Our Investment in Maven Robotics

We’re announcing our lead investment in Maven Robotics’ Series A. Maven designs, builds and deploys intelligent robots for logistics and manufacturing, and has come out of stealth with the secret to building general-purpose robotics for enterprise customers.

It’s not enough to have the best AI. You need the best solution, and achieving that requires a team of experts across the entire robotics stack.

That conclusion came from two questions: why has an industry with this much talent and capital produced so few robots doing real work, and what does it take to be one of the exceptions?

In this piece, we’ll answer both questions as we lay out our Maven investment thesis. We’ll also explain:

• What it takes to close an enterprise customer

• The robotics industry’s research problem

• Maven’s origin story and the team behind it

• Maven’s strategy for building a general-purpose system that delivers value from day one

1. Robots have a job to do

Somewhere in all the excitement, the industry has forgotten that robots are here to do work. A robot is a piece of capital equipment. It gets bought to do a specific job and it gets judged the way every other piece of capital equipment is judged, on output, reliability, and cost.

In the long-term robots will be ubiquitous, today the most attractive work is not evenly distributed. It’s found in large enterprises.

Figure 1: Where the work is: Homes and small enterprises are unstructured, low density and price constrained. The repeatable, high volume work sits in large enterprises. [1]

You can crudely segment the total addressable market of labor into consumer and commercial. Consumer spaces (homes) are unstructured and unsupervised, and most households don’t have the budget or work for equipment that costs tens or even hundreds of thousands to produce. This doesn’t make for a great initial target market, and small enterprises are much the same. The real prize is large enterprises:

• Density: A large customer does the same task thousands of times a day and can spend hundreds of millions on that one task. An SME does fifty different things a few times each.

• Repeatability: One enterprise buyer means one integration, then the same deployment across dozens of sites. An SME is a one-off sale every time.

• Capital and process: Enterprises have an automation budget and balance sheet to afford capital purchases. SMEs are often more limited.

The enterprise problem

Just because large enterprises are attractive doesn’t make them easy. Their scale can make them bureaucratic, slow, and risk averse.

The biggest trap for robotics startups that go whale hunting is the innovation project. You get to sell to people whose job is to be curious and enthusiastic about playing with new tech. The problem is that most innovation projects don’t lead to robots at scale doing work. They are often set without clear goals, in lab conditions, where success is a demo plus a press release. Converting innovation projects into volume contracts can be as hard as winning the volume contract from scratch.

Winning a real contract that scales is completely different. Enterprise sales require managing ten stakeholders, each with different priorities, but the same desire not to get fired for taking on unnecessary risk. The contracts come with strict SLAs, penalties, and liabilities for failure. When an enterprise finally signs, it is because a company can clear a far higher bar on four dimensions:

Achieving this across one robot is hard. Across a fleet it is harder. Across every site globally, harder still. That bar is why so much of this industry is stuck in pilots, and why clearing it takes more than good robots.

2. What it takes to solve the problem

What makes clearing this bar so challenging is that you cannot succeed without multidisciplinary excellence. Enterprises want to buy an outcome, and a single small failure anywhere in the value chain kills performance. This heavily favors full stack companies who control value delivery end to end, and it is where model-only companies are at a disadvantage. Multiple suppliers create an accountability gap, and systems designed independently create a performance gap.

Companies that crack large enterprises must master the whole complex system that is a robotics business. Being expert at one thing is not enough. You need to be expert at eight.

The biggest challenge with innovation projects is that the goal posts are constantly being shifted. Every high reliability industry knows this as the march of nines.

In many industries, 99.99% is the goal. Each nine you add costs roughly ten times the effort of the last, and you cannot reach the next one by repeating what you did in the last phase. It takes a different approach for nine. Your lab, a customer’s lab, a trial line, real production, then real production across every site are five steps up a curve, not five repeats of the same work.

Figure 2: The march of nines. Each environment demands roughly an order of magnitude more effort than the last, and most innovation projects stop before real production.

