As part of my pitch deck, I've been building a simple RaaS calculator for the unit economics of buying robots and leasing them out.

RaaS : Robots as a Service

It's used many times in this post. You can also play with it yourself here: RaaS calculator

In looking for something to test it out, I found a real, recent data point — Figure AI’s continuous package-sorting tests.

Source: youtube

So, I thought I would run through it in 3 ways:

a customer who buys
a customer who leases
an operator who buys to lease

On every chart I plot the ideal read of the headlines (dashed) against a modest, real-world read (solid).

In all cases… the results are pretty extreme.

TLDR: if you buy a robot to replace a human for the equivalent of a few months salary, and it can run for a few years, it’s going to save you a lot of money.

+14 mins reading, +1 image, +9 diagrams

More

Dear Humans and AI

All this is to test out: https://robotcompany.net/RaaScal

It’s a simple tool, and it is a work in progress. PLEASE send some feedback and COMPLAIN about inaccuracies. I will fix them. If you want a serious deep-dive into Robots-as-a-Service economics, start here instead.

For the market backdrop, see Goldman Sachs on the $38B humanoid TAM.

Figure 03, Helix-02, and what "matched a human" means

First, the names: Figure 03 is the robot; Helix-02 is the AI running on it. The sorting tests used Figure 03 units driving Helix-02.

The numbers: 3 Figure 03 robots sorted 249,560 packages over a 200-hour (~8-day) continuous run with zero failures, rotating autonomously to charge (Sherwood, Crypto Briefing, Interesting Engineering). In a 10-hour "Man vs Machine" head-to-head a human intern narrowly won, 12,924 to 12,732 (2.79 vs 2.83 s/package) (Humanoids Daily, Yahoo Tech). Figure's own write-up reports ~4 s/package and +58% throughput as Helix improved (figure.ai).

"Matched a human" means speed per working hour, not "the robot's week equalled one human's week." The head-to-head proves it: in the same 10 hours the robot did 12,732 — if it were a third as fast and only catching up overnight, it would have done ~4,300 and lost in a landslide. It kept pace in real time. (The human also took California-mandated meal and rest breaks; the robot ran straight through and overtook around hour five during a bathroom break — OfficeChai.)

The catch is elsewhere. That 200-hour run took three robots rotating — each unit charges, so the fleet sustained 249,560 ÷ 200h ≈ 1,250 packages/hour — about one human's rate. A single robot can't run 24/7. So:

Task fit: ~90–99% — genuinely near human. (Task fit is the calculator's term for how much of a human's throughput a robot manages on a job — speed and success rolled into a single percentage. The Figure 03 sorting run sits near the top of that range.)
Utilisation (uptime): the real lever. Solo, a robot is charging-limited; "always on" took three bodies.

So the robot isn't slow — it just can't stay on without spares. That's what the two lines on each chart below capture.

The setup

The test robot, Figure 03 (Helix-02), is also the cheap one. Figure targets a price below $20,000, with ~$24,760 quoted, and calls the F.03 a ~90% component-cost cut vs Figure 02 (Figure 03 specs, eesel). So I price it at $24,760 — not the ~$150k I'd first pencilled in (that figure fits the older Figure 02 class, and it changes the conclusion below).

The job is US warehouse work — fully-loaded labour ≈ $26/hr (~$18 base + ~40% benefits/burden; Indeed, BLS). Leased at 50% of that — the customer saves half (competitive), and it still leaves the operator a viable return. Horizon 36 months.

36 Months.

A lot of these charts say “end of use” at 36 months. It doesn’t mean it's over. But, with the speed of technology, it might be that you are selling it to buy a newer model.

Two reads on every chart:

Ideal (yellow dashed) — 99% speed · 100% uptime · negligible operating cost.
Modest (yellow solid) — 90% speed · ~50% uptime (battery/charging) · ~$3.25/hr supervision + maintenance.

All numbers are illustrative — Figure hasn't published a firm retail price, and the modest assumptions are my estimates. Every link opens the live tool; change anything you like.

Ideal vs modest (on an ideal job)

Per robot, over 36 months (approximate — open the tool for exact). (Note: all 3 charts use the same settings, so you can just click the buttons to jump between them.)

