Overview
Data centres have become the physical substrate of AI. The bottleneck is no longer just chips; it is power availability, thermal density, liquid cooling, networking, switchgear, optical interconnect, grid interconnection, backup power, land, water and construction speed.
The best investable angle is often not the hyperscaler itself, but the infrastructure supplier that benefits from multiple hyperscalers and cloud customers. Vertiv, Modine and Celestica are category anchors; smaller electrical, optical and manufacturing suppliers become interesting when they sit on a real bottleneck.
StrongAI demand pull
Very Highcapital intensity
Criticalpower/cooling bottleneck
Highvaluation risk
Stock Table
| Rank | Company | Ticker | Role | Category | Research view |
|---|
| 1 | Vertiv | VRT | Power, cooling, thermal systems and data-centre infrastructure | Category anchor | Best public AI data-centre infrastructure anchor; valuation discipline required. |
| 2 | Modine | MOD | Thermal management, data-centre cooling and HVAC systems | Cooling / thermal | Strong data-centre sales momentum; working capital and capacity-expansion costs matter. |
| 3 | Celestica | CLS | AI data-centre infrastructure, switches, servers and advanced manufacturing | AI infrastructure manufacturing | Major AI infrastructure beneficiary; no longer small but highly relevant. |
| 4 | Credo Technology | CRDO | High-speed connectivity, SerDes and optical/DSP infrastructure | Interconnect | Pure data-movement bottleneck; high growth and valuation sensitivity. |
| 5 | Coherent | COHR | Optical components, lasers, transceivers and photonics | Optical infrastructure | AI optical networking and photonics exposure; cyclical balance-sheet considerations. |
| 6 | Bel Fuse | BELFA / BELFB | Power conversion, magnetic components, connectivity | Power components | Interesting smaller power/connectivity supplier; verify data-centre mix carefully. |
| 7 | nVent | NVT | Electrical enclosures, power distribution and infrastructure | Electrical infrastructure | Strong electrification/data-centre adjacency; larger, quality industrial. |
| 8 | Digital Realty / Equinix | DLR / EQIX | Data-centre real estate and colocation | REIT reference | Infrastructure anchors, but valuation and power availability matter. |
Value Chain Map
| Layer | What it supplies | Names | Investment note |
|---|
| Power and electrical systems | UPS, switchgear, power distribution, busbars, backup systems | Vertiv, nVent, Bel Fuse | Power is becoming the AI bottleneck. |
| Cooling | Air cooling, liquid cooling, heat exchangers, thermal management | Vertiv, Modine, nVent | High-density AI racks push cooling requirements higher. |
| Compute manufacturing | Servers, switches, systems integration, hyperscale manufacturing | Celestica, Flex, Jabil reference | Less glamorous but deeply embedded in AI infrastructure. |
| Networking/interconnect | Optics, cables, DSPs, SerDes, switches, transceivers | Credo, Coherent, Celestica | Data movement can become more limiting than raw compute. |
| Real estate / edge | Hyperscale sites, colocation, edge data centres | DLR, EQIX, private operators | Power availability and leasing spreads determine returns. |
Sub-Themes
- AI data-centre power systems
- Liquid cooling and thermal management
- High-speed optical networking
- AI server and switch manufacturing
- Backup power and grid interconnection
- Edge data centres and colocation
Market Forces
- Hyperscaler capex: cloud AI buildouts are the main demand driver.
- Power scarcity: grid constraints delay or reprice data-centre projects.
- Rack density: hotter AI racks pull demand toward liquid cooling and advanced thermal systems.
- Optical bandwidth: AI clusters need faster, lower-power data movement.
- Capex digestion: an AI infrastructure pause would hit the most extended suppliers.
