Power-First Data Centers: Amazon’s $715 B AI Spend, a 500% RAM Price Surge, and 66 GW under construction (2026)

The era of predictably cheap and abundant computer memory has ended, triggering a fundamental reordering of the data center industry. The unprecedented demand for Artificial Intelligence (AI) compute has created a severe memory shortage, with prices for standard RAM surging by as much as 500% in late 2025 and early 2026. This “RAMageddon” is not merely a component issue; it is a market shock that has shifted the primary bottleneck in data center development from processing power to electrical power. As hyperscalers like Amazon, Google, Microsoft, and Meta commit to spending between $715 billion and $1 trillion on AI infrastructure in 2026, over 60% of this capital is now directed not at servers, but at securing power, land, and cooling. The dominant strategy for 2026 is no longer about compute density, but about navigating a power-constrained supercycle where securing gigawatts of energy is the key to market leadership.

Key Developments Shaping the 2026 Data Center Market

The collision of AI demand and physical infrastructure limits has produced several defining market trends in 2026. These developments illustrate a clear shift in investment priorities, where the cost and availability of memory and power now dictate the pace and structure of the entire digital infrastructure landscape.

1. Hyperscaler AI Spending Surge

Companies: Amazon, Google, Meta, Microsoft
Metric: $715 billion in planned 2026 AI infrastructure spending, with some estimates reaching $1 trillion. This follows a 78% year-over-year increase in data center Cap Ex in Q 1 2026.
Application: Building out massive data center fleets to support proprietary and commercial AI services.
Source: Big Tech Plans $715 Billion AI Infrastructure Spend in 2026, AI Infrastructure Buildouts and Memory Cost Inflation Drove …

2. The “RAMageddon” Price Shock

Metric: 300-500% price increase for standard DDR 4 and DDR 5 memory between Q 3 2025 and early 2026.
Details: A standard 32 GB DDR 5 kit, which cost approximately $100, surged to as high as $450.
Application: Affects the entire technology market, from enterprise servers to consumer PCs, laptops, and gaming consoles.
Source: RAM Prices 2026: Why Memory Skyrocketed 400% (DDR 5 …

3. Memory Supply Seizure by AI

Company: SK Hynix (as market commentator)
Metric: AI data centers are consuming an estimated 70% of the global high-bandwidth memory (HBM) and advanced DDR 5 supply.
Details: The shortage is so profound that market leaders warn it could persist beyond 2030, fundamentally altering supply chains. The memory market is forecast to generate $551 billion in 2026, double the revenue of contract chip manufacturers.
Application: AI accelerators and servers for large model training and inference.
Source: 2026 Memory Chip Shortage: SK Hynix Warns It May Last Past 2030, Memory makers are set to earn $551 billion from the AI boom …

4. North American Data Center Capacity Expansion

Metric: Over 66 GW of data center capacity is currently under construction in North America alone.
Details: This represents a massive buildout cycle attempting to meet AI-driven demand. Globally, active capacity is projected to grow sixfold from 24.4 GW in 2025 to 147.1 GW by 2035.
Application: Large-scale data center campuses for hyperscale and AI workloads.
Source: North America Data Center Report Midyear 2026 – JLL, Forecasting Data Center Capacity Increases in the AI Era

5. Capital Expenditure Shifts to Power and Buildings

Metric: More than 60% of AI infrastructure capital expenditure is now allocated to physical infrastructure.
Details: This includes power procurement and generation, high-density cooling systems, land acquisition, and building construction, marking a decisive shift away from a compute-centric cost model.
Application: Foundational elements of data center construction and operation.
Source: Hyperscaler Capex 2026: Where the $300 Billion in AI …

Table: 2026 Data Center Market Shock Events

The Power-First Paradigm: How a Memory Crisis Redefined Infrastructure

The memory shortage has acted as a catalyst, accelerating the transition to a “power-first” investment model. With compute hardware both expensive and supply-constrained, the new competitive frontier is the acquisition and management of the physical assets needed to power and cool massive AI fleets. This changes the valuation calculus for the entire industry.

Beyond Compute Density

For years, the primary metric of a data center’s value was compute density—how many racks and servers could be fit into a given footprint. In 2026, the most valuable assets are power purchase agreements (PPAs), land with substation access, and water rights for cooling. With over 60% of AI Cap Ex now dedicated to these foundational elements, the value chain has been inverted. Companies that own or can secure land and power hold the ultimate leverage over hyperscalers and AI developers, a stark reversal from the chip-centric model of the past decade.