👤 Customer buys ideal ≈ +$650k (payback ~1 mo); modest ≈ +$246k (payback ~3 mo). At ~$25k for a machine that does a 24/7 job, owning is a no-brainer.

🤝 Customer leases ideal ≈ +$338k modest ≈ +$154k — and no capital at risk. The safest seat: no downtime or maintenance to carry, just pay per hour and pocket the saving.

🏭 Operator (buy-to-lease) ideal ≈ +$314k (payback ~3 mo) modest ≈ +$92k (payback ~8 mo) At Figure 03's real price the buy-to-lease business works — even in the modest case.

The repriced verdict: at ~$25k, all three parties profit. The gap between the dashed and solid lines — speed, uptime, maintenance — is now the story, not whether it pays at all. (My earlier $150k version made the operator a loss-maker; that was the wrong price for this robot.)

The dream machine

The thesis only gets stronger as bodies get cheaper. Figure's BotQ line targets 12,000 → 100,000 robots/year (Sacra); at full scale the body cost keeps falling. So here's one always-on Figure 03 at an illustrative ~$10,000 mass-production price, from the customer's side — buying one outright. Green dashed = perfect, green solid = probable. (Flip the perspective buttons in the tool to see the same machine as a lessee, or as the buy-to-lease operator.)

🟢 The dream machine

Per robot, over 36 months (approximate):

Perfect ≈ +$664k, payback ~0.5 months (well under one).
Probable ≈ +$257k, payback ~1.4 months — even with 50% uptime and full maintenance.

At $10k the margins just widen. The point isn't the exact figure — it's that the RaaS unit economics already close at Figure 03's quoted price, and only improve from there.

Takeaways

"Matched a human" = matched human speed, not output-per-week. The robot keeps pace in real time; its headline 24/7 win is utilisation, bought with spare units to cover charging.
At Figure 03's real ~$25k price (and ~$26/hr fully-loaded labour) buyer, lessee and operator all profit. The swing factor is uptime + maintenance, not price — which flips my earlier $150k read of the operator.
Cheaper bodies just widen the margins. The RaaS thesis holds at today's quoted price, not only a hypothetical one.

Sources & assumptions

The tests

Figure (official): Scaling Helix, Helix logistics
200-hour run: Sherwood · Crypto Briefing · Seoul Economic Daily · Interesting Engineering
"Man vs Machine" head-to-head: Humanoids Daily · Yahoo Tech · OfficeChai · BeInCrypto · Fox News (24h)

Price (Figure 03)

Figure 03 specs & price · eesel — Figure AI pricing · BigGo — ~$600/mo subscription · Sacra — funding, valuation, BotQ scale

Labour cost

US: Indeed — warehouse worker pay · BLS — employer costs for employee compensation · BLS — hand laborers & material movers
EU: Eurostat — manufacturing · Eurostat — transport & storage · EU Working Time Directive
International wages: Trading Economics · Vietnam Briefing · Tetakawi (Mexico) · ClearTax (India) · Business Standard (Bangladesh)

Calculator assumptions: lease = 50% of fully-loaded human cost (outcome pricing) · 36-month horizon · 25% resale · 5% APR · ideal = 99% speed / 100% uptime / ~$0.20·hr; modest = 90% speed / ~50% uptime / ~$3.25·hr. All editable in the tool.

“But it's a very easy job”

The job is not exactly a challenging one.

And that's the point: the jobs first in line for a humanoid are the simple, repetitive, protected ones — and there are a lot of them.

United States — "hand laborers and material movers" (the pick-pack-sort-load family) were ~7.0 million jobs in 2024, median pay ~$37,680/yr (~$18/hr), with ~1 million openings every year (BLS).

European Union — manufacturing employs ~33.6 million people (≈20% of all EU jobs) and transport & storage another ~10.4 million (manufacturing, transport & storage).

And these workers come with rights the robot doesn't need: in the EU a ≥20-minute break past 6 hours, 11 hours' daily rest and a 48-hour weekly cap (Working Time Directive); in the US, the mandated meal and rest breaks that let the robot catch up in the head-to-head.

