Why Humanoids Will Outgrow Every Forecast & The ROI Math Behind Humanoids in Manufacturing

Beyond headline names like Nvidia and Tesla, investors are beginning to map out the broader humanoid robotics landscape. In recent months, we’ve spoken with LPs and public-market funds who acknowledge it may feel early as a 2025 investment theme—the common view being that true scale manufacturing is still a couple of years out. Yet a growing group is already charting where to place future bets, not with the intent of deploying capital today, but to be ready by 2026–2027 when production volumes ramp, commercial contracts materialize, and cost curves bend.

These investors aren't allocating today—but they are building conviction, tracking proof-of-concept deployments (e.g., Figure with BMW, Agility with Amazon partners), and running sensitivity models on pricing, unit margins, and TAM unlock as robots shift from demo to deployment.

In our view, this is the right mental model: humanoid robotics won’t generate near-term IRR, but it will reprice entire labor-exposed verticals over the next 3-5 years. Investors would be well-served to treat this like an options strategy—keep the playbook ready, monitor manufacturing milestones closely, and be prepared to underwrite an informed position in 2026 when early movers begin to break away from the prototype pack.

“We’re going to have millions of humanoid robots soon.”
— Sam Altman, CEO of OpenAI, investor in Figure and 1X

He says in his blog,

Source: Sam Altman’s blog

Altman isn’t betting on narrow pilots—he’s underwriting platform shifts, in the same way GPT-3 was a prelude to ChatGPT. His investment in 1X reflects a view that domestic, safe, low-cost humanoids could be the next iPhone-scale moment.

Meanwhile, Nvidia’s CEO Jensen Huang has gone even further:

“The age of generalist robotics is here. These systems will perceive, reason, and act in the real world.”
— Jensen Huang

In recent interview, Elon Musk discusses the Tesla Optimus robot, predicting that humanoid robots will be the "biggest product ever" with "insatiable" demand, as everyone will want a personal robot like C-3PO or R2-D2.

Musk sets a target of producing one million Optimus robots by 2030.

What we’ll cover today:

1. Market Sizing: Embodied AI & Humanoids

   • BofA & McKinsey forecasts — and why we think they’re conservative

   • Sub-segment growth: warehousing, retail, healthcare, domestic, defense

2. Key Players & Roadmaps (2025–2028)

   • Morgan Stanley’s 100 humanoid enablers

   • Deep dive: Tesla, Figure, Agility, Apptronik, 1X, Sanctuary, Nvidia

   • How they differ, volume targets, and investor inflection points

3. Manufacturing First: Two Paths to Payback

   • Single-shift vs. multi-shift ROI

   • Payback sensitivity tables

In recent months, we’ve been hearing a consistent theme from clients: how fast can humanoids deliver ROI in manufacturing? The question isn’t whether the technology works — it’s how quickly it can translate into cost savings, efficiency gains, and scalable deployment on the factory floor.

Please note: The insights presented in this article are derived from confidential consultations our team has conducted with clients across private equity, hedge funds, startups, and investment banks, facilitated through specialized expert networks. Due to our agreements with these networks, we cannot reveal specific names from these discussions. Therefore, we offer a summarized version of these insights, ensuring valuable content while upholding our confidentiality commitments.

Market Sizing: Embodied AI & Humanoid Robotics

When it comes to sizing this market, the numbers are already staggering — and yet they might still be too small. Bank of America and McKinsey have put out forecasts that give us a starting point, but the real story lies in how quickly embodied AI is moving from research to real-world deployment.

Let’s start with the umbrella category. “Embodied AI”—the stack that fuses perception, language and action in physical machines—is projected by MarketsandMarkets to grow from $4.44B in 2025 to $23.06B by 2030 (39% CAGR). That total includes robots (humanoids, service, mobile), exoskeletons and software, and it sets the ceiling for how big humanoids can get as a sub-slice.

Cut just the humanoids and third-party estimates diverge sharply based on scope. MarketsandMarkets models a jump from $2.92B (2025) to $15.26B (2030), 39.2% CAGR, while Grand View Research’s stricter cut (hardware-heavy, narrower revenue capture) runs $1.55B (2024) to $4.04B (2030), ~17–18% CAGR. Sell-side and macro shops are more conservative on near-term dollars: Goldman Sachs frames an early U.S. adoption window at “at least ~$6B” in 10 years, largely logistics/retail led.

