Streaming services and news organizations run enormous computing operations while owning very little of the hardware. That choice follows from the shape of their demand rather than from fashion.
Demand arrives in spikes
Media traffic is not steady. A finale, a breaking story or a live event can multiply normal load within minutes and return to baseline hours later.
Owned infrastructure must be sized for the peak, which means paying continuously for capacity used a few times a year. The idle majority is pure cost.
Rented capacity converts that into a variable expense. The company pays for the spike when it happens and stops paying when it ends.
Capital expense and operating expense are treated differently
Buying servers is a capital outlay depreciated over years, committing money before anyone knows whether a service will grow into it.
Cloud spending is an operating expense that scales with usage, which is easier to justify internally and easier to reverse if a product underperforms.
For a business launching services that may not survive their first year, the ability to unwind a commitment quickly is worth paying a premium for.
Delivery is a separate market
Moving video to viewers is handled by content delivery networks, which place copies of files close to the audience and charge for the bandwidth they serve.
Building an equivalent global footprint requires relationships with hundreds of networks, physical presence in many markets and constant renegotiation. Very few media companies have the scale to justify it.
Those that do usually build partially, keeping a private network for predictable baseline traffic while renting from commercial providers for peaks and hard-to-reach regions.
Egress pricing shapes architecture
Cloud providers typically charge little to receive data and considerably more to send it out. For a video business, outbound traffic is the entire product.
This asymmetry makes the transfer bill a first-order concern rather than a line item. Encoding choices, cache strategy and file sizes are decided partly against it.
It also creates lock-in. Moving a large media library between providers means paying to extract it, which is a real barrier to renegotiating a contract from strength.
The rent-versus-own line moves with scale
At low volume, renting is unambiguously cheaper. At very high and predictable volume, the arithmetic reverses and owning hardware becomes defensible again.
Several large streaming and social operations have repatriated parts of their workload for this reason, keeping variable and experimental systems on rented capacity.
The steady state most companies reach is hybrid. Predictable load runs on owned or reserved capacity, and everything unpredictable stays somewhere it can be switched off.