A consolidated analysis of three studies: my market and regulations analysis, the first document (meeting scenario), and the second document (a map of fourteen ideas). Everything that could be verified has been checked against primary sources. Everything that remained unverified has been stated aloud.
Three sources answered different questions. The first document answered "what to say at the meeting." The second answered "what can generally be built from two companies." My analysis answered "what of this is true according to primary sources and who is already doing it." Here they are combined into one text, because separately they contradict each other in one place and diverge in priorities.
Five facts, on which the entire position rests, have been verified against primary sources: the US Treasury's proposal from October 22, 2026 (which explicitly includes both agency wrappers and physical AI systems), the FCC's ban on importing foreign-produced advanced robotics, the tariff arithmetic on Chinese servo drives, the November 10, 2026 deadline for rare earths, and three companies cited as market proof: FOUNTAIN Ventures, MISUMI Ventures, and RobCo. All three exist, all three have been verified. This is important: the weakness of the first document is not in fabricated facts.
The second document declares the WFOE bank as unopened and builds its entire sequence of plans on this: "money is not accepted anywhere." This is incorrect.The only thing that remains open is whether this account accepts dollars or if a separate foreign currency account with a different number is needed for foreign currency. This changes the plan: four ideas that the second document put in the "after the bank" queue are now available.
Part two, about competitors, and the section on honest risks are marked as internal. They are not shown to the fund. Everything else is assembled in a way that can be on the table: facts about the subject, our position, proposal, and request. The last section, "One proposal to the fund," is the page worth showing.
Paying customers: zero. Revenue: zero. The base of eleven and a half thousand accounts on the platform is synthetic and not shown to the fund. Match states directly on its page that the room is filled with agent examples, not clients. Any document that hides this will break at the first question from the fund's lawyer.
In the first half of 2026, approximately fifty billion dollars were invested in physical AI, and the robotics and physical AI market in 2026 is estimated at over thirty-three billion. The US share of global venture funding for robotics in 2025 is about fifty-two percent. It sounds like money is flowing freely. In reality, the picture is skewed, and this skew is precisely what determines what to discuss with the fund.
The second number is scarier than the first. According to industry estimates, up to ninety-seven percent of hardware startups in the US have closed in recent years. The reason is almost always the same: the team brought the demonstration to a working prototype and could not transition to repeatable production. This is not a problem of intellect; it is a problem of the workshop, supply, and batch quality.
A fund that invests in physical AI at the earliest stages lives with the same headache: its portfolio can invent but cannot produce. It shouldn't be sold "yet another robotics company." It should be sold a way out of this pain. Everything else in this document is subservient to one task: to be the one who closes the gap between demonstration and batch production.
The gap between a working prototype and a serial production batch is described similarly by several houses. Bessemer formulates it as the moment of the "second generation": a laboratory model ceases to be a toy and transforms into a product that must withstand repetition. Eclipse formulates it differently and more usefully for us: one should not invest in the best head, but in the shoulders, that is, in what this head relies on. The customer, according to their own observation, buys not the elegance of the technology, but the level of operation, reliability, and revenue.
The fund doesn't pay for the team having connections with Chinese factories. It pays for the team reaching the next investment milestone faster.
This is the definition of our value in one line. Not "we know factories," but "we reduce the time to a repeatable batch and confirmed cost price." The fund will immediately feel the difference between these two formulations.
Chinese dominance in components for humanoid robots is not an opinion, but a measured fact, and it is confirmed by several independent sources. Below are the shares that are directly related to actuators and motors, i.e., what any physical AI team purchases.
To this are added two numbers that directly concern the cost. A robot assembled entirely on a Western supply chain costs about one hundred thirty thousand dollars in raw materials. The same assembly from a Chinese supply chain is about forty-six thousand. The difference is almost threefold, and this is precisely what explains why no Western team can afford to completely ignore Shenzhen.
The stronger China is in the supply chain, the harsher the US retaliatory rules. This is precisely why the next section is not about opportunities, but about walls. Approaching an American fund with a story like “we have access to the Chinese workshop” without explaining these walls means getting the very first question for which there is no answer.