In a recent report, McKinsey assessed the adoption of general-purpose robotics across industries and geographies, projecting the market could reach $370 billion by 2040 in its base case. That estimate rests on linear modeling assumptions — moderate improvements in training data, hardware costs, integration ease, and cultural adoption — the same toolkit traditional consulting and banking firms have long relied on.

Source: McKinsey Report

But humanoids and AI do not follow linear curves. Their progress is shaped by nonlinear breakthroughs in intelligence, hardware design, and scale effects that compound faster than conventional models can capture. For this reason, we believe McKinsey’s projections significantly underrate the true trajectory of the market. The reality is that AI and robotics will likely accelerate adoption in ways that will surprise even seasoned forecasters.

If you ask me to put a number on it, I can’t — but what I can say with confidence is that the trajectory toward AGI will be exponential over the next two years.

In an earlier blog, we noted Anthropic’s CEO projecting that AGI could emerge as early as 2026. That prediction isn’t just a curiosity about AI timelines — it has direct implications for humanoid robotics. The closer we move toward AGI, the more powerful the intelligence layer behind humanoids becomes. Tasks that today require careful scripting, structured workflows, or narrow training data will, in an AGI-enabled world, be handled with far greater adaptability and autonomy. This means the hardware investments companies are making in humanoids today will only compound in value as the intelligence running them rapidly improves. In short, the timelines for AGI and for mass humanoid adoption are not independent — they reinforce one another. The sooner advanced intelligence arrives, the sooner humanoids move from factory pilots to mainstream deployment.

Industrial logistics remains the most immediate USD opportunity, but customer‑service and healthcare segments show the steepest growth curves as humanoids cross OSHA safety and FDA device hurdles around 2027.

We believe the consensus market estimates understate both the pace and scale of humanoid adoption. When you overlay vendor-stated timelines and production targets against the conservative CAGR models, the gap is striking. Several companies have outlined bold growth targets. For instance, a U.S. player is preparing to boost production of its flagship humanoid from roughly 1,200 units in 2025 to about 7,500 units by 2027, with long-term plans to reach 10,000 units per year at a new factory. In China, another firm is building a facility designed to produce between 1,000 and 3,000 units annually once operational. These developments underscore how quickly production lines are being established to meet the expected surge in demand.

This acceleration isn’t hypothetical—it’s anchored in capital already deployed. Production tooling, factory build-outs, and simulation infrastructure are in place, and most top players have signed anchor customers in logistics, automotive, or manufacturing. The effect is a compressed adoption curve, where segments like warehouse automation or retail service could transition from pilot to multi-site deployment in 24–36 months, not the 5–7 years modeled in traditional industrial robotics adoption. In short, this space is moving faster than most market studies assume, and investors risk being late if they wait for “proof” in published industry data.

Later in this section, you’ll see how manufacturing ROI is not only achievable but immediate. Extend that logic forward, and it’s clear why demand will accelerate — quickly outpacing the cautious forecasts put out by third parties.

Sub-Segment Analysis: Where Capital Will Flow

Breaking down the U.S. humanoid opportunity by end-market makes it clear that the first meaningful dollars will cluster in a few high-ROI niches—while later segments offer much steeper growth curves.

Take-away: logistics leads in immediate dollar capture, but retail service and healthcare offer the sharpest long-term slope once hardware cost curves fall and software matures.

Source: Morgan Stanley’s Humanoid 100

Bank of America’s report estimated in early 2025 that the hardware bill of materials (BOM) for a typical humanoid could reach roughly $35,000 per unit by the end of 2025, assuming most parts are sourced from Chinese suppliers. That estimate reflects a reference design including:

• about 16 rotary actuators with harmonic drives,

• 14 linear actuators built on planetary roller screw mechanisms,

• a dexterous hand with six degrees of freedom,

• and a vision stack combining a depth camera with LiDAR.
  The underlying assumption is that local Chinese manufacturing will drive cost efficiencies.

