Analysis of the First Public Financial Report in Humanoid Robotics: Agility Robotics' Books on the Eve of Its IPO

By | Smark Robot Edition

On November 5, 2026, Nasdaq will ring the bell for a company.

Agility Robotics, ticker AGLT, will be introduced this way — the world's first pure-play humanoid robotics company to list in the United States and to have real commercial deployments.

"Real commercial deployments" is meant to separate it from companies that have only demo videos.

"Listed in the United States" is meant to separate it from our peers who have already listed on the STAR Market.

After these three exclusions, not much is left.

Full-year 2025 confirmed revenue was $1.782 million.

Against a pre-money valuation of $2.5 billion, that is a price-to-sales ratio of over 1,400x. About 63.5% of it ($1.1319 million) came from related parties, and sales to non-related parties were only $650,100.

The trust account shrank from about $424 million to about $164 million.

When the two sides signed on June 24, the stated basis was "assuming no redemptions," with the trust able to contribute about $424 million.

By the October 30 announcement, only about $164 million actually remained.

SPAC shareholders redeemed more than 60%, and total fundraising fell from "over $620 million" to about $465 million.

This is the first time the embodied intelligence industry has been pushed under the harsh light of the public market and forced to answer questions it had never been asked before.

Who are this company's competitors?

Every argument about humanoid robot commercialization ultimately comes down to the same anchor: comparing the hourly cost of a machine with the hourly wage of a human.

Agility itself calculates it this way.

At the October 6 investor day, the chief financial officer laid out the full framework.

A Digit leased under RaaS (robotics-as-a-service) costs about $8,500 per month, or about $102,000 a year, plus a one-time deployment fee of about $25,000.

Over a five-year lifecycle, that generates cumulative revenue of about $500,000.

By contrast, the human labor cost is a fully burdened hourly wage of about $30.5.

The robot's hourly cost is about 60%, 50%, or even less of human labor.

Humanoid robots have already beaten humans on cost.

Let us look at the data from GXO's Flowery Branch site in Georgia.

The site went live in mid-2024 and was the world's first formal commercial humanoid robot deployment and the first RaaS contract.

By November 2025, Digit 4 had completed a cumulative milestone of 100,000 tote transfers at this site, with task accuracy of about 98%.

November 2025 span ≈ 17 months ≈ 510 days 100,000 ÷ 510 ≈ 196 per day

A skilled warehouse worker's output on similar motions (pick—carry—place) is usually on the order of dozens of items per hour, or hundreds per shift.

Even if all of the site's output were attributed to the machine, one site's daily output would only equal one human worker's single-shift workload.

On an "equal workload" basis, Digit's current cost may be 2 to 7 times that of human labor.

This range is a sensitivity analysis rather than a point estimate — because no company has ever disclosed how many boxes a single machine moves per hour.

GXO's CEO said at the August earnings call that the company had run 45 humanoid robot pilots and had "not yet achieved ROI on humanoid robots," estimating that it would take a few more years. More than two weeks later, GXO still signed up two new sites.

So as a rational buyer, why keep placing orders?

On October 26, Agility and GXO announced an expansion of their partnership to two e-commerce fulfillment sites in North Carolina and Tennessee, starting with Digit 4 hardware in early 2027 and migrating once Digit 5 is commercialized.

Meanwhile, GXO is also testing Apptronik's Apollo and Reflex Robotics' humanoid robots in its own facilities.

What GXO is buying is a ticket onto a learning curve.

They know today's unit economics do not hold, and they also know this category will be cheaper and more capable within two or three years. Now is the time to accumulate operating experience, on-site data, and safety processes before the curve steepens.

Although customers are buying "the future," the supplier must also have the capability and must survive until that future arrives.

A dedicated sorting machine only sorts all day, and sits idle when done.

A humanoid robot can in theory: move totes in the morning, depalletize in the afternoon, check machines in the evening, and do quality inspection and palletizing at night.

Digit 5's official task list includes depalletizing, machine tending, kitting and sequencing, quality inspection, and palletizing — the only ROI path out for this category.

