Source: Apple has cut its spending on AWS to around $370M in 2018 from $775M in 2017, in its bid to become self-sufficient in cloud operations
Apple has cut its spending on Amazon Web Services dramatically over the past year, to around $370 million in 2018 from $775 million in 2017 …
Context & Ripple Effects
Apple has been working down its AWS dependence for years: it signed a $400M-$600M deal with Google Cloud back in 2016 explicitly to reduce reliance on Amazon, and this report shows that strategy biting — AWS spend roughly halved in a single year. A parallel CNBC report days earlier put 2018 spend at $350M with a $30M+ monthly run-rate, so Apple remains one of AWS's largest customers even after the cut.
First-order effects
- Apple cuts its AWS bill by roughly $400M year-over-year, redirecting spend toward its own data centers as it pursues self-sufficiency in cloud operations.
- Amazon loses meaningful revenue from one of its flagship large customers, though at $30M+/month Apple still ranks among AWS's biggest accounts.
Second-order effects
- Google emerges as the structural beneficiary of Apple's diversification — a pattern that compounds later, when Apple's Google Cloud storage spend grows ~50% to ~$300M in 2021, dwarfing ByteDance and Spotify.
- Hyperscalers are pushed to compete harder for Apple-class workloads, since any single provider can lose half a nine-figure account in twelve months.
Third-order effects
- The self-sufficiency ceiling becomes visible: Apple shed commodity storage and hosting spend but by 2024 was spending ~$7B per year renting Nvidia chips from AWS and Google, meaning AI-era compute re-locks even the most capable in-house builder into rivals' infrastructure.
- If the pattern holds, hyperscale cloud economics split into a replaceable commodity layer (where big tenants build their own) and a rented frontier-compute layer (where they cannot), reshaping what 'cloud independence' can mean.
The trend: Big tech's drive toward cloud self-sufficiency reliably trims commodity workloads in-house but keeps getting reversed by each new compute wave that only rival hyperscalers can supply.