Why AI Expansion Is Forcing a Return to Fossil Fuels

The exponential rise of generative artificial intelligence has triggered a profound operational crisis for the global technology sector: a direct collision between insatiable computational demand and hard-fought corporate net-zero commitments. As the race for AI dominance necessitates the rapid expansion of hyperscale data centers, technology giants are confronting severely constrained electrical grids and looming regulatory shifts, such as the Greenhouse Gas Protocol's proposed "hourly matching" mandates for clean energy.

Advertisement

Facing government estimates that data center electricity consumption could nearly triple by 2028, industry leaders are increasingly reverting to natural gas and coal to ensure uninterrupted power, threatening to lock in decades of new carbon emissions. This unfolding energy paradox forces a critical reckoning for investors and stakeholders, highlighting the urgent need for advanced grid flexibility technologies, workload-aware efficiency metrics, and massive investments in carbon removal to salvage the sector's precarious climate targets.

The AI Power Paradox: Rising Demand vs. Net Zero Goals

As artificial intelligence rapidly scales, the technology sector faces a widening gap between its ambitious climate commitments and its escalating energy requirements. This dynamic is fundamentally reshaping how the industry approaches its path to net zero.

Surging emissions across industry leaders

Despite early optimism regarding clean energy transitions, major technology companies are acknowledging significant deviations from their initial 2030 climate targets. Milestones that were once framed as highly achievable—such as powering all operations exclusively with clean energy or reaching carbon-negative status—are now being recalibrated, with firms like Google and Microsoft publicly describing these efforts as a "moonshot" or a marathon rather than a sprint.

Tracking the 50%+ jump at Google and Meta

The accelerated deployment of resource-intensive AI models has led to substantial increases in greenhouse gas emissions across the sector. Over the initial five-year period of their climate commitments, Meta and Google saw their emissions surge by over 60% and nearly 50%, respectively. Similarly, Amazon reported a 33% increase, while Microsoft's emissions rose by more than 23%, reflecting a broader industry trend where total operational emissions are climbing despite these companies purchasing record amounts of clean energy.

The 2028 Electricity Projection: AI's strain on the US grid

The infrastructure required to support this computational growth is placing an unprecedented burden on existing electrical grids. In 2024, data centers accounted for approximately 4.6% of all electricity consumed in the United States. According to government estimates, this consumption share is projected to nearly triple by the year 2028, with some analysts forecasting that nationwide electricity use could rise by as much as 20% over the next decade largely due to data center expansion.

GHG Protocol updates and the transition to "Hourly Matching"

Compounding the challenge of rising physical energy demand are proposed regulatory updates to the Greenhouse Gas (GHG) Protocol, particularly concerning Scope 2 carbon accounting. The tech industry has historically relied on renewable energy certificates and power purchase agreements to meet clean energy goals. However, proposed changes would introduce strict requirements ensuring that zero-carbon power sources are located in the same region as a company’s data center and match its exact hours of operation. Under this "hourly matching" framework, solar credits could only be applied against energy consumed during daylight hours, severely complicating compliance for facilities that run continuously.

Fossil Fuel Dependencies and Grid Constraints

The immediate need to secure massive, uninterrupted power supplies for AI data centers is colliding with severely constrained electrical grids. Consequently, technology giants are increasingly turning to traditional fossil fuels to bridge the energy gap, threatening to lock in long-term carbon emissions despite their sustainability pledges.

The resurgence of Natural Gas for AI infrastructure

Driven by a race to maintain competitive advantage, the tech industry is overseeing a rapid expansion of fossil fuel power generation. In 2024, natural gas supplied over 40% of the electricity used by U.S. data centers, while coal accounted for 30% of their power consumption globally. Utilities are now constructing new natural gas plants specifically to meet this concentrated demand, while some tech firms are directly planning on-site gas generation. This infrastructure shift is directly contributing to broader climate impacts, with independent research linking AI expansion to a 2.4% increase in U.S. fossil fuel emissions last year. Because natural gas plants require roughly 30 years to recover their capital investments, this trend risks delaying the broader transition to clean energy.

