The consequence is a two-front bottleneck for the AI industry. New AI clusters cannot be built without guaranteed power, and even when they are, a new "silicon wall" of tight GPU supply is emerging to throttle what can be deployed inside them .

The Electrical Backbone of AI Is Breaking

The core issue is not a shortage of capital or demand. Alphabet, Amazon, Meta, and Microsoft are expected to spend over $650 billion in 2026 to expand AI capacity . The problem is that the unglamorous, physical components required to power a data center are in critically short supply.

• Transformer Shortages: Transformers, the heavy electrical gear that steps down high-voltage grid power for use in server halls, are one of the most acute bottlenecks. China remains the world's largest producer of this equipment, and trade-war dynamics have tightened supply chains considerably .
• Grid Interconnection Delays: Getting a new facility approved to draw massive amounts of power from the regional grid is a multi-year process in many markets. Interconnection queues are clogged, and transmission upgrades and permitting delays are slowing deployments further .
• A Systemic Gap: The Uptime Institute warned in its 2026 predictions that the widening gap between data center construction timelines and power infrastructure availability will exacerbate challenges across the industry, a problem its research suggests cannot be solved by technology alone .

This has forced a strategic pivot. Infrastructure investors are now abandoning standalone data center deals in favor of vertically integrated strategies that pair digital assets with direct power generation. The site selection process, once a real estate and latency equation, is now dominated by power availability .

Deloitte estimates that by 2035, power demand from U.S. AI data centers could grow more than thirtyfold, reaching 123 gigawatts, up from just 4 gigawatts in 2024 . The long-term trajectory is clear, but the short-term reality is a volatile, uncertain planning environment where multiple bottlenecks are tightening simultaneously .

Inside the AI Server Super-Cycle with Wiwynn

While the grid struggles, the manufacturers building the AI servers are experiencing an unprecedented business boom. Wiwynn Corporation, a Taiwanese cloud IT infrastructure provider, is a prime example of a company riding this wave while trying to outmaneuver the very constraints holding its clients back.

Wiwynn's financial results for fiscal year 2025 showed explosive growth driven almost entirely by AI demand. Consolidated revenue hit NT$950.66 billion (approx. US$29B), a 163.7% jump year-over-year, with AI-related products contributing more than half of the total. Profit after tax surged 124.4% to NT$51.12 billion, with EPS reaching NT$275.06 .

This momentum continued into the first quarter of 2026, where the company reported a record consolidated revenue of NT$276.51 billion—a 62% increase from the same quarter the previous year. Profit after tax grew 44.1% to NT$14.11 billion, with a basic EPS of NT$75.95 .

To address rising AI component costs, Wiwynn implemented a memory procurement agency model in April 2026, a structural change that excludes memory revenue from top-line figures going forward. The company noted that this shift will not affect full-year profitability, and it expects double-digit shipment growth to continue .

Hedging Tariffs with a Global Manufacturing Footprint

Wiwynn's expansion strategy is a direct response to the geopolitical and logistical forces shaking the AI supply chain. Like many Taiwanese ODMs, it is racing to build production capacity closer to its largest customers—North American hyperscalers—to mitigate tariff risks and supply chain fragility.

The centerpiece of this expansion is its new U.S. assembly plant in Socorro, Texas. Initially announced as a $152 million project creating 514 high-skill jobs, the scope later expanded to a $300 million assembly plant . By the end of 2025, the Texas facility had ramped up operations, giving Wiwynn a crucial competitive advantage in securing hyperscaler contracts .

This U.S. push is part of a broader global buildout. Key developments include:

• $900M Combined Investment: Wiwynn and Compal Electronics received approval for a joint $900 million investment to bolster server manufacturing in the U.S. and Mexico. Wiwynn will allocate $220 million to its Mexican unit and $250 million to its U.S. operations .
• Mexico Expansion: Production lines at Wiwynn's Mexico plant were expanded in the second half of 2025 .
• Malaysian Beachhead: The company secured a major AI data center contract with Southeast Asian conglomerate YTL at its Johor, Malaysia facility, diversifying its geographic revenue base .

At its May 2026 shareholders meeting, Wiwynn affirmed it remains "highly confident in the long-term demand for data centers" and is continuing to invest in diversified platforms including next-generation cooling and co-packaged optics .

For now, the industry is caught in a paradox. The companies building the physical backbone of AI, like Wiwynn, are producing at record levels. But the electrical backbone required to turn those servers on is stretched to its limit, creating a high-stakes race between manufacturing output and grid capacity that will define the next chapter of the AI buildout.

