From Land, Ocean, Water, Space -Funding & Actions: PHYSICAL AI SIGNAL REPORT June 2026

Author: David Cao Publisher: Physical AI Builders | 50builders.ai Supported by: Physical AI Builders Community & F50

June closed a question the market had been dancing around since Q1: is Physical AI a robotics story or a manufacturing story? The answer that emerged this month is "both, running on the same balance sheet." Generalist AI raised $400M for an embodied-AI model layer three months after its $500M Series A, doubling its valuation to roughly $2B in the space of a single quarter. Mach Industries closed a $300M Series C at $1.8B, a 4x step-up in twelve months, on the strength of a "10 vehicles a week to 1,000 a week" production promise rather than disclosed revenue. And NVIDIA didn't announce a chip — it announced Halos, a full-stack safety certification architecture for humanoids working next to people on a factory floor. None of these three signals, on their own, would define a month. Together, they say the center of gravity in Physical AI has shifted from "can the model learn the task" to "can the company build and certify the hardware fast enough to keep up with the model."

The domain with the most capital and corporate motion this month was, again, Land — robotics and industrial automation absorbed the largest checks and the deepest bench of announcements. But the more consequential structural news came from Maritime and Space, where regulators and government buyers, not venture investors, set the pace. The IMO adopted its MASS Code for autonomous ships in May, taking effect July 1 — a genuine first for binding international autonomy regulation. And NASA quietly redirected nearly $1B of its Moon Base budget into hard commercial contracts (Astrolab and Lunar Outpost each landing roughly $220M for lunar terrain vehicles), which matters more for the sector's maturity arc than any single funding round.

The one thing almost nobody wrote about this month: DoD's Defense Autonomous Warfare Group budget request jumped from $225.9M in FY26 to a proposed $54.6B in FY27 — a 24,000% increase buried inside a broader RDT&E document. That's not a typo, and it's not yet a signed appropriation, but it is the single largest forward-looking signal in the entire sector this month, dwarfing every venture round combined.

This Month's Defining Signal: Physical AI stopped being rewarded for intelligence alone and started being rewarded for the ability to manufacture, certify, and deploy that intelligence at industrial scale.

🏭 Generalist AI's valuation quadrupled in about seven months. San Mateo-based Generalist AI raised $500M in March and came back for $400M in early June, reportedly near a $2B valuation, led by Radical Ventures. The company builds embodied-AI models that sit above the robot hardware layer. Implication: capital is increasingly willing to pay a premium for the model layer independent of any single humanoid platform — a bet that intelligence, not actuators, is the scarce resource, even as production capacity gets rewarded elsewhere.

🏭 Robotics funding this trailing year hit $4.58B across pure-play companies, with Collaborative Robots absorbing 34% of dollars on just 11% of deal count. North America captured 82% of all disclosed capital. Implication: the market is not spreading bets — it's stacking them on a small number of humanoid and cobot platforms it believes will win categories outright.

🌊 Cellula Robotics and Infinity Fuel Cell completed a 2,023 km fully submerged AUV mission and separately won a Defense Innovation Unit CAMP contract for long-endurance maritime autonomy. Implication: subsea endurance — not payload sophistication — is becoming the metric defense buyers actually procure against.

🚢 The IMO's MASS Code, finalized in May 2026, takes effect July 1 — establishing the first international safety framework for maritime autonomous surface ships, with a path from voluntary to mandatory status by 2032. Implication: shipowners and autonomy vendors now have a regulatory runway to build against, which typically precedes a wave of commercial deployment decisions.

✈️ Mach Industries' Series C priced a pre-revenue, five-program hardware company at $1.8B — a 4x step-up from its Series B twelve months prior — while DoD's proposed FY27 budget shows the Defense Autonomous Warfare Group request jumping from $225.9M to $54.6B. Implication: defense-tech valuations are increasingly forward-pricing procurement budgets that haven't been appropriated yet. That's a bet on policy momentum as much as on hardware execution.

🚀 NASA committed roughly $220M each to Astrolab and Lunar Outpost for lunar terrain vehicle development, part of a nearly $1B Moon Base Phase One package. Implication: after years of pilot programs and small Series rounds, lunar mobility now has government check sizes that resemble a real procurement program rather than a science demonstration.

