Physical AI Funding Tracker: What Has the Capital Actually Produced?

The short answer

Physical AI capital is accelerating faster than public operating evidence. The eight companies in this tracker now span private rounds, strategic investment, and Unitree’s priced Shanghai initial public offering. Their disclosed capital milestones range from a $1.7 billion round with little product detail to task-focused systems already reporting field output.

The important distinction is simple: capital raised is an input, not a deployment result. A funding announcement can verify the amount, investors, and stated use of proceeds. It cannot by itself verify product availability, customer demand, manufacturing output, uptime, autonomy, safety, or return on investment.

This tracker compares:

• Atoms, which raised $1.7 billion while keeping product and deployment details broad;
• Humanoid, which raised $152 million and named Bosch as a contract-manufacturing partner;
• Walden Robotics, which launched with approximately $300 million and claims factory deployments;
• Apptronik, which raised $350 million and established customer pilots plus a Jabil manufacturing relationship;
• Agility Robotics, which raised $150 million in 2022 and later entered a multi-year commercial Digit agreement;
• Gritt, which launched with $32.4 million and reports completed solar-installation work;
• Sereact, whose ongoing Series B reached $116 million as it reported more than 200 live systems;
• Unitree, whose Shanghai IPO was priced at RMB 150.8 per share, implying a valuation near RMB 61 billion and planned proceeds of about RMB 6.1 billion.

Status reflects sources reviewed on August 7, 2026. Valuations are included only when attributable reporting disclosed them. Company-reported output remains labeled as such. Reuters independently reported Unitree’s offer terms, but the primary exchange filing was not retrieved for this update.

Funding and milestone comparison

Company	Disclosed financing event	Product and manufacturing signal	Strongest commercial or operating evidence	Main evidence gap
Atoms	$1.7B round led by Andreessen Horowitz, reported July 2026	Broad physical-world software and industrial productivity scope	No named customer or deployment milestone located in the reviewed announcement	Robot architecture, product, customer, schedule, operating target
Humanoid	$152M Series A at a reported $1.35B valuation	Bosch named as contract-manufacturing partner	Industrial trials and preorders are reported, but comparable shipped-unit and recurring-operation evidence is not public	Contract definitions, shipped systems, named customer results, runtime
Walden Robotics	Approximately $300M at launch	Toyota Research Institute spinout with large-behavior-model and robot platform work	Company says robots were deployed in Toyota facilities and were working with customers	Robot count, task results, customer confirmation, interventions, runtime
Apptronik	$350M Series A announced in February 2025	Jabil agreement covers Apollo manufacturing and factory deployment exploration	Named pilot relationships and defined candidate tasks	Shipped production units, recurring throughput, intervention, uptime
Agility Robotics	$150M Series B announced in April 2022	RoboFab manufacturing facility and Digit production program followed	GXO confirms a multi-year commercial agreement; Agility reports 100,000+ totes moved	Fleet size, system-hours, interventions, uptime, economics
Gritt	$32.4M total launch financing	Robotic arms and perception mounted on existing construction equipment	Two field systems reported; company reports 30,000+ solar panels and 18+ MW installed	Named sites, duration, intervention, audited productivity and safety
Sereact	Ongoing Series B reported at $116M after Zalando investment	Cortex software works across multiple warehouse robot configurations	Company reports 200+ live systems and 1B+ production picks	Per-customer breakdown, time window, errors, intervention, independent audit
Unitree	Shanghai IPO priced at RMB 150.8 per share; about RMB 6.1B in planned proceeds	G1 and H1 humanoid platforms plus established quadruped products	Public product pages establish commercial access; no normalized customer-site humanoid operating record is established here	Revenue and shipment mix, customer confirmation, runtime, intervention, ROI

These rows are not a ranking by company quality. They show how differently capital and evidence can line up. A smaller financing total can accompany field output, while a much larger round can precede a clearly disclosed product.

Unitree: public-market capital is not a humanoid deployment metric

Reuters reported that Unitree priced its Shanghai initial public offering at RMB 150.8 per share, implying a company valuation near RMB 61 billion and planned proceeds of about RMB 6.1 billion. Those figures establish the offer terms as independently reported. They do not establish how much revenue comes from quadrupeds versus humanoids, how many G1 or H1 systems have been accepted by customers, or how those systems perform in recurring work.

This distinction matters because Unitree already has more visible commercial access than many humanoid developers. Its product pages list the G1 with a starting price and present the H1 family through a commercial product path. Our humanoid availability tracker classifies those products separately from production deployment evidence.

