Ask an investor why enterprise software is the greatest business model ever conceived, and you will receive a speech about “capital lightness.”
There are no blast furnaces, no warehouses of unsold inventory waiting to rot, and no fleet of diesel vans depreciating in an industrial estate in Slough. You write the code once, upload it to the internet, and sell it an infinite number of times at a marginal cost that rounds cleanly to zero. The gross margins sit comfortably north of 80%, and a generation of investors whose predecessors analysed railways, paper mills, and oil refineries fell head-over-heels in love with an industry that seemed to produce compounding cash flows without ever demanding capital reinvestment.
This is a wonderful bedtime story. It is also an optical illusion manufactured entirely by the quirks of double-entry bookkeeping. Software businesses consume vast rivers of capital at industrial rates; they simply hire accountants who file the machinery under general payroll.
To understand the trick, consider a simple accounting comparison.
Firm A manufactures precision automotive brackets. It spends £50,000,000 designing, machining, and installing an automated hydraulic stamping press.
Firm B builds an enterprise logistics routing engine. It spends £50,000,000 paying one hundred senior engineers to design, build, and test a distributed database system.
Both machines take two years to build. Both cost £50,000,000 in cash. Both require £5,000,000 every year in specialised maintenance to prevent systemic failure. Firm A replaces hydraulic seals, repacks bearings, and recalibrates sensors. Firm B patches zero-day vulnerabilities, updates expiring cryptographic certificates, migrates schemas, and fixes integration regressions caused by upstream platform updates.
Yet standard accounting regimes treat these identical economic acts as if they belong to different universes.
If an asset can be dropped on your foot, accountants recognise it as capital. Firm A puts the £50,000,000 stamping press onto its balance sheet under Property, Plant, and Equipment, depreciating it by £5,000,000 annually.
Firm B benefits from an immaculate conception. Because symbolic logic cannot be dropped on an auditor’s foot, GAAP and IFRS decreed that the £50,000,000 should be flushed straight through the income statement as Research and Development. The software engine vanishes from the balance sheet the instant the cash leaves the bank, sitting alongside recruiter fees and cold brew.
Once operational, Firm B announces an 85% gross margin. Analysts marvel at the sheer capital efficiency. The £5,000,000 in annual engineering labour required to keep the platform from collapsing is filed under general operating expenses.
Firm A has a factory on its balance sheet. Firm B has a factory too; it simply disguises it as payroll.
Every now and then, the institutional machinery catches a glimpse of the reality behind the disguise.
In the Tax Cuts and Jobs Act of 2017, the United States Congress made a quiet change to Section 174 of the Internal Revenue Code. Effective for tax years beginning after 2021, companies were suddenly forbidden from immediately expensing their research and experimental costs, which the statute explicitly defined as including software development. Instead, companies were required to capitalise software payroll and amortise it over five years for domestic engineers, or fifteen years for foreign research.
For three taxable years, the IRS accidentally behaved like a rigorous industrial economist. The tax authorities looked at Silicon Valley and said: Wait a minute. You are building capital equipment. Treat your developers like stamping presses and pay your taxes accordingly.
The result was pure corporate panic. Startups that were break-even on cash suddenly received massive tax bills on fictional taxable income, because the payroll used to construct their core systems could no longer be deducted in the year the cash was spent. A seed-funded startup with $2,000,000 in revenue and $2,000,000 in engineering salaries suddenly found itself taxed as if it had generated $1,600,000 in net profit.
The software lobby descended on Capitol Hill, explaining that treating software engineering as asset construction was an assault on innovation. In July 2025, Congress folded, restoring immediate domestic expensing. The tech industry breathed a sigh of relief, and politicians agreed to restore the polite fiction that a £50,000,000 codebase is just an ephemeral operating cost.
The standard intellectual defence of the capital-light doctrine rests on physical durability. Steel rusts, bearings grind down, and concrete crumbles. Code, being nothing more than strings of ASCII characters frozen in silicon, is theoretically immortal. A piece of code written today will run identically in twenty years.
