The cloud has established itself in ten years as a dogma, flexibility, speed, economies of scale, but as companies reach their maturity, another obvious fact becomes evident: the economics of the cloud are not those of their models.
The cloud model: consumption versus depreciation
The principle of “pay-as-you-go”, paying for what you consume, appeals to hypergrowth startups. It allows you to experiment with your projects without tying up capital, to adjust capacity to demand and to focus on the product, but this model crumbles when growth slows and the load becomes predictable. Variable costs then turn into disguised fixed costs, without the visibility or stability of material depreciation.
37signalsthe American publisher of Basecamp and Hey, has experienced this. Its cloud bill peaked at $3.7 million per yearbefore the company decides to repatriate its infrastructure. Since then, its costs have fallen to just over a million dollars annually, while improving performance. “We wasted more than ten million dollars between 2018 and 2023,” recognizes its co-founder David Heinemeier Hansson.
A misalignment of interests
The cloud economy relies on continued growth in consumption, so the more companies store, transfer or compute, the more providers earn. Conversely, for companies looking to reduce their marginal cost per userto stabilize their expenses, to capitalize on their assets, the cloud model becomes paradoxical:
the more efficient a company is, the less it benefits from the cloud.
THE egress fees (data output fees)THE duplication and resilience costsnecessary to guarantee availability in several zones. or the long-term contractual commitmentsoften misunderstood (S3, EC2 reserved instances), transform the promise of flexibility into contractual dependence. For profitable and stable companies, the “all rental” logic becomes economically incoherent.
The return of productive capital
Mature companies are thus rediscovering the virtues of amortized hardware, energy control and performance on dedicated servers. However, the return to ownership does not reflect a technological decline, but a rational optimization: better to own what you master than to praise what you endure. This is the logic of Dropbox, Cloudflare or 37signals, players who now see their infrastructures as levers of efficiency, and not as incurred expenditure items.
Towards a new technological realism
If the cloud does not disappear, it becomes a tactical agility tool, useful for experimenting, deploying or absorbing a peak load, and less a single model. In a context where profitability is once again becoming central, companies can review their trade-offs: cloud for speed, ownership for stability. The time has come to reconcile the logic of the product with that of the infrastructurebefore the bill speaks for itself.
