Why you're addicted to cloud computing

Cloud computing is profitable not because it’s inherently cheap, but because providers engineer “lock-in” so customers keep paying—often long after switching would look attractive on paper. The core business dynamic is simple: Amazon’s retail business runs on thin margins, while Amazon Web Services (AWS) becomes a cash cow with operating margins around 30%, and the cloud model depends on keeping startups and enterprises dependent on the same infrastructure.

That dependence starts early. New customers get hooked through free tiers and credits—enough to let a young company build quickly without immediate cost pain. The transcript frames this as a deliberate long game: if a startup survives, the provider can earn far more than the initial free credits. It also highlights how cloud pricing is intentionally complex, making it difficult to estimate the true cost of migrating away. For enterprises, long-term discounts and bundled support further increase the switching burden, sometimes reaching billions in value when large customers are involved.

Once a workload is running, the biggest barrier to leaving is often egress fees—charges for data leaving a cloud network. Ingress (data coming in) is generally free, but moving large datasets out can create a “massive bill” that wipes out the savings from switching providers. The transcript cites an example: 37signals reportedly evaluated an S3 migration cost of roughly $300,000–$400,000 just to move data. It also notes a recent shift: Google Cloud removed transfer fees for migrations away from Google Cloud, which lowers the immediate cost of exit and may reflect regulatory pressure such as the EU’s Data Act.

Even when egress fees fall, lock-in persists through ecosystem design. Some platforms resell cloud capacity—like boutique providers that sit on top of AWS—so customers still inherit the same underlying constraints. Database and big-data software can also force staying put: if a managed database service charges dramatically more when deployed across clouds, large datasets effectively “choose” the cloud for the customer. Proprietary technology deepens the trap as well. Managed services (for example, running open-source databases through fully managed offerings) rely on cloud-specific scaling, security, and surrounding tooling, and custom APIs often don’t port cleanly.

The transcript lays out three responses. One is to avoid cloud entirely by running your own servers—more expensive upfront, labor-intensive, and typically less fault-tolerant, though it can pay off for some mid-sized businesses. A second approach is to use portable open-source technologies and adopt multicloud or hybrid architectures so workloads can move. The third is to accept the “addiction”: pick a cloud, use its proprietary conveniences, and treat the tradeoff as paying for reduced operational pain rather than lower cost. The takeaway is that cloud lock-in isn’t accidental—it’s built into pricing, data movement economics, and tightly integrated services, and that’s why the addiction model works.