Looking further out, the same research suggested BOM costs could drop significantly—to the $13,000–$17,000 range per unit between 2030 and 2035. Drivers of this decline include scaling effects in production and continued improvements in actuator and sensor design. That trajectory would represent more than a 50% reduction over five years, roughly equivalent to a 14% annualized rate of cost decline.

Source: Bank of America Humanoid BOM

BofA’s forecast suggests humanoid robots will go through a gradual three-stage adoption curve—starting with industrial pilots, expanding into commercial services, and eventually reaching households. The idea is that the first wave (2025–2027) is about factory and logistics deployments where robots prove themselves on tasks like material handling and quality checks. The second wave (late 2020s to early 2030s) envisions broader commercial adoption, supported by integration with large language models to enable more natural interaction. Finally, by the mid-2030s, humanoids are expected to reach everyday use in homes and elder care, with shipments in the tens of millions.

We think BofA’s timeline is conservative. Players like Tesla, Agility, and Figure are already pushing toward scaled manufacturing and shipping commitments that could pull forward mass adoption by years. Once cost curves bend and data from industrial pilots compound into better performance, the commercial and household markets could open much faster than expected. For investors, this means treating humanoids not as a distant “sci-fi” category, but as a market that could inflect meaningfully within the current investment cycle.

Linear projections may capture today’s comfort zone, but humanoids and embodied AI will scale along non-linear, compounding curves. That means the $370B+ estimates are not ceilings — they’re just the floor.

Key Players & Their Execution Roadmaps (2025–2028)

Sizing the market is one thing — but to understand when inflection points hit, we have to look at the players building the machines and infrastructure. The roadmaps of Tesla, Figure, Agility, Apptronik, 1X, Sanctuary, and Nvidia tell us where the future bends.

When investors think about humanoid robots, the spotlight often falls on Nvidia (chips) and Tesla (manufacturing). But the real story lies in the broader value chain—what Morgan Stanley dubs the Humanoid 100. This ecosystem can be thought of in two halves: the Brain (intelligence, chips, AI models) and the Body (actuators, sensors, batteries, motion systems).

Source: Morgan Stanley’s Humanoid 100

The Brain: America’s Stronghold

The U.S. has an undeniable lead in the Brain layer. From Nvidia’s GPUs to Google’s and Microsoft’s large language models, to software and simulation platforms, the U.S. already supplies much of the intelligence powering humanoids. These companies capture investor attention because their technology is already critical to generative AI, cloud, and enterprise workloads.

But here’s the catch: while the Brain is strategically vital, it accounts for only a small percentage of the humanoid’s BOM.

Source: University of Cincinnati research

The financial upside of intelligence is leveraged across many domains, but the robotics-specific hardware opportunity is elsewhere.

The Body: China

China dominates the Body side of humanoid robotics. Roughly 60% of global suppliers of actuators, gears, sensors, and magnets are Chinese. That’s no accident—Beijing has aligned subsidies, rare-earth processing, and industrial clusters to ensure humanoid robotics are “Made in China” from the ground up.

Our clients are asking who in US is best positioned to capture it? From Tesla’s vertically integrated Optimus to Nvidia’s ecosystem-enabling GR00T and Thor, the humanoid landscape is more diverse than it appears from the outside. Understanding how these companies differ—in their volumes, timelines, and capital strategies—is essential to separating high-conviction bets from high-risk hype."

Here's how the US players currently shapes up:

RaaS (Robotics-as-a-Service) will likely be the dominant business model in the first phase of humanoid adoption, because it reduces adoption friction for customers who don’t want to spend $50K+ per unit up front. Let’s quickly look at how the highlighted in yellow RaaS landscape looks:

• Figure AI – already pitching “robots as a workforce service,”

• Agility Robotics – commercializing Digit for warehouses; strong Amazon ties suggest RaaS logistics pilots.

• Apptronik (Apollo) – pushing versatility angle (logistics + retail), likely to pursue per-unit monthly pricing.

• Sanctuary AI / 1X Technologies – both emphasize general-purpose humanoids for “labor on demand,” almost certainly RaaS-driven.

Now, let’s go one by one.