Annual effective working hours per machine ÷ annual calendar hours per machine = calendar fill rate

(annual calendar hours = 8,760 hours)

Known: 65,000 hours, distributed across 9 customer facilities, as of May 2026

→ Calendar fill rate ≈ 8%–27%, with the center roughly around 15%

How many units were placed at each site and how long they ran was not disclosed. So an external estimate from public reports—"the number of Digits that have truly entered pilot or revenue-generating deployments is in the tens"—was used for cross-checking.

So this is only an order-of-magnitude judgment.

But the time a humanoid robot actually works in a year may be less than 20% of the time it exists.

From this, AGLT's thinking may have been too naive:

The commercial proposition of humanoid robots is not how many people's wages to replace, but filling up a machine's daily work schedule.

When Digit 5 was released, one specification was identified by many reports as a performance regression: single-charge battery life dropped from about four hours on Digit 4 to ninety minutes.

It is indeed a step backward.

But looking at charging time together, the conclusion is completely the opposite—a full charge takes only about nine minutes, and the run-to-charge ratio improved from about 2:1 to about 10:1.

At about 99 minutes per "run + charge" cycle, effective working time per day can exceed twenty hours.

Digit 4: 4 hours per charge, run-to-charge ratio ≈ 2:1

→ Upper limit of effective working hours per day is about 16 hours

Digit 5: 90 minutes per run, 9 minutes charging, ratio ≈ 10:1, 99 minutes per cycle

→ About 14.5 cycles in 24 hours × 90 minutes ≈ 21 hours

A very typical engineering trade-off—sacrificing depth per run in exchange for all-day density.

The theoretical upper limit of the calendar fill rate was raised from about 67% (16/24) to about 88% (21/24), but the upper limit does not equal the actual value.

The higher the upper limit, the sharper the question "why is the actual value still so low" becomes—because the bottleneck is no longer in the battery life dimension.

Battery life is no longer the main contradiction constraining the calendar fill rate. The next real round of competition is in "task scheduling and autonomy."

A $300 million bet agreement

Digit 5 has locked in more than $300 million in multi-year orders.

Reading the S-4 filing together with the investor day content, the structure is as follows:

From a single, undisclosed customer, and that customer is explicitly described in the S-4 as a related party.

Corresponding to about 1,000 Digit 5 units, in the form of a three-year RaaS contract, with the service period spread over four years.

For each unit the customer orders, Agility must issue it 453 shares of common stock warrants, totaling 453,000 warrants, vesting gradually as the robots are actually deployed.

The robots need to be able to execute certain use cases, and each use case unlocks a new batch of robots. The throughput target for each use case must be agreed item by item with the customer.

The company is still evaluating the accounting treatment of the transaction, and the final transaction price has not yet been confirmed.

The investor day changed to early customers receiving machines in the first half of 2027, with general availability not until the end of 2027 or even early 2028.

The six items taken together are in fact one thing: a conditional contract with a cap of $300 million, spread over four years, unlocked item by item based on capability, with equity incentives attached.

The customer is a related party, meaning it is most likely a shareholder, or a party closely related to a shareholder.

The customer gets the order and also gets the warrants.

This customer is Agility's stock trader, with part of its compensation coming from equity appreciation.

If part of the payment is deemed to be a sales incentive related to the equity arrangement rather than independent consideration, it may need to be deducted from revenue rather than counted as revenue.

The "revenue quality" of the $300 million still needs another layer of discount.

Note: The above is only an interpretation of the structure of publicly disclosed terms. The S-4 itself notes that the company is "still evaluating the accounting treatment of the transaction," and the final characterization should be based on the company's subsequent disclosures. No judgment is made here about the motives of any party; it only points out what is worth questioning in revenue recognition.

The situation Agility faces: the customer (GXO) publicly says it has not yet recouped its investment, it recognized only $1.78 million in revenue in 2025, and the capabilities its product promises (fence-free operation, multi-tasking) have not yet been validated at scale.

If it had signed a hard contract of "1,000 units, three years, unconditional delivery," then once Digit 5 fails to meet the throughput target for a certain use case, breach risk, refund risk, and reputational risk would all hit it at the same time.

But "unlocking by use case" structurally shifts the risk onto time—

If it can be done, unlock the next batch, revenue grows naturally, and the customer is satisfied.

