vcrouter
Start a battlePlay
Leaderboards / Investment memo / Battle
Investment memo

Figure AI — humanoid robotics company targeting warehouse and manufacturing labor

September 13, 2026 · $0.13 for both answers and the judge

Grok 4.6 wins

On September 13, 2026, DeepSeek V4 Pro and Grok 4.6 each wrote an investment memo for the same brief, without seeing the other's answer. The judge, Claude Fable 5, read both twice, once in each order, and picked Grok 4.6 both times.

Why, in the judge's words“B is more honest about unknowns, names a fuller competitor set (Apptronik, Unitree, Sanctuary), gives a decisive evidence-gated verdict; A has date inconsistencies.”

The two answers

DeepSeek V4 Pro
$0.008 · 28s · 1,881 tokens

MEMORANDUM

To: Investment Committee From: [Venture Investor] Date: October 26, 2023 Subject: Figure AI – Series A Investment Consideration

Thesis

Figure AI represents a high-conviction bet on the physical manifestation of artificial intelligence. We are currently witnessing a bifurcation in the labor market: cognitive tasks are being rapidly automated by LLMs, while the physical world remains constrained by a shrinking, aging, and increasingly expensive manual labor pool. Figure’s thesis is that the missing link is not better software or better actuators alone, but a vertically integrated "embodied AI" stack—a humanoid form factor driven by a foundation model.

This is a fund-returner thesis because the TAM is effectively the global labor market for dangerous, dull, and dirty jobs. If Figure succeeds in deploying a general-purpose humanoid that can perform tasks at a cost-per-hour lower than a human warehouse worker, they are not building a robot company; they are building a labor arbitrage platform. The scale of that outcome dwarfs nearly every vertical SaaS or fintech opportunity currently in the market. The capital intensity is high, but the "winner-take-most" potential in the physical labor market justifies the risk.

Product & Wedge

Figure’s product is the Figure 01, a general-purpose humanoid robot. Unlike specialized robotic arms (e.g., ABB, KUKA) that are bolted to the floor in automotive plants, Figure 01 is bipedal and designed to operate in environments built for humans.

The strategic wedge is warehousing and manufacturing logistics. This is the ideal beachhead for three reasons:

  1. Structured Chaos: Warehouses are semi-structured. They are not as unpredictable as a home, but not as rigid as an assembly line. This allows for iterative learning without requiring the robot to solve "general world navigation" immediately.
  2. ROI Clarity: The value proposition is simple math. A warehouse worker costs $45k-$60k/year fully loaded, suffers from high turnover, and is subject to unionization and injury. A robot that works 20 hours a day at $5/hour equivalent is a straightforward CFO sale.
  3. Brownfield Deployment: The humanoid form factor is the wedge. Figure does not need to rebuild the warehouse. It can walk to a shelf, pick a tote, and place it on a conveyor. It fits the existing infrastructure.

Market & Competition

The market for warehouse automation is currently served by incumbents like Amazon Robotics (Kiva) and Locus Robotics, but these are "goods-to-person" or AMR (Autonomous Mobile Robot) solutions that require significant infrastructure changes. They move racks; they do not manipulate individual items in a human-like manner.

The true competitive landscape for Figure is the race to "embodied AGI":

  • Tesla Optimus: The most formidable competitor. Tesla has massive capital, manufacturing expertise, and in-house AI compute (Dojo). However, Tesla is a public company with a primary focus on EVs; Optimus is a side quest for now, and their data advantage is currently in driving, not manipulation.
  • Agility Robotics (Digit): The current commercial leader in humanoids. Digit is currently being piloted by Amazon. However, Digit’s design (bird-like legs, inward knees) is optimized for walking, not heavy lifting or fine manipulation. Figure’s design is arguably more robust for industrial work.
  • 1X Technologies (backed by OpenAI): A dark horse. They are focusing on a softer, wheeled or gentle humanoid for the home/office, but their backing by OpenAI gives them a potential software edge.
  • Boston Dynamics: The best kinematics in the world, but historically they have been a research organization (DARPA roots). Under Hyundai, they are pivoting to logistics (Stretch), but the Atlas platform is too expensive and hydraulic for mass deployment.