On July 27, 2026, a US interagency body recognized that all foreign-produced advanced robotics pose an unacceptable risk, and this definition was added to the FCC list. Practical meaning: new models of foreign-produced advanced robots do not receive authorization for import, sale, and marketing in the US until they receive an exception from the department. Exceptions have already been granted, but they are few: to date, two platforms are known, Husqvarna and ANSCER.
Second, less obvious and much more important for our topic: “foreign production” is considered not only a product assembled abroad, but also a product assembled in the US if the cost of domestic components in it does not exceed sixty-five percent of the total costs. From 2029, the threshold will rise to seventy-five. This means an American team cannot simply take Chinese actuators and guess whether its robot is considered domestic.
On October 28, 2026, the administration acknowledged that the import of robotics and industrial equipment poses a national security risk, but did not impose immediate tariffs and allowed one hundred eighty days for negotiations. This is the very window that should be discussed openly: rules are tightening, but not exactly today. A separate Section 232 investigation into robotics and machine tools is underway, which could lead to quotas and licenses.
Any team in the portfolio will have to answer the same question: where is the boundary of “foreign” for a robot assembled from Chinese parts. This is a question for a lawyer, and we do not provide conclusions on it. But the preparation of data for this question itself, the composition of components, the origin of magnets, the share of costs: this is work that can be done today, and it is needed by every team.
China's retaliatory restrictions concern precisely the component that is found in all motors and drives: heavy rare earths and permanent magnets. It is important not to confuse two different regimes here, because they operate on different timelines.
Licensing of seven elements on a “case-by-case” basis: samarium, gadolinium, terbium, dysprosium, lutetium, scandium, yttrium, plus metals, oxides, alloys, and permanent magnet materials. This regime has never been suspended and remains in effect.
Second package: five more elements, manufacturing equipment and technologies, and most importantly, an extraterritorial rule: a license is required for a product made outside of China if the Chinese rare earth content in it is zero point one percent or more by value.
China suspends the second package for one year, following a trade truce. The suspension is announced until November 10, 2026.
A broader political agreement is extended until January 10, 2027, but legal notifications extending the suspensions themselves for late October 2026 have not yet been issued.
The suspension period expires. If the paper doesn't come out, the zero-point-one-percent rule returns, and it even reaches the magnet inside a foreign product.
The background for this is: China supplies about ninety-four percent of the world's sintered permanent magnets. The supply chain outside China by this date covers only single-digit percentages of global demand: one factory in Malaysia, one pilot line in Utah, one alloy manufacturer in England.
This is not a forecast or a discussion, but a deadline visible on the calendar: in eight days, the supply of magnets for motors may become complicated again. To say this aloud, with a date and mechanism, means to show that we read primary sources, not someone else's summaries. Precisely against this background, the demand for honest supply chain verification is growing.
This is the most important section of the document because it defines how to talk at all. An American early-stage investment fund is obliged to ask who owns the AI being developed and where. From autumn 2026, this question has a strict answer in the form of a rule.
On October 22, 2026, the US Treasury Department issued a proposed rule to implement the COINS Act, the law was passed on December 18, 2025, published in the register on October 26, with comments accepted until December 7, 2026. The proposed rule prohibits almost all covered investments by US persons in covered foreign persons who develop any "superintelligence" system, and removes the previously existing simple notification category.
The definition directly includes "SI agent harness," meaning agent harness, and "embodied SI or physical SI," with industrial robots, bipedal and humanoid, quadrupedal, other mobile robots, autonomous vehicles, and unmanned systems listed in parentheses. Modifying or extending another's model is also considered development. Internal non-commercial use of ready-made models is excluded from the definition.
Section nine of the Ministry of Finance's clarification explicitly states that services, licensing, consulting, and procurement in general are not covered transactions unless they are used to circumvent rules. This is the subtle but legal loophole into which our structure fits: an American company owns the software, agents, contracts, and clients, while a Chinese company performs physical, non-automated work, supplier search, factory audit, batch acceptance, and logistics.
“We are not a robotics company and are not asking for investment in China. The software and contracts reside in a US legal entity. The Chinese company manually performs what cannot be automated. According to the Ministry of Finance's clarification, these are services, not a covered transaction.” Five words, spoken first, defuse the most dangerous question of the meeting.