Which raises the obvious objection: plenty of large customers run robotics at scale.

The reality is that these tend to be traditional robotics companies, which manage this complexity by constraining their way to simplicity. They constrain the work to solids with bounded tolerances, and they constrain the environment, the lighting, and the task itself. General purpose robots promise to remove those constraints, which makes them the hardest robots to refine to industrial standards and is why they tend to end up in dead end innovation projects.

This isn’t to oversell the challenge. It is possible to build highly scalable general robots that crack the enterprise but the ones that win have industrialization at the core of their culture.

3. Robotics has a research problem

Our contrarian take is that the robotics industry needs to leave its academic culture behind. The field is dominated by research organizations, and for robotics to crack enterprise customers it must reject the lab and embrace the factory.

The typical shape of a leading robotics company is PhD founders, a highly specialized skill set focused on one model architecture, and progress measured against demos and academic benchmarks. That structure is necessary for pushing the frontier of what is possible, and it has been enormously successful in traditional AI. It just doesn’t work for full stack robotics. Look at how the incentives differ.

Research cultures lend themselves to innovation projects, but they lack many of the skills required to build full stack robots for an industrial site.

Enough scene setting. Our Maven thesis can be distilled to a single statement: winning enterprise customers takes industrial teams, and this is where Maven excels.

4. The team

Maven is not a lab spinout with a few hired operators. It is an industrial company to its core. Together, Maven’s team has a combined 200+ years of experience across AI, robotics, engineering, and autonomous systems. Most of the senior team came through the same program at Apple, and every one of them is a specialist in getting advanced systems deployed at scale.

Figure 3: Programs and products members of the Maven team have worked on. Source: Maven Robotics.

Maven’s origin story

Apple spent roughly a decade on an autonomous systems project, reported to have cost more than $10bn and to have had around 2,000 people attached by the end. It was canceled in February 2024, having never shipped. Apple never confirmed the program publicly, so everything we say about it here comes from press reporting. [2]

Apple does not hire good engineers, it hires the top person in each discipline required. When the program was canceled, one of the best teams at productizing autonomous systems became available all at once.

Maven’s CEO and co-founder Hamza Derbas was an early employee on this project, and over a decade he directed 100+ engineers across the full stack of autonomous systems. With the program over, he called his brother Khalid, who had spent his career scaling businesses in private equity, with a question. Should Hamza go back to the automotive industry, or should the two brothers fulfill a lifelong dream and start a business together?

Figure 4: Maven co-founders Hamza and Khalid Derbas. Source: Maven Robotics.

Once that decision was made, Hamza was in the perfect position to choose from teammates looking for their next project. 

Maven is full of PhDs who could fit into any of the top labs, but what unites this team is the kind of frontier they work at. They specialize in what can be deployed and relied on in environments where failure is unacceptable. Arresting gear that stops a fighter jet on a carrier deck, battery systems that keep an eVTOL in the air, surgical manipulators operating inside a patient, missile defense, EUV lithography, functional safety at Apple and at Archer. That is the frontier that matters when you are building a product to meet customer KPIs at scale.

Figure 5: The Maven team. Source: Maven Robotics.

This founding team also had the advantage of having worked together for a decade. This allowed them to move fast with zero ramping. As a result, they went from founding to the customer floor in less than two years without touching an innovation project.

5. Product

Deep, then wide

Maven is going after the big prize: Physical AGI. They have built a general-purpose robotics system designed for enterprise customers, and core to their approach is a strategy we call “deep, then wide”.

Research focused companies tend to go wide first with the hope of going deep later. This means impressive breadth across tasks, work pieces, and even robot form factor achieved by sacrificing deployment KPIs. The mindset: if the breadth is there it doesn’t matter much that the robot is slow, or that it fails one time in ten. The problem, is that the last 10% is where much of the effort lives.