Anyone who has ever operated production software knows this is a fairy tale.
Physical machines suffer wear because they interact with friction, moisture, and heat. Software suffers wear because it sits inside a computational environment that refuses to stand still.
Lock an industrial stamping press in a dry warehouse for three years. When you return and throw the main breaker, the machine will run. Deploy a production web application to a server, lock the codebase for three years, and walk away. When you return, the TLS certificates have expired, the dependencies have accumulated fifty critical security vulnerabilities, upstream payment APIs have deprecated your payload schema, and the application is dead.
We call the labour required to fight this entropy “maintenance,” which is a word borrowed straight from plant management. A manufacturing plant manager spends maintenance capex to replace worn seals so the assembly line keeps running. A VP of Engineering spends maintenance capex on software developers to refactor dependencies so the database keeps responding. The activities are identical. The difference is that one specialist wears high-visibility gear while the other wears Patagonia fleece.
The second great illusion of modern capital lightness involves the physical infrastructure.
When a B2B SaaS company boasts that it owns no physical servers, it is celebrating an accounting trick: it outsourced its capital expenditure to four publicly traded landlords.
The four cloud hyperscalers (Amazon, Microsoft, Alphabet, and Meta) guided to roughly $700,000,000,000 in aggregate capital expenditure for 2026. This scale of capital deployment sits in a category reserved for the Apollo Programme crossed with the 1840s railway mania. These firms are buying gigawatts of dedicated nuclear capacity, constructing high-voltage substations, laying thousands of miles of private dark fibre, and buying accelerators by the shipping container.
Every SaaS founder claiming to run an “asset-light” enterprise simply receives a fractional invoice for this infrastructure on an American Express card every thirty days.
Consider an airline that leases every aircraft from a financing consortium. It owns zero aeroplanes. Its balance sheet carries no aluminium and no jet engines. The marketing team could claim to be an asset-light transportation network, but the underlying business remains violently capital intensive. The capital expenditure simply moved onto the balance sheet of the lessor, paid for via lease fees disguised as operating expenses.
Cloud computing is an equipment lease on a planetary scale.
The purest comedy of the era belongs to the allocators.
An institutional venture capitalist has precisely one function in the global economy: to allocate capital. For a quarter of a century, the highest-status, highest-multiple asset class in finance was one where allocators looked each other in the eye and solemnly agreed that the genius of the model was its freedom from capital requirements.
Then they wired £100,000,000 into a company’s bank account for a Series C round.
The destination of that capital was never mysterious. The entire quantum went directly into payroll accounts, funding two hundred engineers to design, construct, verify, and patch a massive intangible machine. Investors were pouring vast rivers of capital into building software factories, while reading audited financial statements that allowed them to pretend the factory did not exist.
The delusion was irresistible because the arithmetic was so flattering. An 80% gross margin reveals nothing about freedom from capital; it simply indicates that expenditures were incurred before the first customer arrived. A hydroelectric dam operates at near-zero marginal cost per kilowatt-hour, and a suspension bridge incurs near-zero marginal cost per vehicle. Both generate gross margins that would flatter an enterprise software vendor, precisely because massive capital was sunk upfront.
Convincing yourself that the enterprise is “capital light” transforms the standard cost structure of infrastructure into an apparent perpetual motion machine. Everyone agreed to participate in the charade: founders accepted the capital, engineers built the factory, accountants filed the construction under general operating expenses, and investors returned to their limited partners to boast of backing businesses that required no capital whatsoever.
Software companies construct complex, fragile machines out of symbolic logic, run them on trillions of dollars of rented physical plant, and spend millions every year fighting the relentless decay of a shifting computational environment.
Software was never capital light. Investors simply fell in love with a business that let them fund a factory while pretending they had discovered a perpetual motion machine.