1. NVIDIA is deeply involved in developing humanoid robotics, but not in the way of building the physical robots themselves. Instead, NVIDIA is focusing on providing the AI, software, simulation, and hardware infrastructure needed to accelerate the development and deployment of humanoid robots.

   

   Unlike players focused on selling robots, Nvidia is monetizing simulation (Isaac Lab), training (GR00T), and inference (Thor). Its roadmap through 2030 includes increasingly advanced open-source models, with the Isaac ecosystem paralleling what CUDA did for GPU programming.

2. Tesla’s Optimus project is the most ambitious in scope and production scale, even if it remains somewhat behind on timelines. Built entirely in-house—down to the chips via Dojo supercomputers—Optimus is being deployed first in Tesla’s own factories to automate repetitive tasks. Elon Musk claims the goal is to sell millions per year at price points under $30k, though current prototypes are estimated to cost over $100k each.

   Tesla's advantage lies in control: custom hardware, software, training infrastructure, and manufacturing—all under one roof. But its openness is minimal; Optimus is expected to remain closed-source, offered as a product or service rather than a platform others can build on.

3. Figure AI is perhaps Tesla's most direct challenger. Backed by Microsoft, Nvidia, and Jeff Bezos, Figure has raised over $1.1 billion to build an AI-first general-purpose humanoid robot. Its Figure 01 is being piloted at BMW’s Spartanburg factory, with plans to scale to 100,000 units by 2029.

   

   

   

   

   Source: Figure Masterplan

   1. Brett Adcock, Founder & CEO has 20 years into building technology companies, previously the Founder of Archer ($2.7B IPO) and Vettery ($100M exit). Figure’s advantage is its strong team, cloud-based teleoperation and supervision stack, which allows robots to operate semi-independently with human fallback. This "shared autonomy" lowers deployment risk and enables faster time-to-value. Unlike Tesla, Figure is actively collaborating with Nvidia’s Isaac stack, suggesting a more modular and ecosystem-driven trajectory.

4. While most companies are still in pilot phases, Agility Robotics is already shipping units. Its Digit V4 robot—designed for warehouse tasks like tote handling—is being manufactured at its RoboFab facility in Oregon, capable of producing 10,000 robots per year. Agility sells robots outright (at prices between $250k–$350k) and offers Robot-as-a-Service (RaaS) pricing around $10–12/hour, with the goal of undercutting human warehouse labor. The company is squarely focused on logistics, not general-purpose tasks—offering a more narrow but lucrative TAM in the short term.

Key Take‑aways

1. Price dispersion is enormous (US $6 k → $350 k)—driven less by bill‑of‑materials than by volumes and scope of work. Tesla, Figure, and Unitree are betting on high‑volume consumer economics; Agility and Apptronik remain enterprise‑grade and margin‑heavy.

2. Road‑maps converge on 2027–28 for first large fleets. Every vendor publicly targets mid‑decade production ramps, with 2030 the horizon for “millions” (Tesla, Sanctuary) or six‑figure cumulative shipments (Figure, Nvidia‑ecosystem startups).

3. NVIDIA sits at the center of nearly every plan—supplying simulation (Isaac Lab), foundation models (GR00T), and on‑robot compute (Thor). Even Tesla’s Dojo is, in effect, a competitive alternative datacenter layer rather than a replacement for on‑robot GPUs.

4. Use‑case bifurcation is clear:

   • Industrial/logistics (Agility, Apptronik, Figure, Tesla) where ROI is measured against $30‑per‑hour labour.

   • Services/home (1X, Unitree) where safety, cost, and “approachability” trump payload specs.

5. Business models are diverging between outright hardware sales (Fourier, Unitree), RaaS subscriptions (Agility, Sanctuary), and vertically integrated “labour‑as‑a‑service” platforms (Tesla, Figure).

Elon Musk discusses the Tesla Optimus robot, predicting that humanoid robots will be the "biggest product ever" with "insatiable" demand, as everyone will want a personal robot like C-3PO or R2-D2.

The next three years are less about technology breakthroughs and more about execution at scale. Who can actually ship robots by the thousands, not just demo them on stage? That’s where the winners — and the returns — will separate.