This is not a legal opinion, and we do not provide one. Whether such a separation can truly exempt the structure from the rule and how to properly transfer assets is solely for a lawyer to decide. In a conversation with a fund, this phrase should be uttered precisely as follows: “this is a question for a lawyer, and we will bring the opinion.” Promising exemption from the rule without a lawyer means jeopardizing both yourself and the fund.
The first document in the market section itself cites FOUNTAIN Ventures as proof that the model exists. It then proposes to build exactly this model. This is its main mistake, and it is worth examining in detail: what we have here is not an idea, but an occupied category.
The fountainbuild.ai website describes a coupling that is hard not to recognize. Two nodes: San Francisco, where the founders, labs, and capital reside, and Shenzhen, where manufacturing resides. Between them, a loop of three phases: define in San Francisco, build in Asia, return to raise a round in San Francisco.
Going to a fund that knows physical AI better than us, with the same two-city bundle, but without capital, without an accelerator, without a million shipped units, and without a team from Apple and Tesla, means becoming second with fewer resources. And note: FOUNTAIN has already resolved the very issue with jurisdictions that the second document considered our hidden trump card, and moved it to a separate node, Singapore, where data is labeled and contracts are concluded. We haven't built this yet.
The next player answers a different question: what if a component manufacturer itself opens a fund and gives the portfolio access to its supply chain from day one. On September 17, 2026, MISUMI Americas announced a fifty-million-dollar fund, Oakland and Tokyo, sole contributor: the MISUMI group itself. The fund expects to gather twenty to thirty early-stage companies, checks from five hundred thousand to one and a half million, and to enter alongside leading venture houses, rather than lead a round. Manufacturing support is declared as the investment thesis itself: not a passive check, but a manufacturing partner from the first prototype to mass production.
Nearby is an example that has been around for many years and therefore carries more weight. HAX, part of SOSV, maintains a sourcing and engineering team in Shenzhen for its hardware startups. This is not a promise, but a working practice within the venture model, and it proves the demand for a “Chinese table” as a standard fund service.
Manufacturing as the fund's thesis ceased to be someone's guess and became an announced strategy with money behind it. This means that a conversation about manufacturing support in the fund's office will not be new. What should be new is what we bring on top of it, and on top of it we have only one thing, but a real one: verification and signature.
The third link of the first document, RobCo, is verified and correct, but it proves the wrong thing. On January 29, 2026, Munich-based RobCo closed a Series C for one hundred million dollars at a five hundred million dollar valuation; Lightspeed and Lingotto from the Exor group led the round, with Sequoia, Greenfield, Kindred, Leitmotif, and The Friedkin Group participating. In total, the company has raised about one hundred sixty million since its inception. Manufacturing facilities in San Francisco and Austin have been open since 2025, and the money is going towards expanding deployments and presence in the US.
The meaning for us is different. RobCo is a company that builds robots, which is exactly the side the fund is investing in. It is not a bridge between two countries. To cite it as proof of a "bridge model" means to substitute the argument: it proves that robots are being paid for, and at the same time reminds the fund that its portfolio competes with well-funded players who have both capital and their own assembly.
Sequoia participated in the round, and Lightspeed led it. If the fund with which a meeting is scheduled operates in this same environment, it is certainly familiar with both RobCo and FOUNTAIN. This is an argument in favor of not reinventing already known companies for them, but rather talking about what they lack.
Neither of the two documents answers the question the fund will ask in the first minute: why its team would call us instead of pressing a button with a vendor they already use. Between "Chinese table" and "digital manufacturing" stands a dense row of active companies.
| Player | What it does | How it solves the pain |
|---|---|---|
| Fictiv | digital manufacturing, orders for machining and casting through an interface | quick "make a part" button, within MISUMI's orbit |
| Xometry, Protolabs | networked factories, instant price based on uploaded file | prototype in days without factory correspondence |
| Dragon Innovation | engineering support for transition to production | that very 'bridge', only consulting |
| PCH International | production management in China for Western brands | factories, assembly, supply, years of practice |
| Seeed Studio Fusion | small batches of electronics from Shenzhen | first hundred boards without opening your own legal entity |
| procurement agencies | hundreds of intermediaries, five-eight percent commission on price | cheapest entry, weakest moat |
There is another layer, closest to our topic, and also not named in any of the documents: agent procurement is already funded. Didero raised thirty million in Series A in February 2026 for procurement automation for manufacturers. Keychain raised thirty million in August 2025 for AI-powered supplier search. That is, the idea of 'an agent sends requests and compares offers' is not free, and approaching a fund with it requires knowing they've already heard it from others.
