The Power Allocation

The nuclear micro reactor is no longer theoretical—it's becoming a procurement decision for data center operators. This episode examines the leading micro reactor designs from Oklo and Westinghouse, their technical specifications, and the realistic timelines for commercial deployment. Key topics covered: Oklo Aurora powerhouse: 15-50 MW units with sodium-cooled fast reactor design Westinghouse eVi...

Data centers are becoming heating plants. Waste heat from AI workloads is being captured and redirected into district heating networks, transforming a byproduct into a revenue stream. This episode examines heat recovery projects across Europe and the business case for integrating heat recovery into new builds. Key topics covered: Microsoft-Fortum partnership serving 250,000 people in Finland Googl...

The networking stack for AI infrastructure is being upgraded simultaneously — from backend intra-cluster connections to frontend switch-to-switch links. Facilities that can't support high-speed networking become bottlenecks that throttle hardware worth tens of millions per rack. This episode examines the cabling requirements for AI-scale compute. Key topics covered: Backend connections migrating t...

Nuclear energy is becoming central to data center power strategy. With 15 reactors expected online in 2026 and hyperscalers signing multi-decade agreements, nuclear offers what solar and wind cannot: dispatchable carbon-free baseload power. This episode examines the nuclear pipeline and its implications for AI infrastructure. Key topics covered: BloombergNEF projection of 15 reactors and 12 GW in...

Microgrids are shifting from supplemental backup to primary power strategy. With grid connection delays extending up to seven years in some markets, operators are building their own power infrastructure. This episode examines the architectures, economics, and strategic implications of microgrid adoption. Key topics covered: Grid constraints driving on-site power adoption Hybrid AC/DC microgrid arc...

Liquid cooling is no longer optional for AI infrastructure — it's the baseline requirement. The data center cooling market is projected to reach $31 billion by 2034. This episode examines the technologies, infrastructure implications, and heat recovery opportunities. Key topics covered: 40% of AI data centers deploying liquid cooling in 2026 Cooling market growing from $11.65B to $31B by 2034 Dire...

Data center permitting timelines have tripled in legacy markets, and community opposition has blocked billions in projects. This episode examines the permitting landscape, regulatory fragmentation, and strategies for navigating local politics. Key topics covered: Permitting extending from 6-12 months to 2-3 years New York moratorium bill for 20+ MW data centers Community opposition and misconcepti...

Meta has created Meta Compute to centralize and accelerate its AI infrastructure buildout. With Prometheus in Ohio and Hyperion scaling to 5 gigawatts in Louisiana, Meta is building at scales that rival utility infrastructure projects. This episode examines Meta's compute strategy and its implications. Key topics covered: Meta Compute division reporting directly to Zuckerberg Prometheus superclust...

Fuel cell adoption has moved from small-scale pilots to multi-billion-dollar primary power deployments for AI data centers. This episode examines the technology, business models, and market shift driving fuel cells as a time-to-power solution. Key topics covered: Bloom Energy's $5 billion Brookfield partnership $2.65 billion American Electric Power agreement Solid oxide fuel cell efficiency advant...

AI-driven data centers require about four times more copper than traditional facilities. With a global deficit forecast and prices projected to hit $12,500-$13,000 per ton, copper is becoming a critical constraint on AI infrastructure. This episode examines the copper supply chain and its implications for investors. Key topics covered: AI data centers requiring 4x copper per facility Global refine...

Five tech giants are deploying nearly $700 billion in a single year to build computing infrastructure — more than the GDP of most countries. This episode examines the capital deployment driving the AI buildout and its downstream effects on utilities, construction, and materials. Key topics covered: Amazon leading with $200 billion capex Alphabet nearly doubling to $175-185 billion Meta Compute tar...

PJM—the grid operator for 13 states—now has authorization to redirect power from data centers to households during grid emergencies. This episode examines what this policy shift means for data center operators and the evolving relationship between AI infrastructure and grid reliability. Key topics covered: PJM's new authority to require data centers to switch to backup generators The blurring line...

NVIDIA's Blackwell GPUs demand 10-30x the power density of traditional data centers, rendering most existing facilities obsolete for AI workloads. This episode breaks down the infrastructure overhaul required to support next-generation AI hardware. Key topics covered: Blackwell rack power requirements: 60-140 kilowatts vs. traditional 4-6 kilowatts Why fewer than 5% of data centers can support 50+...