The IPO should make a different set of questions more important, not less. Readers should watch for audited revenue mix, shipped units by robot family, gross margin, manufacturing capacity and utilization, customer concentration, warranty and service obligations, research spending, and evidence from named customer workflows. A market valuation cannot substitute for task attempts, successful output, system-hours, interventions, downtime, incidents, and customer economics.

Evidence classification: independently reported IPO pricing and planned proceeds, plus first-party product availability. The primary exchange filing was not retrieved for this update, and normalized recurring humanoid deployment evidence is not established by the financing event.

Next milestone to watch: public filing data that separates product mix and shipments, followed by a named humanoid customer reporting task, fleet, active hours, throughput, intervention, uptime, safety, and economics.

A milestone framework for robotics financing

Funding coverage often compresses several distinct states into one story. This tracker keeps six milestones separate.

1. Financing closed. The company and investors completed a transaction. The amount and round type should be attributable.
2. Product defined. A named hardware or software product has a documented form, task scope, and generation.
3. Manufacturing pathway established. A facility, contract manufacturer, tooling plan, or production agreement exists.
4. Commercial access opened. Buyers can purchase, lease, subscribe, or enter a defined pilot or service agreement.
5. Customer operation confirmed. A named operator confirms where and how the system is being used.
6. Recurring performance disclosed. Fleet, duration, output, interventions, downtime, maintenance, incidents, and economics are reported with useful denominators.

A company can advance on one track without advancing on another. A contract manufacturer does not prove shipments. A preorder does not prove delivery. A pilot does not prove recurring production. A cumulative task count does not prove uptime without fleet and time denominators.

Atoms: the largest round, with the fewest measurable product details

TechCrunch reported that Atoms raised $1.7 billion in a round led by Andreessen Horowitz, with Bain Capital, Fifth Wall, and Uber participating. Travis Kalanick and Ben Horowitz also published primary posts confirming the announcement.

Public descriptions center on physical-world software and industrial productivity. In the material reviewed for this tracker, Atoms did not identify a specific robot architecture, product generation, named customer contract, deployment date, manufacturing target, or operating benchmark.

That does not imply that no product or customer work exists. It means the financing evidence is much stronger than the public product evidence. The correct current classification is financed, with product and deployment scope not publicly detailed.

Next milestone to watch: a named product or system architecture tied to a customer, task, delivery window, and measurable operating target.

Humanoid: manufacturing intent is clearer than field performance

Reuters reported a $152 million Series A at a $1.35 billion valuation. The report says Bosch would serve as a contract-manufacturing partner and contribute hardware design, production, and supply-chain expertise.

That relationship matters because manufacturing is a real bottleneck for humanoid systems. It is still a pathway, not output. A manufacturing agreement does not reveal completed units, production yield, delivery acceptance, bill of materials, reliability, or service burden.

Reports around the round also cite preorders and industrial deployments. Those terms need definitions before they can be compared with commercial operations. A preorder may be refundable or conditional. A deployment may be a demonstration, test, pilot, paid installation, or recurring workflow.

Evidence classification: financing and contract-manufacturing relationship are attributable. Commercial shipment and recurring operating evidence remain incomplete.

Next milestone to watch: a named customer confirming delivered robot generation, paid status, bounded task, operating duration, interventions, and output.

Walden Robotics: capital arrived with deployment claims, but denominators remain absent

Walden Robotics launched from Toyota Research Institute with approximately $300 million in financing. Its first-party launch says the company began working with customers after its January 2026 spinout and that robots had been deployed in Toyota factories.

That is more specific than a product-free financing announcement, but it still leaves the operating record incomplete. The launch does not publish robot count, factory location, task attempts, completed units, intervention rate, runtime, downtime, safety incidents, or customer-side performance confirmation.

Our full Walden deployment evidence review separates the observed factory video and company claims from recurring production proof.

Evidence classification: financed company with a defined robot-learning platform, customer work, and company-reported factory deployments. Recurring production performance is not established by the reviewed sources.

Next milestone to watch: named operator confirmation with one robot generation, one task, system-hours, successful throughput, and human-assistance records.

Apptronik: financing, pilots, and a manufacturing path are three different signals

Apptronik announced a $350 million Series A in February 2025, co-led by B Capital and Capital Factory. The company said the funding would support scaling Apollo, customer deployments, and operations.

Jabil later announced a collaboration with two distinct parts. The companies would explore manufacturing Apollo at scale, and they would deploy Apollo in Jabil operations for tasks such as inspection, sorting, kitting, lineside delivery, fixture placement, and subassembly.