If it cannot be done, no unlock, no need to deliver machines beyond its capability, no returns or impairment, just slower revenue.

The customer buys the right to exercise in batches according to the other party's actual ability to deliver, and Agility sells a growth path that will not overextend itself.

For the entire industry, this is a new contract paradigm worth studying.

In the past, the robotics industry signed pure sales contracts, with risk on the buyer.

Now it is beginning to sign "capability" contracts, with risk sliced thin and shared by both sides over many years.

The industry is finally beginning to use contractual tools to manage uncertainty.

RaaS turns AGLT into a leasing company

RaaS is the key to understanding Agility, and also the starting point for understanding its risk exposure.

At Investor Day, CFO Michael Beer laid out the unit economics in full:

Digit single-unit five-year economic model (CFO disclosure, product basis)

Revenue side: RaaS monthly fee of about $8,500 (≈$102,000/year) + about $25,000 deployment fee, totaling about $500,000 over five years (about $535,000 on an S-4 basis)

Cost side: materials $150,000 + deployment about $15,000 + software and maintenance about $15,000/year × 5 years = about $240,000 cumulative over five years

Product gross margin: about 50% initially, hardware payback in about 1.5 years

Scale effects: annual production of 1,000 units → gross margin about 70%, payback < 1 year; annual production of 10,000 units → about 75%, payback < half a year

Materials cost curve: Digit 4 currently about $125,000 → about $150,000 at Digit 5 commercial launch (more expensive) → about $75,000 at annual production of 1,000 units → about $30,000 at annual production of 10,000 units

Three cost-reduction levers: in-house cycloidal actuator redesign (about 25% of BOM), standard-parts supply chain partnerships, volume tiered discounts (contributing about a 20% reduction)

Excludes: company-level expenses. The CFO model assumes 2026 R&D of $115 million and sales and administrative expenses of $60 million.

The reason is that the added sensors and safety compute unit make it larger and heavier; safety comes at a cost.

Their gross margin is on a "product basis," not a company basis.

They deducted materials, deployment, and maintenance, but did not deduct R&D, sales, or administrative expenses.

In 2026, those two items add up to $175 million, against 2025 revenue of $1.78 million.

The two differ by nearly a thousand times.

Under the RaaS model, the robots stay on Agility's own balance sheet.

Customers do not buy them outright; Agility must first spend money to build the machines, then slowly recoup it through rent.

The CFO explicitly stated that the deployed fleet needs separate financing, with a target of securing debt funding at a cost below 10%, and listed "being able to access the debt capital markets" as one of the reasons for going public.

At a BOM of $150,000, the hardware cost of deploying a 1,000-unit fleet is $150 million.

Add deployment, field service, remote operations centers, spare parts, and the maintenance system, and the roughly $465 million total raised in this offering could well be more than half consumed by these 1,000 units.

This money is purely life support, sustaining operations by relying on the capital markets staying continuously open.

Both management and the auditor Grant Thornton noted that, excluding this merger financing, there is substantial doubt about the company's ability to continue as a going concern, and it cannot be determined whether it can continue normal operations within one year after the financial statements are issued.

So is the $2.5 billion valuation too expensive?

The CFO's installed-base scenarios are: 2,500 units corresponding to about $250 million in annual revenue, and 10,000 units corresponding to about $1 billion in annual revenue.

But he added a qualifier—"no deployment date attached."

So whether the $2.5 billion valuation is expensive depends entirely on which basis the market chooses:

On an AI/software company basis (price-to-sales): looking at the scenario of 10,000 units corresponding to $1 billion in revenue, $2.5 billion is only about 2.5x forward PS—absurdly cheap.

On an equipment leasing company basis (discounted installed capacity + balance sheet structure): asset-heavy, starting from negative gross margin, reliant on external debt financing, with unit payback starting at 1.5 years—absurdly expensive.

This dispute over basis will directly determine the valuation models of all the companies queuing up behind.

If the market forces a switch back to the "leasing company" basis, then every company centered on RaaS will have to recalculate its balance sheet.

Amazon invested in Agility as early as 2022 through its Industrial Innovation Fund and conducted multiple rounds of Digit testing in its own warehouses—a name mentioned in every article introducing Agility.