The bridge is occupied, the middle is dense, agent procurement is funded. Three of the fourteen ideas from the second document are discarded precisely for this reason, not a legal one. A narrow strip remains where we have no competitor: supply chain verification, provable by documents and signature, and there is already demand for it, created by the FCC ban and the Treasury Department's proposal. The next section is about this.
Many have access to Shenzhen, and this is proven by the previous section. No one has verification. This is not a slogan, but our own field measurement, published on a live page and open for re-verification by anyone interested.
This measurement should be read as follows: the industry is failing not on technologies, but on decisions. No one forbade machines from reading; it's just that no one decided that they were allowed to. This means the product here is not a website redesign, but a door with owner's rules and a question log. This is exactly what we have already built and what can be checked via the link without an account and without a key.
We don't say "we will build infrastructure." We say: here's a measurement, here's a raw file, here's a check that runs on your laptop and doesn't call us. An investor can verify this in their office, and the verification doesn't end with anything bad, because the statement is honest and narrow.
Agent traffic today is about two-tenths of a percent of sessions, and Gartner expects more than forty percent of agent projects to be canceled by the end of 2027. We need to say this ourselves and first. If an investor says it, they will be right, and we will look like we were hiding the main risk of our category.
Of the fourteen ideas in the second document, three remain, and they form not a list of services, but a single funnel for a Western team that needs China. They pass the first layer themselves, no one can give them the second, and our Chinese company closes the third by hand.
The first layer is bought with a button, the third is sold by hundreds of intermediaries for five to eight percent. No one sells the middle layer, and demand for it was not created by us: the FCC ban and the Treasury Department's proposal have turned supply chain purity into a mandatory document. Plus, according to the Treasury Department's clarification, an investor has their own obligation to conduct reasonable and diligent research before a deal, meaning this report is needed by them personally, not just their teams.
The three products are not invented; they are open at their addresses and can be re-verified. This is important because the entire document is built on the promise not to rebuild, but to use what already works.
| Product | What it does today | Honest limit |
|---|---|---|
| Commons | one key to 400+ models from 59 providers, skills as thin files, agent card with what they can and cannot do, spending limit per agent, failed run is not charged | this is an environment for agents, not a procurement pipeline; there is no pipeline in production |
| Match | finds the intersection of two parties without publishing lists: values are transformed before transmission, only transformed values are compared, both parties learn exactly the intersection | the room is filled with agent examples, not clients, this is written on the page itself |
| Connect | sixteen catalog checks, the gateway lives inside the client's systems, each step of the deal is signed by the buyer and seller, signature verification works offline and does not call us | traffic is low, the share of agent sessions is about 0.2 percent; orders are not accepted in the first month |
Connect is the middle layer of the funnel: a signed record that a batch has been verified and what exactly has been verified. Match is a way not to disclose specifications when we connect a Western team with a Chinese factory, and this is the first concern of any engineering team. Commons is where agents from both sides live. Exactly these three things and not a single fabricated one.
Neither “platform,” nor “ecosystem,” nor “marketplace.” Not a single word from this set. Connect directly on the page distances itself from the warehouse, from the marketplace, and from a copy of contract prices, and repeating these words after us means going against our own product.
This is the main practical conclusion of the entire analysis. Separation is not a legal formality, but a condition under which a conversation with an American fund is even possible. Software, agents, contracts, and clients in the American company. The Chinese company manually does what cannot be automated.Legal entity for software, agents, contracts and clients: ARLI, Inc., a Delaware corporation. One question remains open: where the dollar account sits and whether a separate foreign-currency account is needed before the first funds arrive.
First: that a US legal entity by itself changes the country of origin of a product or removes tariffs. Second: that we remove the structure from under a rule; a lawyer decides this. Third: that a client's drawing is protected by our reputation; it is protected by a contract and a document handling regime, and this should be stated in the contract, not by the word “trust.”
