Technologies
The cloud and who runs it
British regulators have measured cloud concentration in detail; nobody measures aggregate downtime, the cost of outages or how much water data centres use.
The cloud is two companies and a long tail. The UK Competition and Markets Authority found in 2025 that Microsoft and Amazon Web Services each held between 30% and 40% of UK cloud services supply, and identified three adverse effects on competition: concentration, barriers to switching and Microsoft's licensing of its own software on rival clouds. That concentration became visible twice in a month in late 2025, when an AWS failure in one region degraded service for about fifteen hours and a Cloudflare error took down core traffic for about three. Both were documented by the companies themselves, the only source available, because nobody independent measures cloud downtime or its cost. The energy figures are better: data centres used roughly 415 TWh in 2024, about 1.5% of world electricity. For a practical example of software used to turn activity into operational metrics, see remote employee monitoring software.
Two firms supply most of the market
Britain has measured cloud concentration more formally than anywhere else, in a market study and then a full investigation.
Microsoft and AWS each held a 30-40% share of UK cloud services supply in 2024, and the CMA recommended prioritising Strategic Market Status investigations into both.
Direction: Increase. Strength of evidence: Strong.
Caveat Shares are ranges because the underlying figures are confidential, and the finding is UK-specific rather than global.
An earlier Ofcom study of a narrower segment found a higher combined share, and examined the fees charged for moving data out of a provider's network.
AWS and Microsoft had a combined 70-80% share of UK cloud infrastructure services in 2022, and egress fees are "likely to be higher than the incremental costs of providing the service".
Direction: Increase. Strength of evidence: Strong.
Caveat A market study has lower evidential standards than a full investigation, and its segment definition differs from the CMA's, so the two figures are not directly comparable.
The two numbers are not in conflict; they measure different things. Infrastructure services are a narrower category than cloud services as a whole, and that detail decides what a share figure means.
The regulator found switching to be the binding problem
Concentration alone is not a finding of harm. The CMA's final decision named three adverse effects: concentration plus entry barriers allowing above-cost returns, technical and commercial barriers to switching and to running more than one cloud, and Microsoft's software licensing disadvantaging rival clouds (CMA, 2025). The firms dispute this and no remedies were imposed — the matter went to a future Strategic Market Status process. Egress pricing is the concrete mechanism: a customer pays to take its own data out, so the cost of leaving rises with the amount stored.
Two failures in a month showed how much rests on how few
Both incidents are documented only by the companies responsible, which publish good technical post-mortems and no estimate of harm.
An AWS failure beginning 19 October 2025 caused roughly 15 hours of degraded service in us-east-1, traced to a latent race condition in DynamoDB's DNS management that produced an empty DNS record and cascaded into EC2, Lambda and other services.
Direction: Increase. Strength of evidence: Strong.
Caveat The provider's own write-up; it quantifies internal error windows but not customer harm, downtime cost or end users affected.
The second failure came from a different layer and a smaller mistake.
Cloudflare suffered roughly three hours of core-traffic failure on 18 November 2025 after a database permissions change doubled a bot-management feature file past a 200-feature limit, its worst outage since 2019 by its chief executive's account.
Direction: Increase. Strength of evidence: Strong.
Caveat Self-reported, and the post does not quantify the share of global traffic affected, so downstream impact remains unmeasured.
A file exceeding a size limit is not an exotic failure mode. What made it consequential was position rather than severity: the failures are ordinary, the blast radius is not. The same asymmetry runs through what data breaches cost.
Electricity use is measured, and the forecasts are wide
The energy question has two good sources, one global and one American, and both mark clearly where measurement stops.
Data centres consumed about 415 TWh globally in 2024, roughly 1.5% of world electricity, projected to reach about 945 TWh by 2030 while accounting for under 10% of global electricity demand growth over that period in the base case.
Direction: Increase. Strength of evidence: Strong.
Caveat The 2030 figure is a scenario, the IEA publishes four divergent cases, and the 2024 baseline is modelled from incomplete facility-level disclosure.
The last clause is the one usually dropped in coverage: data centres are growing quickly and are still a small share of the growth in electricity demand overall. The US national laboratory estimate shows how uncertain the forward numbers are.
US data centres used 176 TWh in 2023, 4.4% of total US electricity, projected to reach 325-580 TWh or 6.7-12% of US electricity by 2028.
Direction: Increase. Strength of evidence: Strong.
Caveat The 2028 range is nearly a factor of two wide, reflecting genuine uncertainty about AI hardware deployment rather than precision.
A range that wide is a statement about what is not known. It belongs with the rest of the environmental cost of digital technology, where solid baselines and speculative projections sit side by side.
Leaving the cloud is anecdote rather than data
The best-known case for moving workloads back in-house is one company writing about itself.
37signals reported cutting its annual cloud bill from $3.2 million to $1.3 million after leaving AWS, claiming almost $10 million of savings over five years.
Direction: Decrease. Strength of evidence: Weak.
Caveat A self-reported case from a company that markets its own on-premises tooling, with the author conceding the comparison is never fully like for like.
One company's bill is not a trend. It is cited because nothing better exists, which is itself the finding.
The short version
- Microsoft and AWS each held 30-40% of UK cloud services supply in 2024, and the CMA found three adverse effects on competition (CMA, 2025).
- Ofcom found the two held a combined 70-80% of the narrower UK cloud infrastructure segment in 2022, and that egress fees likely exceed the cost of the service.
- An AWS regional failure degraded service for about 15 hours in October 2025, and a Cloudflare error broke core traffic for three hours a month later.
- Data centres used roughly 415 TWh in 2024, about 1.5% of world electricity, but under 10% of projected global demand growth to 2030 in the IEA base case.
- Nobody independent measures cloud downtime, its economic cost or data-centre water use, so the most-quoted figures in all three areas come from interested parties.