Power, Not Chips, Now Determines AI Leadership as Grid Bottlenecks Delay 50% of US Data Centers

Google’s 25 MW solar farm partnership with AirTrunk and European Energy Australia, which reached grid completion in May 2026, signals the energy arms race redefining AI infrastructure investment—a pivot from chip procurement to power asset ownership as electricity becomes the binding constraint on machine learning expansion.

The deal follows according to TotalEnergies‘ announcement in February of a 1 GW solar power purchase agreement with Google in Texas—the largest renewable commitment in TotalEnergies’ US history, delivering 28 TWh over 15 years. Combined with a 1.2 GW deal with Clearway Energy and December’s $4.75 billion acquisition of clean energy infrastructure provider Intersect, Google has secured over 3.2 GW of contracted renewable capacity since January 2025. This isn’t diversification. It’s vertical integration.

The Mulwala project exemplifies a model Google is replicating globally: co-locating generation with consumption to bypass grid interconnection queues that now exceed five years in major US markets. “We’re looking at siting next to power generation and creating these industrial parks—still connected to the grid at the substation or interconnection point, but removing the bottlenecks to bringing generation online and also loads online,” Amanda Peterson Corio, Google’s global head of data center energy, told Canary Media.

The Transformer Bottleneck

Behind the strategic shift lies a structural crisis: physical infrastructure cannot keep pace with AI capital deployment. High-power transformer lead times have stretched from 24-30 months pre-2020 to 3-5 years today, while switchgear is sold out through 2028, according to Sightline Climate. The report found that 30-50% of AI Data Centers scheduled for 2026 will be delayed or cancelled, with 11 GW of announced capacity showing no signs of construction.

US imports of medium-voltage switchgear from China jumped from 1,500 units in 2022 to over 8,000 in the first ten months of 2025—a 5.3× increase exposing critical supply chain dependence. Meanwhile, nearly 2,300 GW of generation and storage capacity sits trapped in US interconnection queues, exceeding the country’s entire installed power capacity. Wait times in major regions now surpass five years, per Hanwha Data Centers.

The constraint is forcing capital reallocation at scale. Alphabet, Amazon, Meta, and Microsoft are expected to deploy over $650 billion on AI Infrastructure in 2026, but half of planned US data center builds face delays due to power and grid limitations, according to Tom’s Hardware in April citing Bloomberg data.

Nuclear Restarts and Geothermal Hedges

Microsoft’s response bypassed renewables entirely. In September 2023, the company signed a 20-year, 835 MW power purchase agreement with Constellation Energy to restart Three Mile Island Unit 1 by 2028—providing 100% of output to Microsoft data centers across the PJM region. “This agreement is a major milestone in Microsoft’s efforts to help decarbonize the grid in support of our commitment to become carbon negative,” Bobby Hollis, vice president of energy at Microsoft, stated in Data Center Dynamics.

The deal reflects a broader pivot. Goldman Sachs projects 85-90 GW of new nuclear capacity needed for data centers alone by 2030, with small modular reactor deployments targeting 2028-2030 timelines. Google, meanwhile, signed a 150 MW geothermal power deal with Ormat in February 2026, diversifying beyond intermittent solar and wind to achieve 24/7 carbon-free energy requirements.

Big Tech accounted for nearly 50% of US clean power capacity additions in 2025—over 20 GW of new generation—according to Trellis. Brookfield’s January agreement with Microsoft, delivering 10.5+ GW between 2026 and 2030, represents nearly three times New York State’s current solar capacity.

M&A Feeding Frenzy in Stranded Assets

The tech sector’s energy procurement arms race is triggering consolidation in fragmented renewable markets. Solar-plus-storage mergers and acquisitions surged in 2025, with developer capital rotation—monetizing mature power purchase agreements—accounting for 21% of all renewable asset deal volume, according to PV Magazine USA in March.

The US solar sector remains highly fragmented with over 1,500 owners across 152 GW of operating capacity, creating massive acquisition opportunities for utilities and independent power producers seeking vertical integration. Assets with existing grid interconnection rights trade at significant premiums—in some cases 40-60% above replacement cost—as developers capitalize on queue scarcity.

The constraint extends beyond hyperscalers. A KPMG survey in December found that 34% of semiconductor industry leaders expressed concern about their ability to procure enough energy for fabrication facilities over the next three years—ranking energy supply alongside tariffs as top operational risks.

Geopolitical Arbitrage

The power bottleneck is reshaping competitive geography. Jensen Huang, CEO of NVIDIA, observed that “low energy costs and looser regulation will help China defeat the US in the AI race,” according to MacroMicro. China’s vertically integrated state utilities can allocate power to AI infrastructure without navigating interconnection queues or PPA negotiations—a structural advantage as US projects stall.

OpenAI CEO Sam Altman framed the challenge starkly: “If there are no major breakthroughs in energy technology, artificial intelligence will not reach its next stage.” The comment reflects growing consensus that energy availability—not model architecture or chip supply—determines the ceiling for AI capability expansion through 2030.