Where Does the $200B+ AI Infrastructure Spend Concentrate: Chips, Networking, Power, or Cooling?
In February 2026, NVIDIA reported quarterly revenue of $68 billion — a single quarter that exceeded many Fortune 500 companies' annual sales. Microsoft, Meta, Amazon, and Google collectively committed over $200 billion in AI capital expenditure for 2026 alone. The question for investors: across the full stack of AI infrastructure — from GPUs to networking switches, power systems to cooling — where is the spending actually concentrating, and which companies capture the most?
Why This Theme Matters Now
The AI infrastructure buildout has entered a new phase. Hyperscalers are no longer experimenting — they are scaling production clusters with hundreds of thousands of GPUs, each requiring high-bandwidth networking, megawatts of power delivery, and advanced cooling. Dell reported a $43 billion AI server backlog in Q4 FY26. Vertiv's orders surged 152% year-over-year as data center power and thermal demands spiked. This is not a future trend; it is a capital cycle already in motion, and the spend is accelerating.
The Companies: Where the Dollars Flow
We examined five companies spanning the AI infrastructure stack — chips, networking, servers, power/cooling, and optical interconnects — to identify where spending concentrates most heavily and which positions offer the best investment exposure.
1. NVIDIA (NVDA) — The GPU Gravitational Center
NVIDIA dominates AI compute with its data center GPU platform, capturing the vast majority of AI training and inference chip spending worldwide.
NVIDIA's fiscal 2026 revenue reached $216 billion, up 65% year-over-year, driven by Blackwell architecture demand. Q4 alone generated $68 billion in revenue with 75% gross margins. Networking revenue tripled year-over-year as NVLink and InfiniBand became essential for GPU cluster interconnects. The company is already shipping Rubin platform samples to customers, securing its roadmap through 2027+. The gravitational pull of NVIDIA's ecosystem — CUDA, software stack, and full-rack solutions — makes displacement nearly impossible in the near term.
| Metric | Value |
|---|---|
| Market Cap | $4,521B |
| Revenue (TTM) | $215.9B |
| Revenue Growth | +65% YoY |
| EBITDA Margin | 67% |
| P/E (fwd) | 22.4x |
| 1Y Price Return | +61% |
NVIDIA captures the largest absolute share of AI infrastructure spending. At 22x forward earnings with 65%+ growth, it remains the most direct and liquid way to play the AI capex cycle.
2. Arista Networks (ANET) — The AI Networking Backbone
Arista provides the high-speed Ethernet switching and routing fabric that connects GPU clusters in hyperscale data centers.
Arista's AI networking revenue is guided to $3.25 billion in 2026, up from $1.5 billion in 2025. Total revenue grew 29% to $9 billion in 2025, with 800-gig adoption accelerating and 1.6-terabit switching emerging. The company's 64.6% gross margin reflects premium pricing power in a market where network performance directly limits GPU utilization. Arista's deep co-design relationships with cloud titans — nearly half its revenue — provide strong visibility.
| Metric | Value |
|---|---|
| Market Cap | $174B |
| Revenue (TTM) | $9.0B |
| Revenue Growth | +29% YoY |
| EBITDA Margin | 46% |
| P/E (fwd) | 39.3x |
| 1Y Price Return | +72% |
Networking is the second-largest line item in AI cluster builds after GPUs. Arista's valuation is rich at 39x forward, but its position as the de facto AI Ethernet supplier justifies a premium.
3. Vertiv Holdings (VRT) — Power and Cooling for the AI Data Center
Vertiv designs and manufactures power management, thermal management, and modular data center infrastructure — the physical backbone that keeps AI clusters running.