But according to media reviews of SEC filings, as of June 2026, Amazon is no longer listed among Agility's active commercial deployment customers, and the earlier pilot has ended.

Agility's response was that it has "completed multiple pilots" and said it hopes to continue advancing the partnership after Digit 5 launches.

Amazon is the world's largest and most warehouse-automation-savvy owner of the scenario.

It is not a lack of money, not a lack of scenarios, and not a lack of engineering capability.

They tried for four years, then called a halt.

SEC filings still list Amazon as a deployment or customer case; public information is insufficient to judge the exact current status of the two sides, nor can it be inferred from this that Amazon is waiting for the next-generation product.

It merely points out that the name is still there, but the deployment status has changed.

For Amazon, in its own highly optimized warehouses, the optimal solution may simply not be humanoid at all.

Amazon's warehouses have been optimized to the extreme for human-machine collaboration—shelving, traffic flows, workstations, and conveyor belts have all been redesigned around existing automation.

In this kind of environment, the efficiency advantages of specialized equipment are amplified, while the versatility advantage of humanoids is compressed.

Incidentally, this brings up a very important strategic judgment:

The value density of humanoid robots is directly proportional to "the degree to which the site is designed for humans," and inversely proportional to "the degree to which the site is optimized for machines."

This has a direct implication for the customer strategies of all players: do not go after scenarios that have already been thoroughly eaten up by automation.

Signal Two: Foxconn Has Entered

Foxconn (Hon Hai) led a PIPE of about $200 million and sent the head of its central strategic investment department to attend the October 6 investor day.

But they are simultaneously a shareholder, a potential contract manufacturer, and a potential customer.

What is Foxconn betting on?

I believe it is not that Agility will win, but that "someone will definitely win in this category, and the winner will need me to manufacture it."

This is a typical supply-chain positioning logic, betting on the certainty of the manufacturing link itself—no matter whether Figure, Agility, or someone else ultimately wins, mass production of humanoid robots will sooner or later have to fall to a contract manufacturing system with precision manufacturing and cost-control capabilities.

But for Agility, they are handing over their most valuable asset to a partner that is both a shareholder and a potential competitor.

Once Foxconn masters the mass-production process for humanoid robots, Agility's vertically integrated moat will be diluted.

And one of the core advantages Agility currently claims externally is precisely its self-built U.S. domestic manufacturing assets.

Political pricing has overwhelmed cost pricing

Digit has nearly 6,000 parts, about 75% of which are supplied by U.S. domestic suppliers, and the complete machine is assembled at the RoboFab factory in Salem, Oregon, with a designed annual capacity of up to 10,000 units.

By the common sense of manufacturing, this is bad news.

75% domestic sourcing means giving up the global cost lowlands, especially giving up the Asian supply chain that is already highly clustered in motors, reducers, batteries, and structural components.

At the commercial launch of Digit 5, the material cost per unit was about $150,000, while Unitree's first-generation flagship H1, also a full-size bipedal robot, had an official complete-machine price of less than $90,000.

But in the 2026 U.S. market, this bad news instead made the company a model of how U.S. manufacturing can avoid being strangled at the choke point.

On July 28, 2026, the U.S. Federal Communications Commission (FCC) issued DA26-786, adding "foreign-produced advanced robotic equipment" to the controlled list, explicitly covering humanoid and quadruped robots with environmental perception, autonomous obstacle avoidance, AI computing, and network connectivity capabilities.

Equipment placed on the list in principle cannot obtain the FCC equipment authorization required to enter the U.S. market.

An assessment by a local government department further pointed out that the new rules use production location and the value share of U.S.-made components as the criteria for determination, requiring the domestic component share of end products to exceed 65%, rising to 70% in 2029.

Note: "65%/70% domestic component share" is an analytical assessment of the policy by a local government department, not a direct statement in the original FCC text, and the official FCC text should prevail.

Meanwhile, the industrial data on the Chinese side is as follows:

According to data announced at the 2026 World Robot Conference press conference, China accounted for about 90% of global humanoid robot shipments in 2025, with more than 330 product models.

And according to a July 2026 report by the U.S. Information Technology and Innovation Foundation (ITIF), U.S. companies accounted for less than 5% of humanoid robot shipments in 2025.