🏭 NVIDIA announced Halos for Robotics — a full-stack safety architecture (compute, software, sensors, certification) for humanoids working alongside humans, with Agility as first adopter for its Digit robot deployed at Amazon, GXO, Schaeffler, and Toyota facilities. Implication: safety certification, not raw capability, is emerging as the actual gating factor for humanoid deployment at scale in real facilities.

Generalist AI — $400M, new funding round (San Mateo, CA). Led by Radical Ventures, reportedly near a $2B valuation. Generalist AI builds embodied-AI models designed to help robots learn and generalize across real-world physical tasks, rather than building its own hardware platform. This is the company's second nine-figure raise in roughly three months, following a $500M Series A in March co-led by Accel and Andreessen Horowitz. Why it matters: investors are now willing to underwrite the model layer as a standalone, hardware-agnostic business — a structural echo of how the LLM stack separated from the chip layer. Builder signal: if you're building a robot-agnostic intelligence layer, the fundraising environment has never been more receptive; if you're building hardware without a clear model-layer partnership, expect harder questions about defensibility.

Mecka AI — $60M across two previously unannounced rounds ($25M Series A + $35M follow-on) (New York, NY). Backers include Bessemer Venture Partners and Framework Ventures. Mecka AI collects "embodied AI" motion data from consumer wearables and body sensors to train robotics models, positioning itself as a data supplier rather than a robot or model builder. Why it matters: the data-layer land grab for robotics is intensifying, mirroring the early internet-scale scraping era for LLMs. Builder signal: proprietary, hard-to-replicate motion data is becoming a fundable moat on its own, independent of any downstream robot.

Trend Signal: Land remains the deepest and most liquid of the five domains — $18.8B raised industry-wide year-to-date against $15B for all of 2025 — but capital is increasingly concentrated in the top handful of deals. The story to watch is whether the model-layer companies (Generalist AI, Skild AI, Physical Intelligence) or the vertically-integrated hardware companies (Figure, Apptronik, Neura Robotics) end up capturing the larger share of long-term value.

Domain Watch: No large disclosed venture round closed in Ocean this month — consistent with the domain's pattern of being driven more by defense contracts and endurance demonstrations than headline equity rounds. The signal to track: Cellula Robotics and Infinity Fuel Cell's 2,023-kilometer, 385-hour fully submerged AUV mission, paired with Cellula's selection to deliver a fuel-cell-powered Guardian AUV under the Defense Innovation Unit's CAMP program. Separately, KONGSBERG and DRASS announced a strategic partnership to build scalable, interoperable subsea systems for allied defense needs. Watch for follow-on Series rounds from long-endurance AUV players over the next 1–2 quarters as defense procurement contracts convert into growth capital raises — this is typically the sequence in Ocean: contract first, equity round second.

Domain Watch: No new mega-round in Maritime this month, but the regulatory news is the real story. The IMO's International Code of Safety for Maritime Autonomous Surface Ships (MASS Code), finalized in May, takes effect July 1, 2026, with a stated path to mandatory status by 2032. Saronic continues to be the domain's gravitational center following its $1.75B Series D in March (now deploying capital into its Louisiana shipyard expansion and Port Alpha facility), and the U.S. Navy is transitioning its medium unmanned surface vessels, Seahawk and Sea Hunter, from experimental to fleet-operational status. Singapore's PIER71 Smart Port Challenge opened its tenth cohort this month, shifting explicit focus from "discovery" to "scale," with new venture-to-capital bridging tracks. Builder signal: regulatory clarity from the MASS Code should start unlocking commercial (not just defense) autonomous vessel deployment decisions in H2 2026 — this is the domain most likely to surprise to the upside in Q3–Q4.

Mach Industries — $300M Series C at a $1.8B valuation (Huntington Beach, CA). Led by Infinite Capital, with Ribbit Capital, Sequoia, Khosla Ventures, and Bedrock Capital participating — all insiders returning from earlier rounds. The round was oversubscribed against a $200M target and represents a 4x step-up from the company's ~$470M valuation twelve months earlier. Mach is running five simultaneous autonomous weapons and defense-aviation programs out of a single Huntington Beach facility with roughly 350 employees, and claims a path from 10 vehicles per week today to 1,000 per week within a year. Notably, the company is pre-revenue at production scale — this is a valuation built on execution promise and policy tailwinds, not disclosed backlog. Why it matters: compare this to Anduril's $5B Series H in May at a $61B valuation, backed by $2.2B in actual 2025 revenue — Mach is being priced on a fundamentally different, higher-risk basis. Builder signal: the bar for "defense hardware startup" valuations has moved decisively upward, but due diligence discipline on revenue-versus-narrative is going to matter more, not less, from here.