This is useful because it names a manufacturing partner and candidate workflows. It should not be upgraded into completed mass production or recurring autonomous deployment without delivery counts and operating results.

Evidence classification: substantial financing, a defined humanoid platform, named pilots, and a contract-manufacturing pathway. Public recurring-production metrics remain unavailable in the cited sources.

Next milestone to watch: completed Apollo units, customer acceptance, scheduled versus active hours, successful task rate, intervention, maintenance, and a commercial contract definition.

Agility Robotics: earlier capital followed by a commercial operating record

Agility Robotics announced a $150 million Series B in April 2022. The company said the round would accelerate research, manufacturing, and commercialization of Digit in logistics and warehouse environments.

Later evidence shows movement beyond that financing objective. GXO announced a multi-year robotics-as-a-service agreement for Digit at its Flowery Branch operation. Agility subsequently reported that Digit had moved more than 100,000 totes in live commercial deployment.

This is the strongest capital-to-operation progression in the comparison because a customer confirms the commercial relationship and the vendor reports cumulative output. It is still incomplete. The public sources do not give fleet size, system-hours, totes per robot-hour, interventions, fault recovery, uptime, maintenance, safety incidents, or service economics.

For the broader context, our production evidence comparison normalizes Digit against five other bounded physical AI workflows.

Evidence classification: financed, manufactured, commercially contracted, and operating in a bounded workflow. General-purpose autonomy and normalized reliability are not established.

Next milestone to watch: fleet and time denominators for the tote count, plus intervention, uptime, maintenance, and contract performance data.

Gritt: smaller financing, direct field output

Gritt launched with $32.4 million in pre-seed and Series A financing. Its system mounts robotic arms and perception equipment on existing construction machines to unload, carry, and position solar panels.

Gritt reports more than 30,000 panels and 18 MW installed, plus 2.8 GW contracted. Those figures describe different evidence states. Installed output reflects completed work as reported by the company. Contracted capacity describes future obligations, not completed installation.

TechCrunch reported two field systems and said it spoke with one unnamed customer. The company also makes productivity, cost, breakage, and safety claims, but the reviewed sources do not publish the protocols and denominators needed to audit them.

Evidence classification: financed and field deployed for a bounded construction workflow. Completed output is company-reported, with limited independent corroboration of field presence.

Next milestone to watch: named project sites, active system-hours, interventions, verified panel throughput, breakage, incidents, and a customer baseline.

Sereact: software scale claims with missing per-site detail

Sereact reported that Zalando joined its ongoing Series B, bringing the round to $116 million. Sereact supplies Cortex, a robot intelligence platform used across warehouse picking and returns configurations rather than one fixed robot body.

In the same announcement, the company reported more than 200 live systems across Europe and more than one billion real production picks. It named customers including Daimler Truck, Mercedes-Benz, BMW, MS Direct, Active Ants, DeltiLog, Rohlik Group, and Austrian Post.

Zalando’s investment is a meaningful strategic customer signal. It does not independently validate the installed-base and pick totals. The release does not break down systems or output by customer, define the measurement period, report failed picks, quantify human correction, or provide an outside audit.

Evidence classification: financed software provider with company-reported production scale and named customers. The size of the operating claim is substantial, while reliability and intervention denominators remain unavailable.

Next milestone to watch: customer-side system counts, pick attempts and successes, active hours, human corrections, downtime, and maintenance by deployment class.

What the comparison shows

1. Round size and deployment maturity are weakly coupled

Atoms has the largest disclosed round and the least detailed public product record in this set. Gritt has far less financing and reports completed field output. This is not evidence that one is a better business. It shows why financing and deployment need separate columns.

2. Manufacturing agreements are intermediate milestones

Bosch with Humanoid and Jabil with Apptronik provide manufacturing capability and a plausible route to scale. The decisive evidence comes later: completed units, yield, acceptance, delivery, serviceability, and field reliability.

3. Named customers improve evidence, but do not settle performance

GXO confirms a commercial agreement with Agility. Zalando is a strategic investor in Sereact. Jabil is both a manufacturing and deployment partner for Apptronik. Each relationship establishes something different. None should be treated as an independent audit of every vendor claim.

4. Cumulative output still needs denominators

One billion picks and 100,000 totes establish meaningful activity. Without fleet size, active hours, attempts, interventions, downtime, and maintenance, they cannot establish comparable throughput, uptime, staffing, or economics.