If only one line from the entire document is remembered, let it be this one. It is honest, narrow, and verifiable, and it differs from what any Shenzhen intermediary would say.
Robots are becoming autonomous, but their supply chains are not. We measure whether the trade surface is machine-readable and sign every step of the transaction. Your teams still go to Shenzhen for components, and from July and October 2026, they will need to prove the cleanliness of the chain. We are the layer that makes this provable: we find a supplier without disclosing others' specifications, verify them on-site with a signed record, and manage the procurement.
Physical AI needs a machine-readable industrial supply chain. The robots are becoming autonomous. The supply chains behind them are not.
This is the only phrase from the first document that I would keep verbatim. It is based on measurement, not on a promise, and therefore there is nothing to refute it in the office.
It's important to start from an honest place: today, we do not have a single paying client for these lines and not a single verified rate. Everything below is either a market benchmark or an offer. No number here is a promise of revenue, and they should be stated exactly as such.
| Line | How they charge | Market benchmark | What's needed to start |
|---|---|---|---|
| Supplier dossier | fixed price per report, then subscription for monitoring | market price not established; for major players, it's a corporate subscription | report template, access to Chinese registries, partner lawyer for wording |
| Supply desk | retainer plus percentage of procurement | commission five to eight percent of the price, retainer from three hundred to five thousand dollars per month | contract with an American client, supplier map across five categories |
| Audit and acceptance | per visit or man-day | audit three hundred to eight hundred dollars, inspection one hundred to three hundred fifty per man-day | checklists for actuators and mechanics, partnership with a laboratory |
| Match Room for the fund | fund platform service or participant subscription | price not set | dictionary of intentions for robotics, one fund, five to ten companies |
| Chinese legal entity service | retainer and percentage of turnover | price not set | account exists; the question of account currency and currency control remains |
All numbers mentioned are taken from the public price lists of other companies and are provided as a market benchmark, not as our price. Our price has not been validated by any transaction. If the fund asks "what have you already sold at these prices," there is only one correct answer: nothing, and we state this ourselves.
The order here is fundamental. Starting with investments in our structure is not possible; this is the most difficult and most contentious conversation. One must start with what solves the fund's own problem.
The fund introduces us to two or three teams that have a task for prototypes, product composition, supplier search, or batch quality. We conduct an analysis and provide a specific manufacturing plan. They pay for the analysis and support; for procurement, there is a transparent part billed to the client. This is the fastest money line and the simplest entry to the fund.
Together with one portfolio team and several factories from Shenzhen, we run a full cycle: find, verify, deliver. The outcome is a measurable reduction in time for collecting proposals, preservation of trade secrets, and a digital history of decisions. They pay by subscription or per workflow.
Each new fund deal requires an analysis of the Chinese part of the supply chain before signing. This includes a supplier dossier, cross-referencing with lists, origin of magnets, and FCC ban flags. Here, we work for the fund, not its team, and here we have no competitor with the same on-the-ground access and the same signature on output.
A share in your projects, a joint studio, investments in a Chinese legal entity, and anything that the second document itself marks as difficult. These are schemes with conflicts of interest and export control issues, and the second document rightly advises against starting with them.
The offer should be small and verifiable. Thirty days, one team, one product node, clear output. Such a pilot can be agreed upon at the end of the meeting without a board of directors and without the fund's legal machinery.
| Week | What we do | What's the output |
|---|---|---|
| first | select a product node or specification, sign an NDA and define work boundaries | agreed pilot subject |
| second | search and initially verify three to five suppliers, including a site visit | shortlist with reasons for rejection for each |
| third | collect proposals in machine-readable form, compare price, lead time, MOQ, materials | summary table and signed record for each move |
| fourth | shortlist, cost calculation with duty and delivery, decision on the next step | batch cost figure and lead time |
They are all measurable without our involvement, and all four are of interest to the investor himself. Time to relevant proposals is what he already calculates in his portfolio cycles. The share of those who passed verification is what reduces his risk. The spread of price and lead time is how he explains to his contributors why hardware investments are dangerous.