The ePlane Company — $50M Series C (India). Backing certification of the company's air-ambulance eVTOL variant, the e200X, with the company becoming the first Indian private firm to receive Directorate General of Civil Aviation acceptance for type certification under the country's new eVTOL rules. Why it matters: this is a signal that eVTOL certification pathways are opening meaningfully outside the U.S. and Europe.

Trend Signal: Air is bifurcating cleanly into two theses — defense-aerial systems commanding prime-adjacent, execution-risk valuations (Mach, Anduril), and commercial eVTOL/cargo players progressing more slowly through certification (ePlane, Acodyne's €2.5M pre-seed for heavy-lift eVTOL). The DoD's proposed $70B+ drone and counter-drone budget for FY27 — including the eyebrow-raising DAWG line item jump to $54.6B — is the macro tailwind underwriting nearly every defense-aerial valuation in the domain right now.

Domain Watch — but the largest capital event of the month wasn't equity, it was procurement. NASA's Moon Base Phase One package, unveiled in late May and rippling through June, awarded Astrolab and Lunar Outpost roughly $220M each to finalize lunar terrain vehicle designs, alongside a $75M Firefly Aerospace subcontract to deploy lunar hopper drones on the MoonFall mission. This follows Lunar Outpost's $30M Series B in May (Industrious Ventures-led), which the company closed in under five weeks — the fastest of its three rounds to date. Why it matters: NASA effectively converted years of CLPS-style pilot contracts into a real procurement program with real check sizes, which is the signal space-robotics investors have been waiting for since Artemis was announced. Builder signal: the companies best positioned in Space right now are those with existing flight heritage and multiple contracted missions — Lunar Outpost's eight contracted lunar and cislunar missions is now a genuine moat, not just a marketing line.

Intelligence/Compute Layer. NVIDIA had arguably its most consequential month of the year for Physical AI specifically — not for a chip launch, but for Halos for Robotics, the industry's first full-stack safety certification system spanning AI compute (IGX Thor), software, sensors, and third-party inspection standards (IEC 61508, ISO 13849). Agility Robotics is the first adopter, integrating Halos into its Digit humanoid deployed across Amazon, GXO, Schaeffler, and Toyota Motor Manufacturing Canada facilities. Separately, NVIDIA deepened its industrial-software partnerships (Cadence, Dassault Systèmes, PTC, Siemens, Synopsys) to bring CUDA-X and Omniverse into the CAD-to-simulation pipeline for FANUC, HD Hyundai, Honda, and TSMC, among others.

OEM/Industrial Layer. AeroVironment continued its consolidation strategy in defense aerial systems, following its $4.1B BlueHalo acquisition with a $200M purchase of engineering firm ESAero, folding it into its Precision Strike and Defense Systems group. Saronic continues converting its March $1.75B raise into physical shipbuilding capacity, adding over 300,000 square feet and three new slips in Louisiana. HII demonstrated its Sea Launcher automated launch-and-recovery system for AUVs, a meaningful step in integrating autonomous underwater vehicles with surface fleet operations.

Cross-Domain Corporate Signal of the Month: Safety and certification infrastructure — not raw model capability — is becoming the axis on which large corporate partnerships form, from NVIDIA's Halos framework in Land to the IMO's MASS Code in Maritime. When the compute layer and the regulatory layer both converge on certification as the next unlock, that's usually a leading indicator that commercial deployment, not pilot programs, is the next 12-month story.

U.S. Federal. The Pentagon's proposed FY27 budget allocates more than $70B for military drones and counter-drone systems — the department's largest such investment to date — up from $13.4B for autonomous systems and $3.1B for counter-drone capabilities in FY26. Buried within the RDT&E documentation is a proposed jump in the Defense Autonomous Warfare Group's budget from $225.9M (FY26 actual) to $54.6B (FY27 request), described by DoD officials as a "pathfinder" function testing and integrating autonomy orchestration tools across vendors. This is a request, not an appropriation, and deserves skepticism proportional to its size — but even a fraction of that figure would reshape the Air domain's competitive landscape. NASA's Moon Base Phase One committed nearly $1B across landers, rovers, and technology demonstrations, with missions beginning in the back half of 2026.