Deep, then wide is the other way around. You build a hardware and software platform designed to be general, then focus development on one specific job until it meets the customer’s KPIs. Only then do you move to the next job. It sounds simple, but it really isn’t. The big risk with this approach is that if you don’t build a platform with general potential, you just end up with a special purpose tool.

Getting that balance right means judging, over and over, what to constrain and what to leave general. It is exactly the judgment an industrial team has and a research organization never needed to build.

By embracing the “deep, then wide” approach, Maven’s robots master real tasks that deliver immediate impact for customers, ultimately building generality task by task.

Figure 6: Deep, then wide. Wide then deep spends its effort below the production bar. Deep then wide crosses the bar on one job, then adds breadth from above it.

Built to be built

Maven has run four hardware generations in two years, each with a defined job. Generation 1 proved they could build something fast. Generations 2 & 3 proved they could do work at the right quality and cycle time, while sacrificing on cost. Generation 4 is the full package, optimized for all 4 KPIs.

Vertical integration is central to their strategy. As we covered in our Standard Bots memo, it is what lets a company deliver a holistic solution, where every component is designed to work with the others, rather than bought and bolted together. But integrating too early kills companies. Every part you choose to build yourself is time you are not in the field delivering value.

Maven optimizes for speed to value, so they integrate on a schedule set by whatever is limiting them. In Generation 3, off-the-shelf pneumatic end effectors were capping performance, so they built their own. In Generation 4, cost is the constraint, so they are building their own actuators and moving to high volume processes like casting and injection molding.

The machine

Figure 7: The Maven machine. An omnidirectional wheeled base, a telescoping pillar and two 7-DOF arms carrying modular end effectors. Source: Maven Robotics.

From a distance, it might look like a humanoid with wheels, or an autonomous mobile robot with arms bolted on, but this is an oversimplification. Instead of copying a specific form factor, they have rethought general robotics from first principles. Up close, every element is a deliberate decision aimed at the KPIs that matter. The machine is an omnidirectional wheeled base, a telescoping pillar, and two 7-DOF arms carrying modular end effectors, running a four-layer software stack. The split underneath it is simple: the hardware is specialized for the customer’s site, and the AI is what keeps it general.

Maven’s approach to algorithm design is the key that stops their deep, then wide approach from dead ending into a specialized solution. If each job produced a bespoke program, targeting a new job would require starting from scratch. Maven’s AI stack is built the other way around: skills are abstract primitives rather than task-specific routines, so they can be recombined for new work and capability grows combinatorially rather than linearly.

Every specification point at a KPI the customer measures:

What it delivers

The real prize isn’t engineering, it’s what the engineering has achieved. Maven meets production KPIs on two use cases:

• Mixed case palletizing. Hard because the SKU mix is enormous and every case has different dimensions, weight and surface, so the robot has to pick and place precisely while it is moving.

• Tote handling. High frequency, high precision, and unforgiving of a drop.

Together, those two skills address roughly $80B a year of US labor cost, the equivalent of 1.6 million people working full time [3]. For context, Walmart, Amazon, UPS, Target, Home Depot and Kroger employ around 4.2 million people in the US between them, and a great deal of that work is moving product from one place to another [4].

The value in Maven’s approach is that they are not constrained to two tasks. The development already completed has accelerated their ability to develop five new skills higher up the complexity curve. Maven intends to launch these over the next year, not as demos but as industrially ready processes that meet the KPIs.

Each added layer of complexity widens the pool of work the platform can take on, from fine material handling through to general assembly. Global manufacturing labor is a $7T a year pool, of which assembly, fitting and fastening work is roughly $1.2T [5].

Maven’s incremental approach is what lets them expand their addressable market steadily while delivering value.

Figure 8: How deep, then wide compounds. Source: Maven Robotics [6]

This is why we believe deep, then wide is the winning approach to general-purpose robotics.

Our thesis: real-world data captured on the machine during paid work is the most valuable training input in robotics. The labs are not blind to this, but accessing this class of data is very challenging for companies with a wide, then deep strategy.

