Good morning.

Before we dive in, an announcement worth your attention. 🎉

We’re launching BuildOut AI 2026 — our most ambitious report yet on the Supply Chain of AI. This is a deeply researched, data-backed field guide to the physical limits behind the AI boom: power grids, metals, cooling systems, and construction timelines that ultimately decide how fast AI can actually scale.

This isn’t a trend piece or a surface-level explainer. It’s a ground-up look at the real-world infrastructure that will shape AI’s next decade.

It’s been long in the making — See what’s cooking 👨‍🍳

Microsoft has launched a new initiative to curb the impact of its AI data centers on local communities, pledging that its facilities will not drive up electricity bills.

The company said it will pay full power costs, help fund grid upgrades, and cut water use — while replenishing more water than its data centers consume across the U.S.

Let’s dive into today’s edition.

Google has signed three new long-term power purchase agreements with Clearway Energy Group, securing 1.17 gigawatts of carbon-free energy for its U.S. data centers across Missouri, Texas, and West Virginia.

Project Details: The agreements span up to 20 years and represent more than $2.4 billion in energy infrastructure investment, with construction beginning in 2026 and first projects coming online in 2027–28. The deals expand Google’s existing partnership with Clearway, lifting total contracted capacity between the two to 1.24 GW.

European Expansion: In Europe, Google deepened its clean-energy strategy by extending its partnership with ENGIE in Germany. ENGIE will act as Google’s carbon-free energy manager in Germany, integrating both new and existing power contracts to reduce emissions from AI and cloud operations.

OpenAI said it is strengthening the U.S. AI supply chain by launching a new Request for Proposals focused on domestic manufacturing. The initiative aims to support the physical infrastructure behind advanced AI, including data centers, consumer electronics, and robotics, as demand for compute and energy continues to rise.

Home Advantage: The company is seeking U.S.-based manufacturers to produce key components, including data center hardware, power and cooling systems, electronic assembly, and robotics components. OpenAI said AI infrastructure depends on a broad industrial base — not just chips — and requires reliable domestic capacity to scale quickly and securely.

Big Picture: The move builds on OpenAI’s Stargate initiative, under which it has already announced more than half of its planned 10-gigawatt capacity. OpenAI said the new effort is intended to shorten supply chains, boost resilience, create manufacturing jobs, and reinforce U.S. leadership in artificial intelligence.

Rio Tinto said it will supply copper from an Arizona mine to Amazon for use in artificial intelligence data centers, highlighting how the rapid expansion of AI infrastructure is tightening demand for critical minerals. Copper is essential for wiring, cabling, circuit boards, and power systems used in large-scale data centers.

Under a two-year agreement, manufacturers that supply parts to Amazon Web Services will use copper produced through Rio Tinto’s Nuton leaching technology at a mine owned by Gunnison Copper in Arizona. Financial terms and production volumes were not disclosed. The Nuton process uses bacteria-assisted leaching to extract copper more efficiently from certain rock types.

The deal comes as copper prices hover near record highs, driven by expectations that AI-related growth could boost global copper demand by as much as 50% by 2040.

Meta is making a direct bet on next-generation nuclear power by paying upfront to finance uranium fuel for reactors built by startup Oklo. The deal marks one of the first times a hyperscale tech company has funded nuclear fuel itself.

What’s the plan? The agreement supports Oklo’s plan to develop a 1.2-gigawatt nuclear campus in Ohio, supplying power to the same grid that feeds Meta’s regional data centers.

While financial terms were not disclosed, industry analysts say the move represents a major vote of confidence in advanced nuclear designs and helps Oklo overcome a key bottleneck: securing specialized fuel such as HALEU amid rising prices and geopolitical supply constraints.

Broader Outlook: The deal also highlights broader shifts in energy strategy among tech giants. As electricity demand from AI surges and renewables struggle with intermittency, companies like Meta, Microsoft, Amazon, and Google are increasingly turning to nuclear power — despite regulatory hurdles and unproven reactor technologies — to lock in long-term, stable energy supply.

The Trump administration, alongside a bipartisan group of governors, has urged PJM Interconnection, the largest U.S. power grid operator, to hold a one-time “emergency” auction to supply electricity for rapidly expanding data centers.

Key Details: The proposal would allow data center operators to sign 15-year power purchase agreements tied to new power plants, potentially supporting $15 billion in new generation capacity. Under the plan, data centers would pay for the power whether they use it or not.

However, analysts and PJM itself caution that the proposal is policy signaling rather than immediate reform. Any emergency auction would require approval from the Federal Energy Regulatory Commission, and could take 6–12 months at a minimum.

China exported a record 64.6 billion yuan ($9.3 billion) worth of power transformers last year, up nearly 36% year over year, as a global shortage of the equipment pushed up prices and stretched delivery times. Transformers — essential for stepping electricity up and down for homes, factories, and data centers — have become a critical bottleneck in grid expansion worldwide.

The surge is being driven by rapid data center construction, accelerating electrification, and efforts to modernize aging power grids, particularly in the United States.

The U.S., the world’s largest transformer buyer, faces a roughly 30% supply shortfall, a gap that energy consultants warn could persist into the 2030s. Since 2021, lead times have roughly doubled, reflecting the market's growing constraints.

OpenAI has signed a multiyear, $10 billion agreement with AI chip startup Cerebras Systems to secure large-scale computing infrastructure through 2028. The deal covers roughly 750 megawatts of compute capacity, underscoring the massive energy and hardware requirements for training and running advanced AI models like ChatGPT.

New Partnership: The partnership is part of OpenAI’s broader push to diversify beyond dominant chip suppliers such as Nvidia and AMD. Cerebras builds specialized, wafer-scale chips designed to dramatically speed up AI “inference” — the process by which models generate responses — an area expected to drive the next wave of AI compute demand.

Big Plans: The agreement adds to OpenAI’s rapidly expanding infrastructure commitments, which total roughly $1.5 trillion over the coming decade, far exceeding its current annualized revenue of about $20 billion