The consequence is a two-front bottleneck for the AI industry. New AI clusters cannot be built without guaranteed power, and even when they are, a new "silicon wall" of tight GPU supply is emerging to throttle what can be deployed inside them .

The Electrical Backbone of AI Is Breaking

The core issue is not a shortage of capital or demand. Alphabet, Amazon, Meta, and Microsoft are expected to spend over $650 billion in 2026 to expand AI capacity . The problem is that the unglamorous, physical components required to power a data center are in critically short supply.

• Transformer Shortages: Transformers, the heavy electrical gear that steps down high-voltage grid power for use in server halls, are one of the most acute bottlenecks. China remains the world's largest producer of this equipment, and trade-war dynamics have tightened supply chains considerably .
• Grid Interconnection Delays: Getting a new facility approved to draw massive amounts of power from the regional grid is a multi-year process in many markets. Interconnection queues are clogged, and transmission upgrades and permitting delays are slowing deployments further .
• A Systemic Gap: The Uptime Institute warned in its 2026 predictions that the widening gap between data center construction timelines and power infrastructure availability will exacerbate challenges across the industry, a problem its research suggests cannot be solved by technology alone .

This has forced a strategic pivot. Infrastructure investors are now abandoning standalone data center deals in favor of vertically integrated strategies that pair digital assets with direct power generation. The site selection process, once a real estate and latency equation, is now dominated by power availability .

Deloitte estimates that by 2035, power demand from U.S. AI data centers could grow more than thirtyfold, reaching 123 gigawatts, up from just 4 gigawatts in 2024 . The long-term trajectory is clear, but the short-term reality is a volatile, uncertain planning environment where multiple bottlenecks are tightening simultaneously .

Inside the AI Server Super-Cycle with Wiwynn

While the grid struggles, the manufacturers building the AI servers are experiencing an unprecedented business boom. Wiwynn Corporation, a Taiwanese cloud IT infrastructure provider, is a prime example of a company riding this wave while trying to outmaneuver the very constraints holding its clients back.

Wiwynn's financial results for fiscal year 2025 showed explosive growth driven almost entirely by AI demand. Consolidated revenue hit NT$950.66 billion (approx. US$29B), a 163.7% jump year-over-year, with AI-related products contributing more than half of the total. Profit after tax surged 124.4% to NT$51.12 billion, with EPS reaching NT$275.06 .

This momentum continued into the first quarter of 2026, where the company reported a record consolidated revenue of NT$276.51 billion—a 62% increase from the same quarter the previous year. Profit after tax grew 44.1% to NT$14.11 billion, with a basic EPS of NT$75.95 .

To address rising AI component costs, Wiwynn implemented a memory procurement agency model in April 2026, a structural change that excludes memory revenue from top-line figures going forward. The company noted that this shift will not affect full-year profitability, and it expects double-digit shipment growth to continue .

Hedging Tariffs with a Global Manufacturing Footprint

Wiwynn's expansion strategy is a direct response to the geopolitical and logistical forces shaking the AI supply chain. Like many Taiwanese ODMs, it is racing to build production capacity closer to its largest customers—North American hyperscalers—to mitigate tariff risks and supply chain fragility.

The centerpiece of this expansion is its new U.S. assembly plant in Socorro, Texas. Initially announced as a $152 million project creating 514 high-skill jobs, the scope later expanded to a $300 million assembly plant . By the end of 2025, the Texas facility had ramped up operations, giving Wiwynn a crucial competitive advantage in securing hyperscaler contracts .

This U.S. push is part of a broader global buildout. Key developments include:

• $900M Combined Investment: Wiwynn and Compal Electronics received approval for a joint $900 million investment to bolster server manufacturing in the U.S. and Mexico. Wiwynn will allocate $220 million to its Mexican unit and $250 million to its U.S. operations .
• Mexico Expansion: Production lines at Wiwynn's Mexico plant were expanded in the second half of 2025 .
• Malaysian Beachhead: The company secured a major AI data center contract with Southeast Asian conglomerate YTL at its Johor, Malaysia facility, diversifying its geographic revenue base .

At its May 2026 shareholders meeting, Wiwynn affirmed it remains "highly confident in the long-term demand for data centers" and is continuing to invest in diversified platforms including next-generation cooling and co-packaged optics .

For now, the industry is caught in a paradox. The companies building the physical backbone of AI, like Wiwynn, are producing at record levels. But the electrical backbone required to turn those servers on is stretched to its limit, creating a high-stakes race between manufacturing output and grid capacity that will define the next chapter of the AI buildout.