5. The next financing story should be a milestone story

The most useful follow-up to a funding announcement is not another valuation. It is a dated operating commitment. Companies can make their progress legible by stating what the money is intended to produce, by when, for which customer class, and with what measurable acceptance criteria.

Unitree’s IPO adds a public-market version of the same rule. Offer price, proceeds, and valuation are capital-market facts. Product mix, manufacturing output, accepted shipments, customer operation, reliability, and economics remain separate operating facts.

A due-diligence checklist after any robotics round

For each financed company, record:

1. Round facts: amount, type, close date, lead investors, and whether the figure is new capital or cumulative funding.
2. Product boundary: named generation, hardware form, software role, target task, and excluded capabilities.
3. Manufacturing state: prototype line, contract manufacturer, dedicated facility, validated process, completed units, and accepted deliveries.
4. Commercial access: research access, selected pilot, paid pilot, purchase, lease, service, or recurring contract.
5. Customer evidence: named operator, site, task, start date, and customer-side confirmation.
6. Operating record: fleet, scheduled hours, active hours, attempts, successful output, interventions, resets, downtime, and maintenance.
7. Safety record: risk assessment, applicable standards, incidents, near misses, stoppages, and corrective actions.
8. Economic record: integration cost, support labor, service price, utilization, performance guarantees, and renewal or expansion.
9. Next milestone: a dated, falsifiable outcome rather than a broad intention to scale.

Use our robot safety standards guide when a financing announcement makes safety or human-collaboration claims. Standards apply to defined applications and risk assessments, not to funding totals.

Evidence watchlist

The next disclosures that would materially change this tracker are:

• Atoms naming a product architecture, customer, task, and dated operating milestone;
• Humanoid publishing completed-unit and named-customer operating evidence under the Bosch manufacturing pathway;
• Walden providing customer-side task, fleet, runtime, and intervention records;
• Apptronik reporting Apollo manufacturing output and recurring customer performance;
• Agility normalizing Digit’s tote count by fleet, active hours, interventions, uptime, and cost;
• Gritt identifying project sites and publishing verified productivity, intervention, breakage, and safety denominators;
• Sereact providing customer-side system counts, attempt-level pick results, corrections, and active hours;
• Unitree disclosing revenue and shipment mix by robot family, manufacturing utilization, named humanoid customer results, and normalized operating denominators.

Verdict

The eight capital stories fall into different evidence classes. Unitree has a priced public offering and commercially presented products, but the IPO does not answer humanoid reliability or customer economics. Atoms is strongly financed but publicly under-specified. Humanoid and Apptronik have manufacturing pathways without disclosed scaled output. Walden pairs substantial capital with company-reported deployments. Agility connects earlier financing to a customer-confirmed commercial workflow. Gritt and Sereact report meaningful field or production activity, but their headline totals still lack the denominators needed for reliability and economics.

Funding can buy time, talent, tooling, compute, factories, and market access. It cannot buy proof. Proof arrives when a named system performs a bounded task for a customer, repeatedly, with fleet, time, intervention, safety, and cost records that other people can inspect.

Frequently asked questions

Which physical AI company has raised the most money in this comparison?

Atoms has the largest single disclosed round in this comparison at $1.7 billion. The financing is verified, but the public record reviewed here does not identify a specific robot architecture, named customer commitment, deployment schedule, or operating target.

Does a large robotics funding round prove commercial demand?

No. A financing event proves that investors supplied capital under agreed terms. Commercial demand needs separate evidence such as paid contracts, named customers, renewals, shipped systems, or recurring production work.

Which funded company has the strongest operating evidence?

Agility Robotics has the strongest combined commercial and operating record in this set. GXO confirms a multi-year robotics-as-a-service agreement, and Agility reports more than 100,000 totes moved by Digit. Fleet size, runtime, interventions, uptime, and economics remain undisclosed.

Are manufacturing partnerships the same as robot shipments?

No. A contract manufacturer or scaling agreement can establish a production pathway. It does not establish completed units, customer deliveries, yield, cost, reliability, or recurring field use unless those outcomes are separately reported.

What should investors track after a physical AI funding round?

Track the product generation, manufacturing readiness, named customer commitments, commercial access, shipped systems, system-hours, successful task output, interventions, downtime, safety record, and the next dated milestone. Keep forecasts separate from completed results.

Does Unitree's IPO valuation prove that its humanoid robots work in production?

No. The IPO pricing establishes a public-market financing milestone and an implied company valuation. It does not establish customer-site uptime, task throughput, intervention rates, safety performance, or return on investment for G1, H1, or any other Unitree robot.