One next step, stated aloud and recorded: either an introduction to the founder, or a technical call with the team, or the transfer of an anonymized production request. Not "let's stay in touch," but a specific name and a specific day. A meeting without a stated step is considered unsuccessful, even if the conversation was pleasant.
Do not start by talking about yourself. Let him say what he believes in, what problems he sees, and what prevents his teams from growing. Here we learn which of the three steps he actually needs.
Briefly about the base: an operating company in Shenzhen, an account, direct work with factories. An American company in the process of opening. No numbers, no promises.
Readability check: 94 out of 189 sites do not respond to the machine, sixteen of them don't even declare it. Hence the problem of chain purity and our signed door.
Pilot. Ask directly: is there one company in the portfolio with a live production task. If so, name the weeks and deliverables.
One next step with a name and a date. And one exit question: where in the portfolio does the Chinese part of the chain sit and who verifies it.
The third, fourth, and sixth. They determine whether our infrastructure is applicable to his portfolio at all. If the portfolio's chain is not tied to China and the fund does not help teams with production, then this entire document is not relevant to him, and it's more honest to find that out at the first meeting.
One false legal claim derails the entire conversation with the fund's lawyer, and the fund's lawyer will read us more carefully than a partner. Therefore, the lines are drawn in advance and out loud.
A signed record proves signatures and bytes, not the truthfulness of a string. If the seller signed that a batch was checked, the record will only prove that he signed it. It is better to name this limitation yourself than to hear it from an investor as an accusation.
| Regime | Essence | Who is affected in our story |
|---|---|---|
| US capital outflow effective January 2, 2025 | prohibition or notification for investments by US persons in Chinese persons engaged in AI, semiconductors, quantum; services, licenses, and procurement are generally not covered | any fund investments in a structure that includes a Chinese legal entity |
| COINS, proposal from October 22, 2026 | prohibition of almost all covered investments in Chinese AI system developers, including agent wrappers and physical systems; refining another's model is considered development | our software on a Chinese legal entity; hence the rule to keep software and contracts in the US |
| FCC, effective July 28, 2026 | advanced foreign-made robots do not receive authorization for import and sale in the US; a product is considered domestic if domestic components exceed 65 percent | portfolio teams; demand for supply chain analysis and for the question of where to assemble the product |
| China's export control on rare earths | licenses for seven elements from April 2025, never suspended; second package with 0.1 percent rule suspended until November 10, 2026 | motors, drives, magnets; the most critical deadline in the discussion |
| US Lists UFLPA, BIS, 1260H, SDN | presumption of forced labor, export restrictions, procurement lists | verification of each supplier; the main driver of demand for dossiers |
| China Data | important data requires security assessment before export; research data and images may fall into this category | data collectors and training farms, so we are not taking this line |
| China's Currency Control | capital into a Chinese legal entity goes through a special currency account after registration | any idea requiring money within China |
Do not read the table aloud. Keep it as an answer to an objection: as soon as the fund asks "how does this align with the regulations," name one regime, one deadline, and one phrase about a lawyer. The table exists so you don't have to search for the answer in your head under pressure.
The second document lists fourteen possible lines and honestly states its assumptions. Below is the same numbering so the documents can be compared line by line, but with a verdict after everything we've checked.
| № | Line | Verdict |
|---|---|---|
| 3 | dossier and supplier traceability | we take first: demand is created by the FCC ban and the Treasury's offer; we have no competitor with land access and a signature |
| 1 | supply table for the portfolio | we take second: under the first layer, this is revenue and quick entry to the fund |
| 2 | audit and batch acceptance | we take: cheap entry, covers the risk of missed defects; insurance is needed |
| 5 | Match room for the fund's portfolio | only as a free pilot: no clients, value appears with the number of participants |
| 9 | certification coordination | as a module to the first layer, not sold separately |
| 13B | reverse search for commands for the fund | possibly later, price to be discussed with the fund |
| 4 | Connect for Chinese manufacturers | later: selling to Chinese requires work inside China |
| 7 | Chinese models for robotics teams | no: worst signal for a Western fund, and falls under the new rule |
| 6 | procurement agent | not at this meeting: this is an agency wrapper, only within a US legal entity, and competitors already have money |
| 10 | Chinese legal entity service | later: account exists, currency control issue is open |
| 8 | desk and laboratory in Shenzhen | capital-intensive, not for now; FOUNTAIN already has this built |
| 14 | own accelerator | no: a fund and capital are needed |
| 11 | robots from China to Europe | no: in the USA it's closed by FCC prohibition, and the fund invests in their competitors |
| 12 | data farm and teleoperations in China | no: worst signal for the fund, and research data runs into security assessment |
He rated line 10 and line 4 low on speed precisely because of the unopened account. The account is open, so both lines are raised. Line 5 remains in place not because of the account, but because of the honest absence of clients in the Match room.
