.

• Gen 3 Specs: 22 DoF hands, 50 total actuators (vs. 11 in Gen 2), Tesla AI5 chip (~5× memory bandwidth vs. Gen 2), Grok (xAI) voice integration, price target sub-$20K at scale.
• Timeline: Mass production started Fremont January 2026 (low-volume). External commercial customers expected late 2026 at B2B prices estimated above $100K/unit. Giga Texas factory (10M units/year target) under construction.
• Delay: Tesla missed its self-imposed Q1 prototype reveal deadline. As of March 31, Musk confirmed the robot is 'walking around' but needs 'finishing touches' before public showcase.
• Talent risk: Milan Kovac, former Optimus lead, departed to Boston Dynamics — a meaningful signal worth watching. The Optimus team is large (100+ open roles) but talent churn in competitive robotics is a real risk.

CONTRARIAN TAKE: The 50,000–100,000 unit target for 2026 should be read as a ceiling, not a floor. Tesla's history of production S-curves (Model 3 'production hell') suggests the real ramp will be 2027. The more important data point is reliability: can Optimus Gen 3 hands perform at 24/7 factory conditions? Q2–Q3 deployment results are the real signal.

Figure AI — Geopolitical Brand Launch

Figure 03 became the first humanoid to enter the White House on March 25, walking beside First Lady Melania Trump at the 'Fostering the Future Together' summit. The robot greeted attendees in 11 languages and self-identified as 'a humanoid built for the United States of America.'

• Technical context: Figure 03 runs Helix 02, a Software 2.0 architecture replacing hand-coded C++ heuristics. Voice is shifting from OpenAI integration to Figure's own 'Hark' omni-model (speech + reasoning + physical action unified).
• Manufacturing: BotQ facility targeting ~50,000 units/year by end 2026. Supply chain to be nearly 100% non-Chinese by summer 2026.
• Valuation: Raised $1B+ Series C at $39B valuation. Backed by NVIDIA, Intel Capital, Qualcomm Ventures, Salesforce.
• Strategic implication: The White House appearance was brand strategy as much as technology demonstration. 'Made in America' framing is now a differentiation vector in humanoid robotics — a direct counter-positioning to Chinese humanoid competitors (AGIBOT, Unitree).

4.  INFRASTRUCTURE & POWER: THE CONSTRAINT LAYER

The infrastructure crisis underlying Physical AI is no longer a 2030 forecast. It is a 2026 operational problem.

Grid Stress: NERC Warning + PJM Shortfall

• NERC (North American Electric Reliability Corporation) issued a formal warning in March 2026: AI data center demand creates 'high likelihood, high impact grid risks' including potential cascading outages if the largest facilities remain unregulated.
• PJM Interconnection (65M people, 13 states) projects a 6 GW reliability shortfall in 2027. Capacity market prices have spiked nearly 10× in some regions, driving retail electricity increases above 15% in parts of the service area.
• Retail electricity costs are up 42% since 2019 vs. 29% CPI inflation over the same period. Goldman Sachs projects data center power demand will add 0.1% to core inflation in both 2026 and 2027.

Scale: 15 GW → ~100 GW Pipeline

• The entire US data center sector currently draws fewer than 15 GW. The pipeline of projects under construction, if completed, would push that to ~100 GW — more than a 6× increase. Some single proposed facilities seek more than 5 GW, exceeding the peak load of entire cities.
• NTT Global Data Centers announced plans on March 19 to double its global capacity to 4 GW — a signal the expansion is worldwide, not just North American.
• Meta announced a new 1 GW campus (powers ~750,000 homes). Adani is in active talks with both Meta and Google for large-scale data center partnerships.

Architecture Shift: 800V DC Power

• NVIDIA's Vera Rubin racks are specifying 800V DC distribution. At 1 MW per rack, traditional AC-to-DC conversion requires ~200 kg of copper busbar per rack — physically untenable at gigawatt scale.
• Vertiv's 800V DC ecosystem (integrating with Vera Rubin) commercially available H2 2026. Eaton and Delta shipping 800V DC systems. Most of the industry is still at 400V DC — a meaningful readiness gap.
• ABB + VoltaGrid deal (signed CERAWeek, Houston, March 25): 35 synchronous condensers with flywheel for voltage stabilization in behind-the-meter AI data center power. ABB is becoming the critical stabilization layer for hyperscale AI compute.

Physical AI-Specific Infrastructure Implication

Robot training and deployment requires both training compute (data center) AND edge inference (Jetson Thor, IGX Thor). The infrastructure constraint is therefore two-sided: centralized GPU clusters for foundation model training, and distributed edge compute for real-time robot control. The edge compute layer is currently less discussed but equally constraining — and NVIDIA's 5G edge AI partnership with T-Mobile is the first serious attempt to solve it at population scale.