International. The IMO formally adopted its MASS Code in May, effective July 1, 2026 — the first binding international safety instrument for autonomous ships, with mandatory status targeted for 2032. Singapore's Maritime and Port Authority launched the tenth PIER71 Smart Port Challenge, explicitly restructuring around helping startups cross from pilot to commercial scale via new investor-matching tracks.

Export Control and Supply Chain Watch. The BIS Affiliates Rule, which will require companies to map complex ownership structures used for procurement circumvention, remains in an enforcement pause until late 2026 — but companies are already collecting compliance data ahead of activation. The administration's January 2026 shift on H200-class chip exports to China (from "presumption of denial" to "case-by-case review") remains contested in Congress, with the AI OVERWATCH Act advancing in the House Foreign Affairs Committee on a bipartisan basis. For hardware-heavy Physical AI companies with cross-border supply chains — batteries, rare-earth-adjacent components, advanced compute — this is the single most important policy thread to track into Q3, as NDAA-driven battery restrictions on foreign entities of concern begin phasing in starting January 2028.

Two announcements landed in June that, on the surface, have nothing to do with each other. NVIDIA introduced Halos for Robotics, a full-stack safety certification architecture that lets humanoid robots operate legally and defensibly alongside human workers on a factory floor. The IMO finalized its MASS Code, the first binding international safety framework for autonomous ships, taking effect July 1. One is a corporate product launch; the other is a multilateral regulatory instrument. But both point to the same underlying shift, and it's the most important structural development in Physical AI this month.

For the past several years, the binding constraint on Physical AI deployment has been capability: could the model perceive the environment accurately enough, could the actuator execute the task reliably enough, could the system generalize beyond its training distribution. That constraint hasn't disappeared, but it has been joined — and in some domains, superseded — by a second constraint: can the system be certified as safe enough to deploy at scale around people, cargo, and critical infrastructure, in a way that satisfies insurers, regulators, and enterprise risk committees simultaneously.

This matters because certification bottlenecks behave completely differently from capability bottlenecks. Capability improves continuously and somewhat predictably with more data, better models, and iteration cycles — the kind of curve venture investors are comfortable underwriting. Certification is discontinuous. A robot or vessel can sit fully capable and commercially ready for months or years waiting on a standards body, an insurer's actuarial tables, or a government agency's rulemaking calendar. NVIDIA building Halos in partnership with sensor makers (Infineon, NXP, STMicroelectronics, Texas Instruments), software vendors, and inspection bodies isn't just a product decision — it's an attempt to manufacture certainty around a certification pathway that would otherwise remain a black box for every humanoid company trying to sell into Amazon, GXO, or Toyota facilities. The IMO's MASS Code does the same thing at the level of nation-states: it converts years of "regulatory scoping exercises" into an actual, dateable safety instrument that shipowners and autonomy vendors can build commercial deployment plans against.

What this unlocks: enterprise buyers who have been running humanoid and autonomous-vessel pilots for 18–24 months now have a credible answer to the question "how do we defend this to our insurer and our board." That answer was previously "trust the vendor's internal testing," which is a much harder sell than "this system is certified against IEC 61508 and ISO 13849" or "this vessel operates under an IMO-recognized safety code." Expect the gap between pilot-stage and commercial-deployment-stage Physical AI companies to widen sharply over the next two quarters, with certification-ready platforms pulling decisively ahead of technically comparable competitors that lack the paperwork.

What this forecloses: it raises the bar for new entrants. A well-funded humanoid startup with a technically impressive robot but no path to IEC/ISO certification, or an autonomous-vessel company operating outside the MASS Code framework, is now competing on a fundamentally worse footing than it was six months ago — regardless of how good its underlying AI is. Certification infrastructure, once it exists, tends to favor incumbents and well-capitalized entrants who can afford the compliance overhead, and tends to squeeze out under-capitalized challengers who can't.

Builder/investor action: if you're evaluating a Physical AI company for investment or partnership in Land or Maritime specifically, certification readiness should now sit alongside unit economics and technical differentiation as a primary diligence axis — not an afterthought bolted onto the go-to-market section of the deck. Ask not just "does it work" but "who has agreed, in writing, that it's safe."