Simulation, egocentric video and other data types are substitutes, adopted because the right to capture real data must be earned. The only way to earn real world data is to build a full system that meets enterprise requirements and deploy it on customer sites at scale.

This is the secret to Maven’s approach. By focusing on value delivery, they embed themselves in their customers’ operations and earn the right to capture data the labs cannot reach:

• Deliver value from day one

• Earn trust and embed in the customer’s operations

• Collect real data from real work

• Iterate faster, on the problems that matter

• Unlock the next job, one rung harder

This cycle compounds: the deeper your integration, the more data you capture and the faster you generalize.

Why we led

Three things convinced us.

• The team: a group that has already built the hardest version of what Maven needs, in places where failure was not an option.

• The speed: four hardware generations and onto customer floors in two years.

• The culture: industrialization is not a phase Maven gets to once science is done. It is how they have worked since the first generation.

The industry has spent a decade building organizations designed to learn things. Maven is one built to ship them.

Disclosures

Maven Robotics is one of a number of portfolio investments held by RoboStrategy, Inc. This discussion of a single portfolio company should not be viewed as representative of the Fund’s portfolio as a whole, and the outcomes described here should not be taken as indicative of the outcomes of the Fund’s other investments. Information regarding the Fund’s full portfolio is available in the Fund’s filings with the SEC.

RoboStrategy, Inc. holds an investment in Maven Robotics, and FP Strategies LLC, the Fund’s investment adviser, therefore has a financial interest in, and an incentive to present favorably, the company discussed herein. The Fund led the financing round described in this article, and the fair value of the Fund’s investment in Maven Robotics is based in part on the terms of that round. A fair value derived in part from a transaction that the Fund led is subject to greater conflict than one derived from a transaction among unaffiliated third parties. Nothing herein constitutes investment advice or a recommendation to purchase or sell any security.

This article contains forward-looking statements within the meaning of Section 27A of the Securities Act of 1933 and Section 21E of the Securities Exchange Act of 1934. Forward-looking statements are not historical facts but are based on current expectations, estimates, projections, beliefs, and assumptions about the Fund, our current and prospective portfolio investments, our industry, and our beliefs and assumptions. Statements regarding Maven Robotics’ plans, product roadmap, and anticipated deployments are forward-looking statements of the company and are subject to the same uncertainties. Words such as “anticipates,” “expects,” “intends,” “plans,” “will,” “may,” “continue,” “believes,” “seeks,” “estimates,” “would,” “could,” “should,” “targets,” “projects,” and variations of these words and similar expressions are intended to identify forward-looking statements. These statements are not guarantees of future performance and are subject to risks, uncertainties, and other factors, some of which are beyond our control and are difficult to predict, that could cause actual results to differ materially from those expressed or forecasted in the forward-looking statements. The Fund undertakes no obligation to update or revise any forward-looking statements, whether as a result of new information, future events, or otherwise.

Investing in shares of RoboStrategy, Inc. involves a high degree of risk and is highly speculative. You should read the discussion of the material risks of investing in our common stock in the “Risk Factors” section of the Prospectus. You may lose part or all of your investment.

This article is not an offer to sell or a solicitation of an offer to buy any securities. Any offering of securities may be made only by means of a prospectus meeting the requirements of Section 10 of the Securities Act of 1933, as amended.

Shares of closed-end investment companies frequently trade at a discount to their net asset value (“NAV”). If shares of our common stock trade at a discount to our NAV, investors will face increased risk of loss. There is no assurance that an active trading market will develop or be sustained, or that shares will trade at or above NAV.

The Fund’s investments in private companies are valued at fair value as determined pursuant to procedures and methodologies approved by the Fund’s Board of Directors. Fair values are necessarily subjective, and there is no assurance that such values will be realized or that a value will be at or close to the price at which the security would be sold in a transaction between unaffiliated third parties.

Views expressed herein are those of FP Strategies LLC, the Fund’s investment adviser, and do not necessarily represent the views of RoboStrategy, Inc. or its Board of Directors.