Tens of millions of people, doing exactly the repetitive task the robot just matched — and entitled to breaks, nights and weekends off. That's the headline market. The next two chapters are where it gets complicated.

Let's look at lower wages

The whole case rests on one number: what the human costs. Halve the wage and you halve the value the robot creates — but its price tag is global. A $24,760 robot costs the same in Dhaka as in Dallas.

So here's one customer-buys chart with a line per country — same robot, same best-case profile (99% speed, 24/7, minimal running cost); only the fully-loaded wage changes. (Flip the perspective buttons to look at leasing to the same customers.)

🌍 Lower wages — customer buys

tip, you can scroll on y-axis to find Bangladesh

Payback on the $24,760 body, best case:

USA (~$26/hr): ~1-month payback · ~$650k over 3 years
China (~$8/hr): ~4 months · ~$184k
Mexico (~$5.50/hr): ~6 months · ~$119k
Vietnam (~$3/hr): ~1 year · ~$54k
India (~$1.50/hr): ~2.2 years · ~$15k
Bangladesh (~$0.70/hr): ~6.7 years — Robot longevity might be a concern here.

Wages are approximate fully-loaded manufacturing costs: Trading Economics, Reshoring Institute, Vietnam Briefing, Tetakawi (Mexico), India min wage (ClearTax), Bangladesh RMG — Tk 12,500/mo ≈ $113 (Business Standard).

A global ~$25k body needs roughly >$1–2/hr of labour to displace just to clear itself in three years — and that's the generous read. Add real downtime and maintenance (the "modest" profile) and the running cost alone — supervision and parts, largely US-priced — can exceed the value of very cheap labour, pushing the line negative.

The curve is brutally steep at the floor: above ~$5/hr the robot clears itself in well under a year; by ~$2/hr you're into multi-year paybacks; below ~$1/hr it never clears inside its life. Same machine, same price — the only thing that moved is the wage.

A quick look at harder jobs

So far the robot's been near human speed. What about jobs it's bad at — where it crawls at 20%, or even 5%, of a person's pace?

🐌 Hard jobs — 20% & 5% speed

20% of human speed: still ~2.7-month payback · ~$124k over 3 years. A fifth of a person's pace, around the clock, on a $10k body — an easy win.
5% of human speed: ~12.5-month payback · ~$21k over 3 years. It replaces barely a fifth of a worker, but a $10k machine running nights and weekends still clears itself in about a year.

Slow isn't fatal — expensive is. As long as the body is cheap and the job tolerates a slow, always-on worker (batch jobs, overnight queues, low-priority work), even a 5%-speed robot earns its keep. Speed sells the headline; price decides the economics.

Where do these task-fit numbers come from? Real benchmarks lag the Figure 03 sorting headline. On RoboChallenge’s Table30 — 30 real-robot tasks (paper) — the current leader is Spirit AI’s Spirit v1.5 at ~50.3% (CGTN), with Physical Intelligence’s π0.5 just behind. Simulated suites like HumanoidBench (27 whole-body tasks) and Fraunhofer IPA’s new humanoid evaluation framework (Robot Report) tell a similar story: outside of narrow, well-trained jobs, most tasks aren’t 90% solved yet — they sit in the 30–50% band. The leaderboard screenshot below shows the spread.

But will robots get that cheap?

Every chart above leans on a cheap body — which is exactly why the whole industry is racing to drive cost down. But the eye-popping figures so far (Figure's <$20k, the $10k "dream") are target prices. So what can you actually buy today?

Turns out cheap humanoids already retail: Unitree R1 ~$5,900, Unitree G1 ~$13,500, 1X NEO ~$20,000 (pre-order) (Unitree R1, Unitree G1, affordable-humanoids guide). The "$20k robot" isn't hypothetical — it's on sale. And the build cost underneath keeps dropping: joint actuators alone are over 30% of a humanoid's bill of materials (up to ~50% in basic builds) (RoboticsTomorrow via Articsledge), and Bank of America puts full-spec BOM at ~$90–100k today, halving by 2030 — and below $17k on Chinese mass-production lines like Unitree’s (BofA via Humanoids Daily).