Vertiv's Q4 2025 orders exploded 152% year-over-year, with backlog reaching $15 billion — more than double the prior year. Revenue grew 28% organically, and management guided 2026 adjusted EPS of $6.02 with 22.5% operating margins. The company's SmartRun prefabricated solutions and liquid cooling capabilities position it squarely in the AI power and cooling bottleneck. AI racks consume 40-100kW each versus 10-15kW for traditional servers, making Vertiv's products essential.
| Metric | Value |
|---|---|
| Market Cap | $103B |
| Revenue (TTM) | $10.2B |
| Revenue Growth | +28% YoY |
| EBITDA Margin | 21% |
| P/E (fwd) | 44.6x |
| 1Y Price Return | +214% |
Power and cooling are the binding constraint on AI buildouts. Vertiv's $15B backlog provides multi-year visibility, though the stock's 214% one-year return prices in significant execution.
4. Dell Technologies (DELL) — The AI Server Integrator
Dell builds and ships AI-optimized servers, integrating NVIDIA GPUs into enterprise and hyperscale deployments with its PowerEdge platform.
Dell's AI server business reached $25.2 billion in shipments in FY26 with a $43 billion backlog. Management guides $50 billion in AI revenue for FY27, implying ~100% year-over-year growth. Total FY26 revenue was $113.5 billion, up 19%. While Dell's server margins are thinner than component suppliers — EBITDA margin is just 7% — the company captures enormous volumes as the leading integrator for enterprise AI deployments. Traditional server and storage businesses provide a stable revenue base.
| Metric | Value |
|---|---|
| Market Cap | $99B |
| Revenue (TTM) | $113.5B |
| Revenue Growth | +19% YoY |
| EBITDA Margin | 7% |
| P/E (fwd) | 12.8x |
| 1Y Price Return | +56% |
Dell captures the highest revenue share by dollar volume but the thinnest margins. At 12.8x forward earnings, it is the cheapest way to play AI infrastructure — a volume bet, not a margin bet.
5. Coherent (COHR) — Optical Interconnects for AI Scale-Out
Coherent manufactures optical transceivers and photonic components that enable high-bandwidth data transmission between GPUs, switches, and data center clusters.
Coherent's data center segment is surging on 800-gig and 1.6-terabit transceiver demand, with bookings exceeding 4x current shipments. Q2 FY26 revenue grew 22% year-over-year to $1.69 billion, with indium phosphide production capacity set to double by end of 2026. The company is also advancing co-packaged optics (CPO), which could become the dominant interconnect architecture for next-generation AI clusters. However, Coherent's overall profitability is still recovering — FCF was negative in recent quarters as the company invests heavily in capacity expansion.
| Metric | Value |
|---|---|
| Market Cap | $40B |
| Revenue (TTM) | $6.3B |
| Revenue Growth | +19% YoY |
| EBITDA Margin | 20% |
| P/E (fwd) | 39.6x |
| 1Y Price Return | +273% |
Coherent is the highest-beta play on AI interconnect scaling. The 4x book-to-ship ratio signals massive demand, but negative free cash flow and a 273% run-up make it the highest-risk pick.
The Verdict: Ranking the Picks
The AI infrastructure dollar concentrates overwhelmingly in compute — NVIDIA captures the largest and most profitable share of the $200B+ spend. NVDA remains the highest-conviction pick, combining dominant market position with reasonable forward valuation. ANET is the strongest derivative play, as networking scales proportionally with GPU cluster size. DELL offers the best value entry point at 12.8x forward earnings, though margins are structurally lower. VRT has the most compelling backlog story but is priced for perfection after a 214% run. COHR has the most explosive upside if optical interconnect demand materializes as bookings suggest, but is the most speculative given negative FCF and high valuation.
Risks to Watch
- Hyperscaler capex pullback: If cloud providers moderate 2027 spending plans, all five companies face deceleration risk simultaneously
- NVIDIA supply constraints: GPU shortages could slow the broader buildout, impacting downstream networking, power, and server volumes
- Margin compression in servers: Intense competition between Dell, Supermicro, and ODMs could pressure AI server margins further
What to Monitor
- Hyperscaler capex guidance in Q1 2026 earnings calls — any moderation would signal a cycle peak
- NVIDIA's Rubin ramp timeline and whether it triggers another upgrade cycle or a digestion pause
- Vertiv and Coherent order growth rates — sustained 100%+ growth validates the multi-year thesis; deceleration below 50% suggests demand normalization