An honest list of gaps. It's needed not for a report, but for a conversation: for each point, the correct answer is "we are checking this," not a fabricated number.
An investor reads a list of unverified items as a sign of discipline, not weakness. A fund that cannot name its gaps differs from a fund that lists them, exactly as a founder does in a pitch.
Resolve the account currency issue. Open an American company and submit bank documents. Create a supplier dossier template and free-analyze three to five publicly known Chinese suppliers as an example. Measure twenty to thirty catalogs of Chinese component manufacturers using the same methodology and publish.
One to two pilots with portfolio teams, if introductions are made. First retainer through an American company; client pays factories directly until the account's currency side is resolved. Release the first dossiers as signed records.
Fixed price per supplier and per product composition. Offer the fund a format of "Chinese supply chain analysis before a deal." Populate the Match room with the first suppliers and decide who pays, the fund or the participants.
Verification that no statement from the three sources was lost. Links to sections by name, because numbers shift during editing.
| Source statement | Where verified | Outcome |
|---|---|---|
| China holds over 60 percent of the humanoid robot supply chain | China in the chain: numbers | confirmed, several independent sources |
| FOUNTAIN Ventures is proof of the model | Who is already doing what the first document proposed | confirmed, and this makes the model busy |
| RobCo raised 100 million in January 2026 | Manufacturer side | confirmed, Series C, valuation 500 million |
| MISUMI Ventures, 50 million fund, September 2026 | Capital that comes with manufacturing | confirmed, announced September 17, 2026 |
| The Ministry of Finance proposes to prohibit investments in AI developers, including agency wrappers and physical systems | Capital withdrawal rules | confirmed verbatim, four independent analyses |
| FCC banned the import of advanced foreign-made robots | US Wall | confirmed, domestic component threshold 65 percent |
| Tariff stack on servo drive 37.5 to 42.5 percent | US Wall | confirmed by the arithmetic of three duties |
| Rare earth suspension expires November 10, 2026 | Chinese export control | confirmed; April licenses were not suspended |
| Bank account of Chinese legal entity not opened, money not accepted anywhere | How to read this document | refuted: account opened, Bank of China, 757545002208 |
| Match is filled with agent examples, not clients | What we already have | is confirmed, written on the product page |
| None of the verified websites allow the machine to request a catalog | Where our true advantage lies | is confirmed by our field measurement; 16 out of 189 publish only rules for agents |
Eleven statements were checked, ten were confirmed, one was refuted, and this refutation is in our favor: it removes the most severe restriction of the second document. Three questions remained unchecked and are listed in a separate section: account currency, the status of our software under the new rule, and the fund's name.
Robots are becoming autonomous. There is no supply chain behind them. Your teams still go to Shenzhen for actuators, gearboxes, and magnets, and from July and October 2026, they will need to prove their chain is clean. We make this provable: we find a supplier without revealing others' specifications, verify them on-site, and sign every step of the deal. Software and contracts are with an American company. The Chinese company manually does what cannot be automated. We are not a robotics company and do not ask for investment in China. We ask for an introduction to one or two teams in your portfolio and thirty days to measure the benefit.
I am not a person who opens an American company to sell Chinese manufacturing. I am a founder with an active operational base in Shenzhen and a signing layer above the trading surface. Two companies are not a business; they are two operational nodes. The fund is a partner that provides access to tasks on which this layer is tested.
Physical AI needs a machine-readable industrial supply chain. The robots are becoming autonomous. The supply chains behind them are not.
ARLI, Inc., a Delaware corporation. Questions: hq@arliwork.com.