For many years, hyperscalers assumed that data centre growth would increase gradually. But the sudden need for an additional 100–200 GW of global energy capacity by 2030, along with the demand for almost perfect uptime, has completely changed that expectation. The rapid rise of AI-driven workloads has pushed power requirements far beyond what the industry had predicted, forcing hyperscalers to rethink how they secure reliable and sustainable energy.
As this growth continues, emissions from data centres are expected to rise. At the same time, public pressure for stronger climate action is increasing, pushing companies to consider new decarbonisation strategies. This raises two important questions: which low-carbon solutions are practical and affordable, and which ones will hyperscalers actually choose to implement?
Wood Mackenzie recently released a two-part analysis that explores these issues in depth. The report evaluates different decarbonisation pathways and assesses how willing hyperscalers are to adopt them.Decarbonisation options vary widely in effectiveness. Traditional renewable energy credits are becoming less dependable due to stricter scrutiny on timing and location.
Battery storage continues to advance, but current lithium-ion technologies cannot provide the uninterrupted power supply required by data centres without heavy dependence on the grid.Despite these limitations, several cleaner power sources show strong potential.Near-term options between 2026 and 2029 include nuclear restarts.
Around 27 GW of nuclear capacity worldwide was shut down early, including 11.5 GW in the US. Restarting some of these facilities could provide hyperscalers with fast access to firm, carbon-free energy. Recent agreements with companies such as Constellation, Talen Energy, NextEra Energy and Vistra suggest growing interest in this direction.
Solid oxide fuel cells paired with carbon capture are another possibility. They offer quick deployment, often in under a year, which is appealing for companies expanding at high speed. However, their costs remain high, and global manufacturing capacity is still limited. For example, Bloom Energy plans to reach about 2 GW of annual capacity, which is far below overall industry needs.
Carbon capture applied to gas-fired power generation also stands out as a flexible near-term solution. Since gas will continue to be a major source of new data centre power, carbon capture systems can either be built into new plants or added later, making it easier to reduce emissions without affecting reliability.In the medium term, between 2030 and 2035, new technologies are expected to create more options.
Enhanced geothermal systems, next-generation carbon capture and long-duration energy storage could lead to major cost reductions and broader deployment. Small modular reactors and new types of traditional nuclear reactors also show long-term promise, but they carry high cost and development risks, making their future uncertain.
Given the wide range of options, some technologies stand out more than others. If hyperscalers want to decarbonise beyond nuclear restarts, they should focus on carbon capture on gas plants, enhanced geothermal and long-duration storage. These options are the most likely to scale effectively while maintaining cost competitiveness.
Current cost estimates vary significantly. Greenfield nuclear projects can reach around US$55 per MWh, but they require more than a decade to develop. Gas with carbon capture falls around US$135 to US$145 per MWh, making it one of the strongest near-term choices due to its balance of cost, reliability and deployment speed.
In contrast, pairing renewables with long-life batteries to meet strict uptime requirements can exceed US$400 per MWh, making it far less practical.A major factor in the decarbonisation challenge is the willingness of hyperscalers to invest. While these companies have significant financial resources, they also face intense competition in the AI market. This limits their ability to prioritise deep decarbonisation over expansion or profitability.
As a result, many currently rely on renewable credits and carbon removal offsets to claim low operational emissions, even though actual emissions remain higher. For example, US data centres have an estimated carbon intensity of about 548 kg of CO₂ per MWh, compared to the national grid average of 370 kg per MWh. Proposed updates to the GHG Protocol may also reduce the ability to rely on RECs to cover direct emissions.
Energy use in the sector is expected to more than double from 400 TWh in 2024 to 800 TWh by 2030, and could exceed 3,500 TWh by 2050. In the short term, hyperscalers will continue to use renewable investments, RECs and carbon removal to offset rising demand. In the longer term, their adoption of deeper decarbonisation technologies will depend heavily on regulatory pressure, public expectations and evolving climate policies. Overall, hyperscalers are focusing on building diverse and flexible energy portfolios. But achieving credible decarbonisation may require earlier and faster action than many companies currently anticipate.
Subscribe to get the latest posts sent to your email.