BOTTLENECK: Power interconnection timelines (currently 2–5 year queues at many US utilities) are the single largest deployment constraint for Physical AI infrastructure. Modular, grid-independent power systems (e.g., Data Power Supply's factory-to-site-in-4-weeks model) are an undervalued near-term solution until utility-scale interconnections come online.

5.  UNDER-THE-RADAR SIGNAL  (Most Important Section)

HIGH-CONVICTION, OVERLOOKED: The Skild Brain / Foxconn / NVIDIA Houston Deployment — and What It Means for the Intelligence Layer Economics

What It Is

In March 2026, Skild AI (Pittsburgh, $14B valuation) deployed its 'Skild Brain' foundation model onto Foxconn's robotic assembly lines in Houston, Texas — the lines building NVIDIA Blackwell GPU server systems. This marks the first publicly documented large-scale commercial deployment of a hardware-agnostic robot intelligence foundation model in a high-precision, high-throughput manufacturing environment.

The Core Technical Breakthrough

The Skild Brain is 'omni-bodied' — it controls robots it has never trained on by adapting in real time to new body forms. The Houston deployment is not a controlled demo. It is production-line robotics operating under commercial SLA pressure.

• Data flywheel effect: Each assembly task generates real-world operational data that feeds back into model training. The more Foxconn lines Skild powers, the smarter the model becomes for every other customer. This is a compounding returns architecture — identical to how LLM companies benefit from production traffic.
• Compute backbone: NVIDIA Jetson Thor (edge inference) + Isaac Sim (simulation) + Cosmos (synthetic data generation). The entire stack is NVIDIA-native, meaning Skild's data flywheel also accelerates NVIDIA's simulation accuracy. This is a mutual compounding arrangement.

Why the Market Is Underestimating This

• The narrative focus is on humanoid hardware (Tesla Optimus, Figure 03). But the Foxconn deployment proves that software intelligence layers — not hardware — capture the recurring revenue in robotics. Skild charges per deployment; the hardware (Foxconn's arms) is someone else's capex.
• The $14B valuation looks expensive on a revenue multiple today. If the data flywheel compounds over 36 months of Foxconn-scale deployments, the model's performance gap vs. hardware-specific solutions widens every quarter. The valuation is pricing a network effect that hasn't yet fully materialized.
• Industrial robot OEM pricing power (ABB, FANUC, KUKA) depends on proprietary software lock-in. Skild's omni-bodied model breaks that lock-in — one intelligence layer, any hardware. The revenue migration from OEMs to software platforms is already happening; the Foxconn deployment is the first large-scale proof.

Why Now

NVIDIA needed a proof point for its Physical AI platform at GTC. Skild needed a showcase deployment with production credibility. Foxconn needed to demonstrate its ability to build Blackwell GPU servers autonomously at scale. All three parties had perfectly aligned incentives — which is exactly why this happened at this moment, and exactly why it will be replicated across other Foxconn production lines in 2026.

Stress Test: What Kills This Thesis

• Reliability at 24/7 production rates. Lab-quality performance in a controlled demo does not automatically translate to 99.9% uptime on a manufacturing floor. If failure rates are material, operators default to humans.
• NVIDIA dependency risk. Skild's entire training stack is NVIDIA-native. If NVIDIA changes licensing terms, prioritizes its own robotics software, or acquires a competing intelligence layer, Skild's moat narrows.
• Tesla's in-house path. If Optimus Gen 3 at Fremont produces a proprietary intelligence dataset at a scale that matches Skild's omni-bodied training, the 'hardware-agnostic' value proposition becomes less differentiated.

Long-Term Implication

The Foxconn deployment is the 'Android on the first Nexus phone' moment for robot foundation models — not the iPhone moment (that comes when a consumer humanoid ships at scale). It proves the software abstraction layer works in production. Over the next 24 months, watch for Skild (or PI, or another foundation model company) to displace proprietary OEM software across ABB, FANUC, and KUKA fleets. When that happens, the $14B valuation will look like the entry point, not the peak.

KEY PERSON MOVES: TALENT AS A LEADING INDICATOR

In Physical AI, where teams are small and the field is nascent, executive movement is a leading indicator of where the next capabilities will form.