The Rise of On-Site Generation

The strain on public power grids has become a critical barrier to growth, leading to project delays and cancellations totaling over 7 GW of planned capacity in the U.S. In response, operators are aggressively pursuing “behind-the-meter” power strategies. This includes building dedicated microgrids using solutions from firms like Power Secure and deploying on-site generation. While historically used for backup, technologies like fuel cells are now being planned for primary power, as seen with Oracle’s move to use Bloom Energy fuel cells. This trend is a direct effort to bypass grid limitations and secure the power needed for AI expansion.

North America’s 66 GW Buildout and Its Strain on the US Power Grid

North America, particularly the United States, remains the epicenter of the AI data center buildout, with a staggering 66 GW of capacity under construction. This historic expansion is creating unprecedented demand for electricity, pushing regional power grids to their breaking point and exposing a critical national infrastructure vulnerability.

Concentration in Primary Markets

According to JLL research, the bulk of this construction is concentrated in established data center alleys like Northern Virginia, Phoenix, and Dallas. These regions offer favorable business climates and existing fiber infrastructure, but their power capacity is finite. The sheer scale of demand—with single AI campuses requiring hundreds of megawatts, equivalent to the consumption of a small city—is overwhelming local utility planning and leading to multi-year moratoriums on new grid connections in some areas.

Grid Constraints as the New Barrier

The U.S. electric grid was not designed for the exponential load growth driven by AI. The Belfer Center highlights this as a watershed moment, where data center demand is directly impacting grid stability and national energy policy. This has made grid interconnection queues, not silicon fabrication, the most significant bottleneck in deploying new AI capacity. Developers who cannot secure power are sitting on billions in stranded assets, forcing a strategic re-evaluation of site selection toward regions with available power, even if they are outside traditional hubs.

Infrastructure Maturity: From Advanced HBM to Concrete and Cooling

The 2026 market dynamics have shifted the definition of “advanced technology” in the data center space. While cutting-edge chips remain vital, the critical path to deployment now runs through mature, industrial-scale infrastructure like high-density cooling systems and power generation, which are proving far harder to scale than silicon.

HBM and DDR 5 as Table Stakes

High-Bandwidth Memory (HBM) and DDR 5 are technological marvels, essential for feeding data-hungry AI processors. However, with demand outstripping supply by a wide margin, their availability has become a given requirement rather than a competitive differentiator. Access to a stable supply of this memory is now “table stakes” for any serious AI player, often requiring direct, multi-billion-dollar commitments to memory manufacturers like SK Hynix and Samsung.

Cooling’s Critical Role in AI Density

As rack power densities soar past 100 k W to support AI accelerators, traditional air cooling is no longer viable. This has elevated the importance of advanced liquid cooling solutions. The capital expenditure on cooling systems has become a major line item, driving innovation and investment in companies that specialize in direct-to-chip or immersion cooling. Providers like Aligned Data Centers are gaining prominence by engineering facilities specifically for these high-density thermal challenges, making cooling expertise as critical as network architecture.

Hyperscaler Strategy 2027: Will On-Site Generation Render the Grid Obsolete for AI?

The pivotal strategic action for the coming year is the potential large-scale decoupling of AI data centers from the traditional electric grid. To de-risk their multi-trillion-dollar AI roadmaps from public grid constraints, hyperscalers are expected to pivot from merely contracting power to directly funding, owning, and operating their own primary power generation assets at a gigawatt scale.

• Signal: The allocation of over 60% of Cap Ex to physical infrastructure already demonstrates a strategic acceptance of funding foundational assets. The next logical step is moving up the energy value chain from consumer to producer.
• Signal: Early projects like Oracle’s deployment of fuel cells for primary power are proof-of-concept for grid independence. Watch for these to scale from single-site solutions to a portfolio-wide strategy among all major cloud providers.
• Signal: As grid-lock worsens, hyperscalers will likely explore more capital-intensive but long-term solutions, including partnerships to develop dedicated Small Modular Reactors (SMRs) for their data center campuses.
• Signal: Look for an increase in vertical integration, with hyperscalers making direct investments or acquisitions of energy technology companies to secure a proprietary advantage in power generation and management. Some may even explore radical long-term alternatives like space-based AI data centers to bypass terrestrial constraints entirely.

The questions your competitors are already asking

This report covers one angle of the AI infrastructure market. The questions that matter most depend on your work.

• data centers building their own power plants
• new data center construction projects US
• top liquid cooling companies for data centers
• data center projects delayed by power shortages

This report does not answer these. Enki Brief Pro does.

Your question, your angle, your framework. SWOT, PESTL, scenario modelling. The same niche depth, built around the decision your work actually depends on.

Run your first brief in Enki Brief Pro