Texas is on track to reach 40 gigawatts of data center capacity by 2028—a 142% increase from current levels and more than the total capacity of most countries. This episode examines why the Lone Star State is winning the race for AI infrastructure investment. Key topics covered: ERCOT's regulatory advantages over FERC-jurisdictional regions OpenAI's 1.2GW Stargate data center in Milam County with...

The AI infrastructure buildout faces a constraint that capital alone cannot solve: a severe shortage of skilled construction labor. This episode explores how 82% of construction firms report difficulty filling positions, and why projects are being postponed or canceled despite secured funding and permits. Key topics covered: Peak crew sizes jumping from 750 to 4,000-5,000 workers per site 82% of f...

With interconnection queues stretching 4-7 years, AI data center developers are turning to natural gas generation at unprecedented scale. This episode examines how 48 gigawatts of proposed capacity—a third of all U.S. projects—are now designed to bypass the grid entirely using behind-the-meter gas turbines. Key topics covered: The 24x increase in behind-the-meter gas generation projects in two yea...

Traditional European data center hubs face power shortages, land constraints, and regulatory complexity that are pushing new AI development elsewhere. The infrastructure Europe built for the cloud era may not serve the AI era. Key topics covered: Dublin, Amsterdam, Frankfurt: structural constraints on new development Google's €5.5 billion Germany investment struggling to find power Asia-Pacific gr...

Small modular reactors have become the energy solution every AI company wants to talk about. But the timeline reality is stark: the earliest SMR-powered data centers won't come online until late 2027 at the absolute earliest, with more realistic estimates pointing to 2030 or beyond. Key topics covered: Hyperscaler nuclear deals: Meta/Oklo, Amazon/X-energy, Microsoft/Constellation, Google/Kairos Ac...

AI data center investments across the Gulf Cooperation Council countries are projected to exceed $5 billion in 2026. The Middle East has entered the AI infrastructure race with structural advantages that constrained Western markets cannot match. Key topics covered: Saudi Arabia's Humain: $1.2 billion for 250 megawatts of AI capacity OpenAI Stargate UAE: 200MW first phase of a 1 gigawatt cluster So...

OpenAI is publicly warning that the U.S. electrical grid cannot support the pace of AI development. The company estimates the nation needs approximately 100 gigawatts of new generating capacity annually — a buildout rate not seen since post-war electrification. Key topics covered: Data center power consumption projected to reach 123 GW by 2035 (up from 4 GW in 2024) Infrastructure bottlenecks: tra...

A medium-sized data center consumes approximately 110 million gallons of water per year — equivalent to the annual usage of 1,000 households. As AI infrastructure scales, water is emerging as the second binding constraint after electricity, generating community opposition that developers can no longer ignore. Key topics covered: U.S. data center water consumption projected to surge to 68 billion g...

Elon Musk's xAI just announced the largest private investment in Mississippi history: a $20 billion AI supercomputer facility with 2 gigawatts of power capacity. In this episode of The Power Allocation, we analyze why megascale AI infrastructure is moving to where the electrons are. Key topics covered: xAI's $20 billion Mississippi AI training facility Why 2 gigawatts equals the output of two nucl...

Air cooling can no longer keep up with AI. The data center liquid cooling market is growing at 50% annually and has already captured 46% of the market. In this episode of The Power Allocation, we explain why direct-to-chip and immersion cooling are now mandatory for AI infrastructure. Key topics covered: Why AI accelerators generate heat densities air cooling cannot handle The $11 billion liquid c...

The political backlash against AI's power consumption has arrived. Bernie Sanders wants a moratorium on data centers. Ron DeSantis is blocking AI expansion in Florida. And Tom Cotton's DATA Act could let facilities build their own off-grid power. In this episode of The Power Allocation, we map the emerging political fault lines around AI infrastructure. Key topics covered: Senator Sanders' call fo...

Power transformers have become the unexpected bottleneck in AI infrastructure. Lead times now exceed two years, and data centers across Silicon Valley sit dark — fully built but unable to operate. In this episode of The Power Allocation, we examine why electrical equipment is the new constraint on AI growth. Key topics covered: Why transformer lead times have stretched to 24+ months The 70% price...