The gap: Everyone is watching humanoid robot valuations and defense drone rounds. Almost nobody is watching the fact that DoD's Defense Autonomous Warfare Group — the successor to the Biden-era Replicator initiative — just proposed a budget line that goes from $225.9M to $54.6B in a single fiscal year, a roughly 24,000% increase, and that this number has received almost no sustained press coverage relative to its size. Most outlets covering it treated it as one line item among many in a broader defense budget story.

The window: This is a request, not an appropriation, and it will almost certainly be negotiated down substantially in the FY27 budget process. But even directionally, it signals that the Pentagon intends to move from "pathfinder" testing of autonomy orchestration tools (its stated FY26 function) to something resembling full-scale procurement integration across multiple vendors simultaneously. The companies currently "live, right now, testing different systems and orchestration tools for autonomy" with DAWG, per DoD's own comptroller briefing, are positioned to become the default integration layer for how the U.S. military buys and fields autonomous systems across land, sea, and air simultaneously — a cross-domain orchestration role that doesn't map cleanly onto any single vendor's current product category.

The company profile that would win: not a hardware manufacturer, and not a pure model-layer company, but a systems-integration and orchestration platform that can sit above multiple hardware vendors' autonomous platforms and coordinate them under a single command-and-control architecture — closer to what Lunar Outpost's Stargate C3 platform is attempting in Space, but for multi-domain defense autonomy. This category doesn't have an obvious public leader yet.

Why the market is overlooking it: budget line items inside RDT&E documents don't generate the same headlines as a $300M Series C announcement, even when the dollar figure is two orders of magnitude larger. Defense-tech investors are watching company-level fundraising; almost nobody is tracking DoD's internal orchestration-layer budget requests as a leading indicator of which integration platforms will matter in 18 months.

Talent Trend: The clearest through-line in Physical AI hiring right now is scarcity at the intersection of perception, deep learning, and hands-on control experience on real bipedal or legged platforms — recruiters and operators alike describe a candidate pool numbering in the low thousands nationally, with compensation for senior robotics-learning engineers roughly 40–50% higher than 2022 levels. Meanwhile, the OpenAI robotics-leadership churn is worth watching less as a hiring data point and more as an early signal that governance disagreements over defense applications of embodied AI are starting to surface publicly at frontier labs — expect this tension to recur as more labs pursue Pentagon-adjacent contracts.

Which Layer Is Accelerating Fastest: The model/intelligence layer (Generalist AI, Skild AI, and peers) continues to attract capital fastest by dollar velocity — back-to-back nine-figure rounds within months of each other. But the deployment layer — certification, safety architecture, and regulatory frameworks — is accelerating fastest by structural importance, because it's the layer that converts pilots into revenue.

Where Are the New Constraints: Certification and safety infrastructure, as detailed in this month's Cover Story, is now a binding constraint alongside capital and compute. A second, quieter constraint: manufacturing throughput. Mach's "10-to-1,000 vehicles per week" claim and Saronic's shipyard buildout both signal that production capacity, not design capability, increasingly separates well-funded companies from companies that can actually deliver.

Where Is the Next Bottleneck Shifting: Toward government appropriations discipline. Several of this month's largest forward-looking numbers — the $54.6B DAWG request, the sub-domain of NASA's Moon Base billion-dollar package — are requests and commitments that haven't fully cleared the appropriations or contracting process. The next 1–2 quarters will test how much of this headline capital converts into signed, executable programs versus getting renegotiated downward.

State of Physical AI, June 2026: The sector has moved decisively past the "can this work at all" phase and is now negotiating the much harder "can this be certified, manufactured, and paid for at scale" phase — which is a sign of genuine maturity, but also means the easy capital-raising era, where a strong demo and a good story were often enough, is giving way to a period where production capacity, safety certification, and durable government appropriations will separate the companies that compound from the companies that stall out mid-scale.

Note: Several large rounds referenced in this issue (Saronic $1.75B, Neura Robotics $1.4B, Skild AI $1.4B, Anduril $5B) closed in prior months and are referenced for context, not counted as June closings.

Physical AI Signal Report is published on the first of each month by David Cao through the Physical AI Builders Newsletter at 50builders.ai. Supported by the Physical AI Builders community and F50.

For submissions, partnership inquiries, and event information — F50 Physical AI Summit (Silicon Valley · Dallas · Austin) — visit 50builders.ai or f50.io.