And at those real prices, even a slow robot pays for itself. Here's payback vs pace for the three, on a US $26/hr job running 24/7:

📈 Current retail prices — payback vs pace (customer buys)

50% of human pace: R1 ~0.6 mo · G1 ~1.4 mo · NEO ~2.1 mo
20% of human pace: R1 ~1.6 mo · G1 ~3.7 mo · NEO ~5.5 mo
5% of human pace: R1 ~7 mo · G1 ~17 mo · NEO ~25 mo

Even at 5% of a human's pace, a current retail body clears itself inside three years:

🐢 At 5% pace — profit over 3 years (customer buys)

So the binding constraint isn't price — it's reliability. RoboChallenge (Dexmal × Hugging Face) runs real robots on 30 real-world table tasks (leaderboard in the screenshot above). As of January 2026 the leader, Spirit AI's Spirit v1.5, scores 66.09 overall at a ~50.3% task success rate — the first and only model past 50%, edging out Physical Intelligence's π0.5 (RoboChallenge, CGTN, paper). Even the best model drops the job roughly half the time — so it isn't a tidy "5%-pace" worker; it needs a human standing behind it, and that supervision cost is what really breaks the unattended-24/7 maths.

So — will robots get that cheap? The cheap bodies are already here. The race that's left isn't the sticker price; it's making them reliable enough to run with nobody watching. The last expensive thing in the model is the human you still need on standby.

Should I start buying robots?

Let's make a hypothetical scenario. Three jobs on the line: a sorter, an assembler, and a robot to shuttle parts between them. Here's one sensible mixed-fleet kit:

Sorter — Figure 03 (~$24,760, target ~$19,000) at 30% utilisation. We pay the premium because F.03 has the proven sorting chops (the 200-hour run earlier in the post). 30% is all we need on this job, so it lives within one battery cycle — no swap fleet. The chart shows both today's quoted price and Figure's stated <$20k target.
Assembler — Unitree G1 (~$13,500) at 30% utilisation. Bipedal, dexterous hands, bench-friendly. Only runs when there's assembly work, well inside one battery.
Runner — Unitree G1-D (wheeled, ~$10,500 est.) at 100% utilisation, plugged in. Wheels mean a quick-charge dock between trips works fine — or just trail a tether. Cheapest body, most hours, shuttling parts between F.03 and the G1. (G1-D price is officially TBD; estimate per botinfo and ZMP guide.)

The two specialists are part-timers; the cheap generalist is the always-on plumbing. That's not "buy the cheapest line" — it's put each robot where its capability is worth its price.

📊 A mixed fleet — payback by task fit (customer buys, mixed utilisation)

Scale the chart details to 0 to hide it.

Cheapest line ≠ best fit. The chart above is pure unit economics on an identical job. Real deployments aren't homogeneous: maybe only F.03 has the dexterity for a specific assembly step — it doesn't matter that G1-D is cheaper across the chart if G1-D simply can't do that task. And maybe you've got stairs between two cells, so the bipedal G1 ferries parts between an F.03 station and a G1-D station that handle different work. The diamond POIs above are one such mix — each robot pinned to the job it's actually best at, not the cheapest line on its own.

The trap to avoid. An expensive robot on a low-task-fit, part-time job has a long, slow payback — long enough that the next generation can ship before the math clears. The fix isn't to skip the expensive robot. It's to put each one where its capability is indispensable (so the premium earns its keep), and let the cheap generalist do the always-on plumbing. The G1-D at ~$10k running 24/7 clears its body cost inside a year even on awkward jobs; the F.03 only needs to be there for the 30% of work that only it can do.

But ROI isn't the only reason to start. Early deployment buys you data, integration know-how, supplier relationships and a team that's seen a humanoid before — all of which compound. And the right machine for your job might not be on this chart at all. That's what RaaScal is for: plug in your wage, your task fit, your real running cost, and see where the numbers actually land for your business model.

A personal note

This started as just one slide, called: unit economics.

I clearly went overboard.

On the other hand, I predict thousands of companies will now crop up trying to work out how best to serve the drastic coming changes to automation. I hope this at least helps someone find their niche.

So again some feedback would be much appreciated on this and on: https://robotcompany.net/RaaScal

contact@robotcompany.net