L13Pillar 2: Put Chips on Server· Pillar 2: Put Chips on Server

Assemble the Server

Server OEMs & Systems Assembly

Supply Constraint

3/10
3/10

How hard it is to add capacity in this layer. Suppliers, lead times, capital intensity, geographic concentration.

Demand Pull

8/10
8/10

How much of this layer's revenue is AI-driven today and how fast that mix is growing.

Multiple competitors. GPU supply from NVIDIA is the real constraint — not assembly.

Layer Dependencies

Server OEMs take finished GPUs (L06), system memory (L08), storage (L07), power regulators (L09), cooling (L10), timing (L12), and transceivers (L11) and solder them onto motherboards. Dell, HPE, SMCI compete for NVIDIA GPU allocation.

Deep Dive

The Power Grid Reckoning trend starts here. Data centers need three things from a site: power (preferably 100MW+), fiber connectivity, and water for cooling. Finding sites with all three is increasingly difficult as hyperscalers compete for the same limited powered land in Virginia (Dominion territory), Texas (ERCOT), and emerging hubs like Mississippi, Indiana, and Saudi Arabia.

NUAI (New Era Energy, originally tracked as the deep-dive company in this project) exemplifies the powered-land thesis: if you control a site with behind-the-meter power generation (solar, gas, nuclear), you bypass the 5-10 year utility grid interconnection queue. That queue is the binding constraint. Dominion Energy's PJM interconnection queue in Northern Virginia has more than 100GW of projects waiting — years of delay that no amount of money can accelerate.

Fluence Energy (FLNC) provides battery energy storage systems (BESS) that smooth renewable power output on powered-land sites. STEM Inc offers AI-powered energy optimization software. Bloom Energy provides fuel cells as backup/supplemental power. These companies sit at the intersection of "power for AI" and "powered land as the scarce resource."

The macro insight: the AI buildout is fundamentally a real estate and energy procurement problem before it's a technology problem. The most valuable AI infrastructure asset in 2026 may not be a GPU — it may be a 500-acre site with 500MW of utility interconnection rights. That's what hyperscalers are competing for, and that's what makes L13 companies strategically valuable beyond their current revenue.

CHAIN INSIGHT

Utility grid interconnection queues of 5-10 years are the binding constraint. Powered land with behind-the-meter generation bypasses the queue — making land+power the scarce resource.

Companies in This Layer

Speed to market
Super Micro Computer

Dominant AI server assembler. Fastest time-to-market for new NVIDIA platforms. Modular building block design philosophy. Accounting investigation overhang.

Enterprise relationships, global support infrastructure, and financing capability — not technology differentiation
Dell Technologies

Traditional server leader growing AI server content. PowerEdge AI portfolio and enterprise relationships. Strong in enterprise accounts preferring full-service.

Government/HPC moat via Cray is real; ProLiant server moat is weak; Juniper adds networking switching costs but integration unproven
Hewlett Packard Enterprise

HPC heritage with Cray acquisition. Strong in government and research AI deployments. Growing enterprise AI server portfolio.

Enterprise niche
IBM

Power and Z-series servers. AI infrastructure through Watsonx platform. Enterprise AI deployments. Niche but relevant for specific workloads.

Hyperscaler qualification + ZT integration
Sanmina Corporation

Contract electronics manufacturer. Builds servers, networking equipment, and other complex systems for OEMs. Benefits from overall AI hardware volume growth.

ODM-Direct model, 800G/1.6T networking design wins, hyperscaler engineering depth
Celestica

AI server and networking hardware ODM-Direct for hyperscalers — designs and builds 800G/1.6T switches, liquid-cooled rack systems, and custom AI compute platforms.