Traction & Business Signal

  • Capital: Figure has raised a significant Series A (~$70M) led by Parkway Venture Capital, with participation from notable tech founders and AI investors. This signals strong syndicate confidence.
  • Talent: The team is elite. CTO Jerry Pratt has decades of robotics experience (IHMC). The AI team is composed of ex-DeepMind and Google Brain researchers. This is a "best-in-class" team signal.
  • Partnerships: Figure announced a commercial agreement with BMW in early 2024 to deploy robots in the Spartanburg, SC manufacturing plant. This is a massive signal—it validates that a top-tier manufacturer is willing to test the tech in a real production environment.
  • Revenue: Unknown. The BMW deal is likely a pilot or co-development agreement, not a high-volume purchase order. There is no public data on units deployed or revenue recognized.

Risks (The 3 Deal Killers)

  1. The "Manipulation Gap" (Software is Harder than Walking): The market is obsessed with walking and balancing. Walking is largely solved. The actual bottleneck is generalized manipulation—picking up a crumpled plastic bag, a shiny metal bolt, or a soft fabric without explicit programming. If the "foundation model" for physical interaction does not emerge, Figure 01 becomes a very expensive paperweight that can walk but cannot work. We are betting on a software breakthrough that has not happened yet.
  2. Capital Intensity & Unit Economics: Hardware is hard. Tooling a humanoid for mass production requires hundreds of millions of dollars before positive gross margins are achieved. If the cost per unit cannot drop below ~$50k quickly, the ROI for customers evaporates. Figure risks being caught in a "valley of death" where they have demand but cannot finance the manufacturing scale-up without massive dilution or a strategic acquisition.
  3. The Tesla Factor: Tesla has the "full stack" advantage. They can manufacture the actuators, design the chips, train the models, and sell the robot at cost (or a loss) to gain market share, funded by their automotive cash flow. If Tesla decides to aggressively prioritize Optimus for logistics in the next 24 months, Figure will be fighting a war of attrition against a company with 100x the capital.

Verdict

INVEST (Conditional)

We should invest in Figure AI at the Series A. The market timing is right: LLMs have provided the missing "brain" architecture for robotics, and the labor shortage is a secular tailwind. The team is uniquely qualified to execute on the integration of AI and hardware.

However, this is a binary bet. We are not investing in a robotics company; we are investing in the probability that "System 2" reasoning can be applied to physical space.

The one question I need answered before wiring funds:

"When Figure 01 fails a manipulation task in the BMW plant (e.g., picking a specific hose from a bin of tangled hoses), what is the specific data flywheel mechanism that ensures the next attempt is more likely to succeed, and how does that loop differ from the 'scripted' automation of traditional robotics?"

If the answer relies on "more teleoperation" or "better prompt engineering," we pass. If the answer demonstrates a self-supervised learning loop where the robot learns from its own sensor data and failures, we invest.

Grok 4.6Winner
$0.009 · 31s · 1,392 tokens

Investment memo: Figure AI (hypothetical Series A) Humanoid robotics for warehouse and manufacturing labor. 720 words.

Thesis

Figure could be a fund-returner if it becomes the first company to ship a general-purpose humanoid that reliably does economically useful work in unstructured industrial environments at a cost below a fully loaded warehouse or factory worker. Labor is scarce, expensive, and politically constrained; a robot that uses existing aisles, racks, and tools without facility redesign captures a multi-hundred-billion TAM with software-like margins once hardware is commoditized. The bet is that end-to-end neural control plus a vertically integrated body beats both specialist AMRs and slower academic humanoids. If Figure reaches even low-single-digit thousands of units in real plants with 80%+ uptime on a handful of SKUs, the valuation path looks like an industrial Tesla, not a robotics science project.

