Decarbonising data centres: hyperscalers’ power race will increase emissions but spur more direct decarbonisation

For years, hyperscalers expected data centre growth to be incremental. The unforeseen need for 100-200 GW of new global energy capacity by 2030, however, in addition to almost 100% uptime availability, has necessitated a paradigm shift. 

While the surging growth in AI-driven power demand means global data centre emissions will almost certainly increase, public appetite for decarbonisation means some direct and indirect offsetting measures are likely. The key questions are: which decarbonisation pathways are economically and technically viable, and will hyperscalers actually pursue them? 

In a recent two-part report, Wood Mackenzie analysts examined viable decarbonisation pathways and hyperscaler willingness to adopt them. Fill in the form to receive a complimentary extract from the report and read on for a brief introduction. 

Viable decarbonisation pathways: what works and what doesn't 

Increased scrutiny on the timing and geography of renewable energy credits (RECs) renders them less viable for hyperscalers. Battery storage technology continues to improve but remains incongruent with data centre uptime requirements when combined with renewables - traditional lithium-ion batteries yield insufficient uptime without backup firming from the grid. 

Several decarbonised power sources show more promise: 

Near-term options (2026-2029): 

• Nuclear restarts: Some of the 27 GW (11.5 GW in the US) of prematurely shutdown nuclear capacity could be restarted, offering relatively fast access to firm, decarbonised power. Recent hyperscaler deals with Constellation, Talen, Nextera and Vistra signal strong interest. 
• Solid oxide fuel cells (SOFCs) with carbon capture: Promise fast speed-to-power - less than one year—but may be expensive and are unlikely to scale rapidly. Bloom Energy aims to reach just 2 GW of annual manufacturing capacity this year. 
• Carbon capture on gas combustion: Gas dominates most near-term data centre power capacity additions. Carbon capture can be added either from the start or retrofitted later, offering valuable optionality. 

Medium-term breakthroughs (2030-2035):

Between 2030 and 2035, hyperscalers should have more options. Enhanced geothermal, next-generation carbon capture technologies and long duration energy storage (LDES) batteries could deliver transformational cost reductions. First-of-a-kind (FOAK) small modular reactors (SMRs) and new traditional reactors show promise but face the highest risk of not being cost-competitive and therefore not scaling. 

The recommendation: If hyperscalers want to decarbonise beyond nuclear restarts or extensions, they should prioritise carbon capture, followed by enhanced geothermal and potentially LDES storage. These technologies are most likely to scale and be cost competitive. 

Our analysis shows levelised costs ranging from US$55/MWh for greenfield nuclear (though with 10+ year development timelines) to over US$400/MWh for renewables plus long-life batteries when accounting for uptime requirements. CCUS on gas combustion falls in the middle at around US$135-145/MWh, offering a compelling balance of cost, speed-to-power and scalability. 

Hyperscaler willingness remains the critical variable 

Despite having greater financial means to decarbonise than most industrial energy users, hyperscalers face a fundamental tension: they will not sacrifice market share or profits for decarbonisation while competing for AI dominance. 

Today, most hyperscalers claim near-zero Scope 1 and 2 data centre emissions through RECs and carbon removal offsets, yet US data centres have an average estimated carbon intensity of 548 kg CO₂ per MWh - 48% above the national grid average of 370 kg per MWh. Proposed changes to the GHG Protocol will likely reduce hyperscaler ability to cover their direct emissions with RECs. 

We project data centre power use to double or more to 800 TWh by 2030 from 400 TWh in 2024, with growth continuing to reach over 3,500 TWh by 2050. Before investing in direct decarbonisation, hyperscalers will counteract rising demand through renewable investments, carbon removal offsets and RECs. Medium to long term, the extent to which they invest in decarbonised power sources whose electrons they actually use will depend on regulatory pressure, public perception and evolving climate policy. 

Ultimately, in a world that is undecided whether to decarbonise and who is on the hook for it, hyperscalers are targeting energy diversification and optionality. Achieving credible decarbonisation when needed may require taking action sooner and faster than expected. 

Learn more 

To receive a complimentary extract from our report and read more about Wood Mackenzie’s views on hyperscaler decarbonisation, fill in the form at the top of the page.