• Milan Kovac (former Tesla Optimus lead) → Boston Dynamics. Kovac was a critical architect of Optimus's AI stack. His move to Boston Dynamics — now part of Hyundai's robotics ecosystem — is a signal that Boston Dynamics is accelerating its humanoid intelligence roadmap.
• Michael Perry (ex-Boston Dynamics VP of Marketing and Strategy) → Persona AI. First noted in SXSW coverage; Persona AI is building a social humanoid platform. Perry's brand-building expertise at Boston Dynamics suggests Persona is prioritizing consumer and enterprise narrative.
• Figure AI replacing OpenAI voice integration with in-house 'Hark' omni-model. This is a talent/IP concentration signal: Figure is retaining AI inference capability in-house rather than paying per-call to an external provider. Margin and defensibility implications are significant.

6.  APRIL 2026 OUTLOOK: WHAT TO WATCH

Physical Intelligence Round Close

Watch for formal close announcement of the ~$1B round. Founders Fund participation would be a signal that frontier AI capital — not just robotics-specialist VCs — has fully committed to the physical world. Bull: closes at $11B+, accelerates PI's simulation infrastructure. Bear: terms change, round size reduced, signals investor recalibration on pre-revenue robotics multiples.

Tesla Optimus Gen 3 Prototype Reveal

Musk confirmed the robot is 'walking around' with 'finishing touches' needed. Reveal expected April–May 2026. Bull: Gen 3 hands demonstrate reliable dexterous manipulation in factory video. Bear: reveal is another controlled demo with limited real-world reliability evidence.

NVIDIA GR00T N2 / GitHub Release

NVIDIA's Physical AI Data Factory Blueprint and OSMO framework scheduled for GitHub release in April. Watch adoption rate by robotics startups and industrial partners. Bull: rapid open-source community adoption creates network effects that entrench NVIDIA's simulation stack. Bear: slow adoption exposes a gap between NVIDIA's platform ambition and developer workflow reality.

Skild / Foxconn Houston Deployment Performance Data

No public data has been released on the Foxconn line performance since the March announcement. April earnings releases for NVIDIA and Foxconn parent Hon Hai may include qualitative commentary. Watch for defect rates, uptime data, or any indication of expanded line deployment. This is the highest-signal data point in the entire sector right now.

PJM / NERC Grid Policy Actions

Following NERC's formal warning and PJM's 6 GW shortfall projection, state legislatures in Virginia, Georgia, Indiana, and Washington are considering data center impact fee legislation. Any movement on federal interconnection queue reform (FERC) could materially accelerate or delay data center buildout timelines. Bull: FERC fast-tracks transmission reform, unlocking queued capacity. Bear: political gridlock plus utility rate hikes trigger backlash that slows data center permitting.

Figure AI BotQ Production Ramp

Figure's claimed 50,000 units/year target for BotQ requires supply chain execution at a pace no humanoid startup has demonstrated. April–June is the first real window to validate or disprove this timeline. Watch for any commercial pilot announcements.

7.  CLOSING INSIGHT: SYSTEM-LEVEL TAKE

Which Layer Is Accelerating Fastest?

The intelligence layer. In 90 days, the three leading robot foundation model companies (Physical Intelligence, Skild AI, and implicitly Tesla's in-house stack) collectively added approximately $10–15B in valuation. No hardware company has re-rated at that speed. The market has made its thesis clear: the robot intelligence layer is where value will compound, not the actuator or chassis layer.

Where Are the New Constraints Emerging?

Three constraints are converging simultaneously:

• Power: The US grid cannot support the training infrastructure required to build the next generation of robot foundation models at the pace capital demands. NERC's warning is the canary. The constraint is already showing up in interconnection queue timelines and retail electricity prices.
• Safety validation: No regulatory framework exists to certify a general-purpose humanoid robot for unstructured commercial environments. The path from 'impressive demo' to 'insurable deployment' is undefined. This is the hidden bottleneck in every humanoid commercial timeline.
• Real-world data: Simulation has improved dramatically (NVIDIA Cosmos 3), but there is no substitute for diverse real-world operational data. Physical Intelligence's refusal to commercialize prematurely is a correct strategic read: the company that compiles the most diverse task data in the next 24 months will have a compounding advantage that is very difficult to replicate.

Where Is the Next Bottleneck Shifting?

Today the bottleneck is data quality and diversity (solved partly by simulation, partly by deployment scale). In 12–18 months, as foundation models mature, the bottleneck will shift to safety certification and insurance underwriting — the regulatory and actuarial infrastructure required to put general-purpose robots in environments with humans. No company in the sector is investing seriously in this layer yet. That is where the next unlocking opportunity is.

STATE OF THE MARKET — MARCH 2026: The intelligence layer won the capital race; now the constraint is whether the physical world — its power grids, its regulatory frameworks, its factory floors — can absorb the robots fast enough to justify the valuations.

Physical AI Builders  |  50builders.ai  |  lu.ma/physical-ai  |  physical.f50.ai

© 2026 F50 / Physical AI Builders. For institutional use. Not investment advice.