Why Robots Don't Deploy At Scale | The Main Gap

In November 2025, Benjamin Bolte sent a farewell email.

K-Scale Labs had raised $6.25M across three rounds in nine months. It had over $2M in pre-orders, including from the head of OpenAI's robotics business. It tried to raise a $10–15M bridge to mass production. The raise failed. By November, $400K was left in the account. The CEO refunded every deposit, laid off the team, and open-sourced the entire stack on the way out.

Three months later, Amazon — which has every conceivable advantage in capital, talent, supply chain, and deployment infrastructure — quietly killed Blue Jay, its ceiling-mounted multi-arm picking system, citing "high manufacturing costs and operational issues." Same year, BMW finished an eleven-month pilot with Figure 02 at Spartanburg. 30,000 X3s assisted, 90,000 components moved, 1,250 operating hours, 1.2 million steps. The robot worked. The pilot worked. And then BMW did what every serious enterprise customer does after a successful pilot: they planned the next pilot. Leipzig. Summer 2026. A different robot, a different use case. Not a fleet purchase. Another pilot.

These three stories look like different kinds of failure. They're not. They're the same problem at three points in the chain.

The technology mostly works. The unit economics mostly close. The companies still die — and the deployments still don't scale — because the capital that should be flowing to productive machines is locked behind the wrong rails.

This essay is about those rails. What's broken, why even S&P 500 buyers can't get around it, and what the survivors of the next two years will have figured out about the unique financial properties of a machine that earns.

‍

The Main Gap is Financing 

Start with the asset class itself. Industrial robot installations hit an all-time high market value in 2026. Roughly 4.6 million robots are operating in factories worldwide. The robot financing market is on track to hit $13.73B by 2030. The category is not an experiment. It is productive infrastructure.

And yet.

Every layer of this market is telling you the same thing: the financing rails were built for a different kind of asset. They underwrite static industrial equipment, owned by creditworthy industrial buyers, depreciating on five-to-seven-year schedules with predictable salvage values and an established maintenance ecosystem. Almost none of those properties hold for a 2026 humanoid, mobile manipulator, autonomous fleet, or AI-driven sensor network.

The gap shows up at four seams.

‍

Even S&P 500 Buyers Are Stretched

Start at the top. JPMorgan's Smothering Heights 2026 outlook documents that AI's share of capex among the top US tech buyers ran 40% in 2023, 55% in 2024, and 70% in 2025. Goldman Sachs Research notes S&P 500 capex is now on track to reach 75% of cash flows — a level last seen during the late-1990s telecom buildout. As Goldman's 2026 investment outlook puts it, AI capex has largely been internally financed to date, but the increasing reliance on debt warrants close monitoring.

Columbia Threadneedle estimates that of the ~$6 trillion in projected AI infrastructure spending through 2030, roughly $1.5 trillion will need to come from external financing — public credit, private credit, off-balance-sheet structures, asset-backed lending. RBC tracks the same shift: Big Tech capex as a share of revenue is at its highest level in over a decade, marking a departure from the asset-light models that supported premium valuations through the 2010s.

Meta's net debt to EBITDA rose from negative at the start of 2025 to 37% after its October bond issue, and 63% when consolidating its Hyperion off-balance-sheet vehicle. Meta is one of the strongest credits on earth and it is restructuring itself around a single capex push.

That is the top of the market. Now look at the bottom. JPMorgan's same outlook flags that retailers and mid-market manufacturers with high debt levels and low tech integration are struggling to keep pace, facing the double-edged sword of high interest rates and falling relative productivity. The companies that most need automation — to close labor gaps, defend margins, compete on output per worker — are the companies whose balance sheets can't absorb the capex line.

If hyperscalers are tapping bond markets and SPV structures to fund AI infrastructure, and mid-market manufacturers can't fund automation through traditional debt at all, the conclusion is uncomfortable but unavoidable: the existing capital stack cannot finance the rollout of physical AI at the speed the technology is arriving. Something has to give.

‍

Banks won't Lend Against an Asset They Can't Price

The traditional industrial robot financing market — FANUC arms, Universal Robots cobots, ABB welders — works fine. Lessors like NFS Capital, ENGS Commercial Finance, and DLL underwrite these as standard equipment with 12–84 month terms, predictable depreciation, and an established secondary market. Section 179 tax treatment is dialed in. 80% of US companies obtain equipment through financing. For the mature category, the rails exist.

They don't exist for what's coming next. The closest analytical analogue is autonomous trucking, where Hansel Leasing notes banks hesitate to issue traditional loans because residual values are uncertain and long-term maintenance costs are not yet proven. The same logic applies tenfold to humanoids. IPO Club's 2026 humanoid research argues the true economic depreciation window is 3 years, not 6, because embedded AI compute, sensors, and power management face rapid obsolescence cycles. SVRC's 2026 leasing guide is direct: for rapidly improving categories, leasing protects you from buying hardware that's obsolete in 18 months — but lessors face exactly the same residual-value uncertainty.

Insurance compounds it. Underwriters can't write standard policies on humanoid fleets without historical actuarial data, which means most 2026–2027 deployments stay confined to controlled environments under vendor-led safety regimes. Hidden costs add 36–42% to the sticker price of a humanoid over three years — integration, software licensing, training, insurance — and most of those costs aren't easily lease-financeable line items in any case.

The result: BMW × Figure, Agility × Toyota Motor Manufacturing Canada, Apptronik × Mercedes — every flagship 2026 deployment is structured as vendor-led RaaS, not buyer-led equipment finance. The vendor carries the residual risk on its balance sheet because no one else will. That works for a Toyota or a BMW. It caps the entire industry at the vendor's balance-sheet capacity. Cardinal Robotics offering enterprise robotics finance at 5% annualized — versus 9%+ traditional cost of debt — exists precisely because the standard rails won't price the asset at all.

‍

SMEs and Non-Industrial Buyers Are Entirely Shut Out

If S&P 500 companies are stretched and humanoid buyers can't get bank financing, the SME picture is worse. The G7 SME AI Adoption Blueprint identifies access to capital, flexible financing, and affordable AI tools as a top-tier adoption barrier. Academic studies of Industry 4.0 adoption among manufacturing SMEs document systemic financing barriers from both banks and government programs. The World Bank's framing of the SME financing gap — reliance on indirect bank channels, high barriers to direct options, inefficient capital allocation as a critical bottleneck — applies to robotics adoption with full force.

The mechanism is straightforward and brutal: a small Polish injection-molding shop that would benefit massively from a $35K cobot, paying for itself in 18 months, can't necessarily clear the underwriting bar at its local bank — even though the cobot would generate verifiable cashflow from day one. The asset is productive. The buyer isn't bankable. The bank doesn't underwrite the asset; it underwrites the buyer.

Then there's the entire category of buyer who isn't a manufacturer at all. A Bundesliga stadium operator who wants to deploy stadium-grade IoT — 8K 360° cameras, edge infrastructure, monetizable B2C and B2B feeds — has clear demand and a clear revenue model. They don't have a capex line for it, and their CFO won't approve one. A hospital that would benefit from autonomous logistics. A retail chain that needs inventory robots. An agricultural cooperative that wants robotic harvesting. None of them fit the manufacturer-equipment finance template, and none of their bankers know how to price the assets.

The combined gap is structural. Capital wants exposure to robotics-as-an-asset-class. Robots are generating verifiable revenue. The rails between the two are running on paperwork and underwriting models built for static industrial equipment held on creditworthy balance sheets. Three of those four conditions are no longer the world we're living in.