Atomic Banking - Here Come the Jetsons
As enterprises (including banks) push their technology boundaries with new workloads and AI adoption, our demand for energy increases (as do questions about how it is created). At the tip of that spear are the cloud data centers of major hyperscalers.
In the United States, the Federal Energy Regulatory Commission reports that demand for electricity is now projected to grow at 4.7% per year for the next 5 years – that’s a big jump from the 2022 forecast of only 2.6% per year. Much of this new demand is driven by plans for increased industrial capacity, and the increasing amounts of electricity required to power data centers.
AI has grabbed the headlines over the past two years, in part due to concerns about the compute power requirements. This increasing demand for energy comes at a time when energy policy is pushing for cleaner energy sources. This puts the hyperscalers in an interesting spot – meeting the compute resource needs of AI (and other workloads moving to the cloud) and meeting green commitments at a time when, for most countries, the vast majority of electricity is generated from fossil fuels.
The past few months witnessed an interesting news cycle about cloud data centers. More specifically – the investments and partnerships announced by big tech and the major hyperscalers about powering their data centers with a nuclear technology - Small Modular Reactors (SMR).
- Amazon announced three “nuclear” initiatives – one of which is to co-locate a data center adjacent to a nuclear energy facility in Pennsylvania.
- Google announced it would purchase nuclear energy from SMRs operated by Kairos Power.
- Microsoft published a policy brief, Accelerating a Carbon-Free Future in late 2023 – and followed this up with a recent announcement that it would buy 100% of the electricity output from Constellation Energy’s revival of the Three Mile Island nuclear power plant.
- Oracle announced it plans to build a “gigawatt-scale” data center powered by SMRs
Of course, nuclear energy is far from new. Depending on the geographic location, bank data centers may already be drawing on electricity generated by nuclear power stations. This is almost guaranteed in France, where 64% of the nation’s electricity is generated by nuclear energy, but not in Germany, where nuclear power has been phased out in waves since the early 2000s. In the US, approximately 18% of electricity supply is from nuclear reactors. 1
But why SMRs vs other energy sources? There are three key drivers:
- Power delivery requirements. Data centers require high minimum levels of uninterrupted power (baseload power), plus uninterrupted variable energy supply to meet peak demands. This is essential for 24x7x365 uninterrupted operations.
- Footprint – compared to other energy sources (especially renewables at scale), SMRs have a relatively small physical footprint.
- Operating life – SMRs are projected to have operating lives of up to 30 years - far beyond alternative renewable energy sources.
Bringing this back to banking… I wrote recently about the increasing challenges presented by digital and data sovereignty and the somewhat conflicting requirements for resilience and portability of banking technology. Data center and cloud infrastructure decisions are long-term commitments for banks. It may be the 2030s before these SMRs come online, but that isn’t long for data center planning purposes.
In an age where governments talk about national energy security, regulators talk about digital sovereignty and resilience (which will increasingly impact bank data center strategies), the hyperscalers are now talking (albeit indirectly) about data center energy security and sovereignty. Let’s face it – major disruption in electricity supply to hyperscaler data centers could have far-reaching consequences – and not just for banks. BYOP (bring your own power) could become a thing. Green energy and energy security will become increasingly important considerations in financial institutions’ data center strategies.
1https://www.iea.org/countries
Image generated by GenAI