Manufacturing First: Two Paths to Payback

Of all possible applications, manufacturing is where humanoids land first. Why? Because the ROI math is too good to ignore. To see why, let’s walk through two scenarios — one where robots cover a single human’s shift, and one where they work 24/7 across multiple shifts.

McKinsey highlights manufacturing as the sector with the strongest near-term demand for humanoid robotics. That shouldn’t come as a surprise. Factory environments combine high labor intensity with structured, repeatable tasks — exactly the kind of work robots can take on most effectively.

Source: McKinsey

Let’s take a closer look at what clients are really digging into when it comes to manufacturing ROI — the assumptions, the payback math, and the scenarios that make humanoids so compelling on the factory floor. To make it concrete, we’ll look at two scenarios in the manufacturing use case — where these robots are most likely to gain traction first.

• In the first scenario, we assume robots simply replace a single human on a standard shift (about 2,000 hours per year).

• In the second, we assume robots run across multiple shifts, effectively covering the work of several humans by operating closer to 24/7.

These two views let us see just how much utilization drives payback — and why factories, with their round-the-clock production schedules, are poised to benefit the fastest.

Key Assumptions

1. Single Shift and Multi Shift Scenario : Robot = assumed to replace either:

   • 1 FTE (single-shift equivalence), or

   • ~3 FTEs (multi-shift) if run 24/7 (≈6,000 hours/year).

2. Wages: Range $10–$40/hour to reflect loaded cost (The federal minimum wage remains at $7.25/hour plus ~30% benefits)

3. Work Hours: Human = 2,000 hours/year (single-shift worker)

4. Robot Cost: $15K–$30K upfront, plus $2K/year maintenance (ignored in payback since effect is small).

5. Productivity: 1 robot = 1 human equivalent per shift; linear scaling.

6. No downtime assumption (conservative — integration/ramp-up could extend breakeven by 3–6 months).

Now that we’ve laid out the assumptions, let’s review the sensitivity tables. They highlight just how dramatically the payback window changes depending on two factors: the wage level and whether robots are deployed in single-shift or multi-shift operations. At lower wages ($10/hr), the economics look respectable but not explosive — payback takes more than a year if robots simply replace one worker per shift. But once you move into multi-shift utilization, the math flips.

Even at modest wage levels, humanoids pay for themselves in just a few months. And at higher wage environments ($30–$40/hr), the payback window shrinks to mere weeks. This is why analysts argue manufacturing will be the first industry to tip: the combination of round-the-clock demand and rising labor costs makes the ROI too attractive to ignore.

These tables are meant to be directional benchmarks, not precise forecasts.

Payback Sensitivity Tables

Single-Shift Scenario (Robot replaces 1 FTE @ 2,000 hrs/yr)

Multi-Shift Scenario (Robot replaces ~3 FTEs @ 6,000 hrs/yr)

Summary:

• Even at low wages ($10/hr), humanoids break even in ~1.5 years single-shift or <6 months multi-shift. At higher wages, ROI becomes extraordinary (payback <3 months at $40/hr, 24/7 utilization)

• Assembly lines are the natural launchpad for humanoids because the work is structured, repetitive, and labor-heavy

If humanoids achieve the promised $20K–$25K unit cost, labor substitution in U.S. manufacturing becomes inevitable — especially in multi-shift facilities where ROI is undeniable.

The results speak for themselves: months, not years, to breakeven. Manufacturing isn’t just the first use case — it’s the beachhead that proves humanoids are economically inevitable.

Conclusion

USCC’s “Humanoid Robots” report paints a picture of accelerating humanoid robotics development, with China emerging as a highly coordinated and aggressive developer. China plans include establishing a world-class innovation ecosystem by 2025 and integrating humanoids into manufacturing supply chains by 2027. These efforts are backed by comprehensive government strategies: subsidies, tax incentives, development zones, and public–private collaboration.

When you put it all together, the story is clear: humanoids are not a far-off bet — they’re the next major wave of embodied AI. Market forecasts may understate the upside, but the execution roadmaps tell us scale is coming faster than most expect. Manufacturing will prove the economics first, with payback measured in months, not years. And behind every robot stands a data center, fueling demand for GPUs and AI infrastructure well beyond LLMs.