Certain information contained herein has been obtained from published sources prepared by third parties and has not been independently verified. Operational information concerning Maven Robotics, including statements regarding its team, deployments, hardware development, and performance against customer key performance indicators, was provided by Maven Robotics and its officers. Statements concerning the discontinued Apple autonomous systems program are based solely on press reporting; Apple has not publicly confirmed that program or its details. Market size figures are estimates. While such information is believed to be reliable, we assume no responsibility for its accuracy or completeness.

1. Figure 1 image sources, left to right: https://inclusiveteach.com/2025/08/04/9-tips-for-a-sensory-friendly-home/; https://www.instagram.com/p/DYeaSJCjQ1k/; https://www.newyorker.com/humor/borowitz-report/mueller-rents-giant-warehouse-to-store-evidence-against-trump

   
   

2. Apple has not publicly confirmed the program. Figures cited reflect contemporaneous press reporting, principally Bloomberg (February 27, 2024) on the cancellation of the program and the approximately 2,000 people attached to it, and The New York Times on a reported spend of more than $10 billion over roughly a decade.

   
   

3. RoboStrategy estimate. The US Bureau of Labor Statistics Employment Projections programme reports 2,988,900 hand laborers and freight, stock and material movers, and 2,764,800 stockers and order fillers (2024 base year), at median annual wages of $40,240 and $37,330 respectively (May 2025 wage data). We estimate that palletizing and depalletizing account for 40% of the effort in the first occupation and 15% in the second, judgements based on operating experience in warehouse environments. Applying a 1.35x multiplier for benefits and other employer costs, which BLS wage figures exclude, gives approximately $86 billion of annual labor cost, equivalent to 1.6 million people working full time at roughly $53,000 all-in.

   
   

4. Company filings, most recent fiscal years: Walmart approximately 1.6 million US associates (FYE January 2026), Amazon more than 1 million US employees, UPS 373,027 US employees, Home Depot approximately 422,500 US associates (FYE February 2026), Target approximately 415,000 (FYE January 2026) and Kroger more than 403,000 (FYE January 2026).

   
   

5. RoboStrategy estimate. Global manufacturing value added was $17.6 trillion in 2025 (World Bank, https://data.worldbank.org/indicator/NV.IND.MANF.CD). Labor’s share of manufacturing value added, implied by national accounts data, is 29% in India (MOSPI Annual Survey of Industries, 2022-23), 47% in the United States (BEA, 2024) and 54% in the EU27 (Eurostat, 2024). We apply 40%, below the developed-market figures because the global manufacturing workforce is concentrated in lower-wage economies, giving a global manufacturing labor bill of approximately $7.0 trillion. As a cross-check from independent data, McKinsey sized global manufacturing labor at $5.1 trillion in 2015; global manufacturing value added has grown 42% since then, from $12.4 trillion, implying $7.2 trillion today, within 3% of our figure. Assemblers and fabricators account for 10.7% of US manufacturing employment, but that is a job title rather than a task, and insertion, fitting and fastening also occur within other production roles, which together account for 30.7% of manufacturing employment (BLS Occupational Employment and Wage Statistics, NAICS 31-33, May 2023, https://www.bls.gov/oes/2023/may/naics2_31-33.htm). We estimate assembly work at 17% of the global manufacturing labor bill, giving $1.2 trillion. Spread across a global manufacturing workforce of roughly 380 million full-time equivalents, that is equivalent to approximately 65 million people working full time.

   
   

6. Data hour figures shown are Maven Robotics estimates for future periods and are forward-looking statements; see Important Disclosures. The Waymo figure shown refers to rider hours, not fleet operating hours: Waymo reported that its riders “enjoyed over 3.8 million hours” in Waymo vehicles during 2025. Source: Waymo, “Delivering more for our riders in a year of incredible growth,” December 10, 2025, https://waymo.com/blog/2025/12/2025-year-in-review/.