Product & wedge

Figure 01/02 is a bipedal humanoid (~5'6", ~60 kg class) with dexterous hands, cameras, and onboard compute aimed at walking, picking, placing, and simple tool use in factories and warehouses. The wedge is not “do everything a human does on day one.” It is: occupy the same physical envelope as a person so customers do not rip out infrastructure, start with a narrow set of repetitive, injury-prone tasks (tote handling, kitting, line loading), and improve via fleet data and foundation models. Vertical integration (actuators, hands, software stack) is the claimed moat versus assembling off-the-shelf parts. Partnerships (BMW manufacturing; earlier OpenAI collaboration, later in-house models) are meant to close the sim-to-real and data loop faster than competitors who treat the body as a research platform.

Market & competition

Addressable market is warehouse + manufacturing labor substitution in the US/EU first (tens of millions of roles; fully loaded cost often $40–80k+/year). Broader “any job a human does in a built-for-humans world” is the long-term story; near-term revenue is task-specific robots sold or leased to logistics and OEMs.

Real competitors: Tesla Optimus (scale, vertical integration, Elon distribution); Agility Robotics Digit (already in Amazon/GXO-style deployments, more conservative biped); Boston Dynamics Atlas (best hardware, historically research/demo-oriented, Hyundai-owned); Apptronik Apollo (NASA/Google heritage, Mercedes interest); Sanctuary AI (Phoenix, cognitive architecture); 1X (NEO, OpenAI-adjacent); Unitree and Fourier (China cost/speed); plus AMR/cobot incumbents (Locus, Geek+, Universal Robots) that win if “good enough without legs” remains true. Figure’s differentiation is speed of iteration and “human form factor + end-to-end learning” versus Digit’s warehouse-first pragmatism or Tesla’s manufacturing scale.

Traction & business signal

Publicly known: high-profile BMW manufacturing partnership and factory footage; large later-stage rounds (hundreds of millions, multi-billion valuation) with Microsoft, Nvidia, Amazon, Bezos, etc.; demos of walking, manipulation, and some language-conditioned tasks; talent from Boston Dynamics, Tesla, Google, etc. Series A-era traction (early pilots, revenue, unit economics, reliability hours) is largely unknown from public sources. No disclosed production volumes, ASPs, gross margins, or independent third-party uptime data at the time of a hypothetical Series A. Brand and capital access are strong; commercial proof is still mostly promised.

Risks (the three that kill the deal)

  1. Reliability and cost never close. Humanoids remain 10–100× too expensive and too fragile for 2-shift warehouse economics. If mean time between failures, battery, actuators, and hands do not hit industrial numbers, customers keep AMRs and humans. This is the modal failure mode for the entire category.
  2. Tesla (or China) wins the stack. Optimus plus Tesla manufacturing and FSD-style data, or Unitree-class cost curves, make Figure’s body a me-too. Vertical integration only matters if you actually out-iterate; otherwise you are a well-funded science project.
  3. Wedge never expands. Narrow tasks get solved by cheaper non-humanoid robots or process redesign. If Figure cannot show a path from “one BMW cell” to general warehouse labor without constant teleop or custom engineering, it is a services company with robot COGS, not a platform.

Secondary risks (safety/regulation, data/IP fights, founder concentration) matter less than these three.

Verdict

Pass at a hypothetical Series A, unless the one question below is answered with hard evidence. Capital and narrative are abundant later; early rounds need proof that the robot does real work, not that it walks in a video. I would need: independent, customer-attested data on hours of productive unsupervised work in a live warehouse or plant (task list, uptime, interventions per hour, and fully loaded cost vs. human), not lab demos. Without that, this is a research bet priced like a product company. If that data exists and unit economics pencil even on a single SKU, the thesis becomes investable and potentially fund-returning.

DeepSeek V4 Pro vs Grok 4.6, overall

Across 29 judged battles, DeepSeek V4 Pro won 22, Grok 4.6 won 5, and 2 were ties. On investment memo alone: 7 to 2, 1 tied.

Compare DeepSeek V4 Pro and Grok 4.6 Judge one yourself, blind

Costs are the provider's list prices at the time of the run.