The United States has used administrative means to artificially create a protected demand pool in an industry where it lags by a large order of magnitude.

As a result, a considerable portion of the $2.5 billion pre-money valuation is not a technology premium, but the scarcity of being "the only compliant pure-blood U.S. domestic target."

CEO Peggy Johnson explicitly stated in an interview that the FCC ban is not only not a problem for Agility, but actually beneficial to its market expansion.

A company voluntarily presenting regulatory barriers as a competitive advantage shows that it itself knows where its value comes from.

One phrase repeatedly emphasized in this transaction is—the only publicly listed pure humanoid robotics company in the United States with active commercial deployment.

This is AGLT's biggest single-point risk, far greater than technology risk.

Once the policy environment loosens, or is substantively circumvented through methods such as third-country assembly, the scarcity will evaporate instantly, and we cannot see this in the financial statements.

Two Temperatures on Opposite Sides of the Pacific

Bringing the lens back to China, in the same week the capital temperature of the embodied intelligence track formed a sharp contrast with that side.

Unitree Technology listed on the STAR Market (688836.SH) on August 19, 2026, at an issue price of RMB 150.80 per share, with a post-issuance market value of about RMB 60.993 billion and fundraising of about RMB 6.099 billion; revenue in 2023–2025 was RMB 159 million, RMB 392 million, and RMB 1.708 billion respectively, with net profit of RMB 288 million in 2025.

Agibot Innovation: According to Caijing, citing investor sources, a new round of financing of about RMB 1–1.5 billion was recently launched at a valuation of about RMB 50 billion; under HKEX rules, a pre-listing financing completed after filing must be held for a full 120 days before listing, so listing within the year is basically hopeless. Agibot officially disclosed 2025 revenue of about RMB 1.05 billion.

Weiyi Intelligent Manufacturing: Originally scheduled to list in Hong Kong on October 28, it announced on October 25 a delay of the global offering and a full refund of subscription payments. Only one cornerstone investor was introduced this time, subscribing about HK$15.7 million. According to market statistics, the IPO break rate on the Hong Kong stock market in the third quarter of 2026 reached 45.2%, with a full-year average of 21.9%.

In addition, some media reported that regulators had issued informal "window guidance" to some investment banks and investment institutions, raising the approval threshold for IPOs of humanoid robot companies and paying more attention to whether order revenue is genuine and sustainable. (This claim has not been formally confirmed by any official source; it is a media paraphrase, and this article only provides it as background, not as factual evidence.)

Agility is going the SPAC route, anchoring on already-deployed paying customers and pipeline.

The SPAC mechanism means it does not have to lay out its revenue structure, gross margin, and customer concentration in the sunlight the way a traditional IPO does.

Chinese companies going the traditional IPO route anchor on revenue structure and going-concern capability.

How much of the revenue is related-party transactions? How much is sustainable repeat purchasing? What proportion is research procurement and government orders? Can gross margin improve with scale?

These questions are indeed hard to answer at the current stage of the industry.

And this difference reveals a conclusion that holds for all Chinese peers:

Once you enter the public market, the power of narrative shifts from progress metrics to "structural metrics."

You can impress the primary market with working hours and tote counts, but you cannot impress the secondary market's continuous disclosure requirements.

Should Chinese peer decision-makers just watch the show?

The first thing to do

Right now, immediately switch the internal KPI from units delivered to "annual effective working hours per unit × unit output cost ratio."

Unit count is the easiest metric to inflate.

Related-party procurement, research procurement, government projects, and demonstration rentals can all push up unit counts, but they do not reflect any economics.

Effective working hours and per-unit throughput, by contrast, will be asked about no matter where you raise money or list in the future.

What is most worth learning from Agility is not its technology, but a metric like "65,000 hours / 100,000 totes" — traceable, reusable, and able to show an incremental slope every quarter.

This is an extremely clever disclosure design: it replaces economics that cannot prove themselves with progress that can.

By contrast, disclosures by Chinese peers are generally too crude.

If external communications could give average daily effective operating hours per unit and average daily task completions per unit, that would be more persuasive than a hundred demo videos.

Second, make multi-skill reuse rate a core KPI

The only ROI path for humanoids is one machine with multiple capabilities.