Regional deployments: Microsoft, Meta, and Google's fossil fuel reliance

Leading technology companies are actively backing specific fossil fuel projects to ensure operational reliability for new hyperscale facilities. In Wisconsin, two new natural gas plants are being constructed to support a Microsoft data center, while three natural gas facilities will power a sprawling Meta site in rural Louisiana. Both companies plan to offset these new emissions through solar investments located in different areas. Similarly, Google is slated to purchase power from a planned natural gas plant at an Illinois corn processing facility, which intends to utilize underground carbon capture and storage technology to mitigate its footprint.

Navigating infrastructure bottlenecks and policy shifts

The deployment of clean energy alternatives is severely hampered by structural and political hurdles. A massive backlog of proposed renewable projects is currently stalled awaiting permission to connect to power grids, creating a near-term infrastructure crunch. This bottleneck is fundamentally a cross-functional operational crisis, where physical grid constraints directly elevate energy costs and complicate corporate compliance and supply chain reporting.

Furthermore, federal policy shifts have exacerbated these infrastructure challenges. Recent administrative actions have disrupted the renewable energy pipeline by canceling grants, permits, and tax breaks for wind and solar development, while simultaneously mandating the continued operation of coal-fired power plants slated for retirement. These changing regulatory and political environments remove the financial and operational predictability companies relied upon when establishing their original net-zero timelines.

Technological Solutions for Energy Optimisation

To mitigate the severe environmental impact of AI's energy consumption, the technology and energy sectors are actively investing in new operational frameworks and grid management tools. From dynamic load balancing to advanced efficiency metrics and carbon markets, these solutions aim to optimize existing infrastructure while securing long-term carbon offsets.

Grid flexibility: strategic acquisitions like Octopus and Uplight

The rapid buildout of data centers necessitates alternative approaches to keeping electrical grids reliable and affordable without waiting for prolonged traditional infrastructure expansions. To help North American utilities manage this surging demand, UK-based Octopus Energy Group acquired a majority stake in Uplight, a Colorado-based grid flexibility provider. Operating in partnership with Schneider Electric, Uplight manages over 8.5 gigawatts of flexible load and serves more than 85 utilities. The acquisition leverages customer engagement and distributed energy resource management solutions to unlock flexible grid capacity significantly faster than physical buildouts allow.

Workload-Aware Metrics: PUE, CUE, and the compute efficiency ratio (CER)

Measuring data center efficiency is evolving well beyond traditional, facility-level benchmarks. While Power Usage Effectiveness (PUE) remains the industry standard—tracking the ratio of total facility energy to IT equipment energy—it can obscure deep inefficiencies, such as when a facility boasts a low PUE but runs servers at very low utilization rates. To address this, operators are increasingly adopting workload-aware metrics like the Compute Efficiency Ratio (CER). CER normalizes actual computational work performed against the energy consumed, making it vital for AI and High-Performance Computing (HPC) workloads where standard PUE cannot accurately track GPU utilization or job throughput. Alongside Carbon Usage Effectiveness (CUE), which measures CO2 emissions relative to IT energy, these metrics provide operators with actionable data to optimize AI deployments.

Biochar and Carbon Dioxide Removals (CDR): Microsoft's 1-Million-Ton Deal

With the power demand from data centers extending the operational life of fossil fuel-fired generation, large-scale carbon removals are increasingly viewed as essential infrastructure to counterbalance unavoidable emissions. Highlighting this strategic pivot, Microsoft recently secured a major offtake agreement with a U.S.-headquartered biochar company to procure one million carbon dioxide removal (CDR) units over a 10-year period. These types of technology-based greenhouse gas removals are becoming critical tools for tech giants attempting to fulfill their long-term net-zero pledges while continuing to scale carbon-intensive AI models.