So internally there should be a clear metric measuring, for the same machine within a year, the number of different task types actually performed and the distribution of time.

This maps directly to calendar fill rate, and directly to why customers should pay a premium for generality.

Raise the per-unit calendar fill rate from its current low level to 50% and make unit economics work.

Third, go seize the power to define safety standards

A Schaeffler field engineer said at Investor Day that, due to the lack of ISO safety standards, we find it necessary to install fencing around any deployed humanoid robot.

She also added that use cases that would block aisles or equipment entrances were excluded.

In one sentence, she killed the core selling point of Digit 5's "fenceless collaboration."

Fenceless equals cost reduction, equals reduced deployment friction.

The fence itself costs money, the space the fence occupies costs money, and use cases that cannot be done because of the fence are a direct loss.

So the implementation of safety standards is currently the single biggest cost-reduction lever in this industry, more direct than cutting BOM from RMB 150,000 to RMB 30,000.

And the current situation is: ISO/CD 25785-1 (the international safety standard for humanoid robots) is still at the committee draft stage.

Agility is participating in the development of the ANSI/A3 TR R15.108 technical report and ISO 25785-1.

NVIDIA Halos for Robotics was released on June 22, 2026, at the Automate conference in Chicago, with Agility as its first launch partner, adopting the IGX Thor and Halos Core architecture.

Standards are being written, and whoever writes the standards will be the one defining market access for the next decade.

Chinese players' presence in the standard-setting process is severely mismatched with their roughly 90% share of shipments.

When going overseas in the future, this will become a threshold harder to circumvent than tariffs.

Fourth, think clearly about the separation of building and using

US labor is expensive and currently has policy protection, making it more suitable to put robots to use first.

China has a strong supply chain and low manufacturing costs, making it more suitable to build robots first, and this division of labor will not change for a considerable time.

But one cognitive error must be guarded against: do not mistake the advantage in building for the story of "using."

Being able to build humanoid robots at low cost and in large volumes, and being able to use them in a specific scenario to generate positive ROI, are two completely different capabilities requiring completely different organizations.

After November 5 is the mark of adulthood

After November 5, the market will no longer be pricing just one company, Agility.

It will become the reference frame for all the companies queuing up behind it.

Figure AI's valuation is reported by the media to be on the order of USD 39 billion, and Apptronik's Series A is said to have cumulatively exceeded USD 935 million at a valuation of about USD 5 billion — these numbers used to exist only in the past.

Once one company's real financials are made public, everyone's valuation will have to be recalibrated.

US Securities and Exchange Commission (SEC) · Churchill Capital Corp XI 8-K (2026-10-30): composition of total fundraising of about USD 465 million, 11/2 vote, 11/4 closing, 11/5 listing, 180-day lock-up, conversion of all existing shareholders' shares

US Securities and Exchange Commission (SEC) · Churchill Capital Corp XI 8-K (2026-06-24): pre-money valuation of USD 2.5 billion, total fundraising of more than USD 620 million, Foxconn leading the PIPE, shareholder list including NVIDIA / Amazon / SoftBank Vision Fund 2 / Schaeffler

Agility Robotics official website: Digit has about 6,000 parts, about 75% sourced domestically in the US, RoboFab annual capacity of up to 10,000 units

US Federal Communications Commission (FCC) · DA26-786 notice (2026-07-28): foreign-produced advanced robotic equipment included in the controlled list

Shanghai Stock Exchange announcement (2026-08-18) · Unitree Technology STAR Market listing trading announcement

Weiyi Intelligent Manufacturing · HKEXnews "Latest Information on the Global Offering" announcement (2026-10-25)

About this article: 1. The factual parts all come from the publicly verifiable sources above; the speculative parts are the author's personal views. 2. All calculations in the text are the author's own derivations based on public figures, and the process is fully shown. 3. This article does not predict the price movement of any security, does not constitute investment advice, and does not constitute an evaluation of or accusation against any company. Neutral wording has been used for unverified matters. 4. The accompanying images are generated images of an abstract metaphorical nature, contain no text, and do not refer to any specific entity, scenario, or individual. 5. The transaction is still subject to the shareholder vote on November 2 and closing on November 4, and the final terms are subject to the company's official announcement.