Technology Deep Dive
AI factories are dense electromechanical systems. GPUs and accelerators require power distribution, voltage conversion, cooling loops, network fabrics, optical modules and high-throughput manufacturing. Bottlenecks move: first chips, then networking, then power, then cooling, then grid interconnection.
| Bottleneck | Why it matters | Public angle |
|---|
| Power availability | Data-centre sites cannot scale without electricity and distribution gear. | Vertiv, nVent, Bel Fuse. |
| Thermal density | AI racks push heat loads beyond traditional air cooling. | Vertiv, Modine. |
| Network bandwidth | Training clusters require low-latency, high-bandwidth data movement. | Credo, Coherent, Celestica. |
| Manufacturing capacity | Hyperscale AI systems require reliable production partners. | Celestica. |
| Water and resilience | Cooling and climate constraints connect data centres to water and energy. | Water/energy pages. |
Company Profiles
1. Vertiv · VRT
Power and cooling infrastructure for data centres
Vertiv is the public-market category anchor for AI data-centre infrastructure. It supplies power, cooling, racks, monitoring and services to data-centre customers.
- Recent evidence: Reports around Q4 2025 highlighted Q4 revenue of about $2.88bn, backlog of roughly $15bn and 2026 sales guidance around $13.25bn–$13.75bn.
- Risks: valuation, hyperscaler capex cycle, margins and execution during rapid growth.
2. Modine · MOD
Thermal management and data-centre cooling
Modine is a data-centre cooling and thermal-management beneficiary. Its Climate Solutions segment has been pulled by rapid data-centre growth.
- Recent evidence: Fiscal Q3 2026 Climate Solutions sales rose 51%, data-centre sales rose 78%, and management expected data-centre revenue to grow more than 70% year-over-year.
- Risks: working capital, capacity expansion, margin compression and debt.
3. Celestica · CLS
AI infrastructure manufacturing and switching systems
Celestica is a major manufacturing and advanced-technology supplier to data-centre infrastructure customers, with high exposure to AI networking and compute systems.
- Recent evidence: 2025 revenue was $12.39bn, up 28%, and management raised 2026 outlook to $17bn revenue and $8.75 adjusted EPS.
- Risks: customer concentration, margin sustainability and hyperscaler spending cycles.
4. Credo Technology · CRDO
High-speed connectivity and data movement
Credo is a data-movement bottleneck name, supplying connectivity solutions for hyperscale AI and cloud networking.
- Risks: valuation, customer concentration, high growth expectations and competitive cycles.
5. Coherent · COHR
Optical components and photonics
Coherent gives exposure to lasers, photonics and optical networking used in data-centre and communications infrastructure.
- Risks: debt, cyclicality, optical-demand timing and customer concentration.
Future Scenarios
Bull case: hyperscale AI capex continues, rack power density rises, and power/cooling/networking suppliers see multi-year backlog conversion.
Base case: data-centre buildout remains strong, but winners rotate between power, cooling, optics and manufacturing as bottlenecks shift.
Bear case: cloud capex pauses, AI utilisation disappoints, and extended infrastructure valuations compress.
Signals to Watch
- Hyperscaler capex guidance
- Vertiv backlog and orders
- Modine data-centre sales and margin
- Celestica AI infrastructure outlook
- Optical networking demand and 800G/1.6T ramps
- Grid interconnection and power constraints
Metrics That Matter
- Backlog and book-to-bill
- Organic data-centre revenue growth
- Adjusted operating margin
- Free cash flow conversion
- Customer concentration
- Capacity expansion capex
Risk Map
- Capex-cycle risk
- Valuation risk
- Customer concentration
- Supply-chain bottlenecks
- Grid and permitting delays
- Margin compression during rapid expansion
Convergence
- Data Centres + AI: physical substrate of AI.
- Data Centres + Energy: power, grid, storage and nuclear.
- Data Centres + Water: cooling and site constraints.
- Data Centres + Computing: photonics and advanced packaging.
- Data Centres + Materials: copper, semiconductors and thermal materials.
Research Library
Summary
Data Centres & Digital Infrastructure is one of the most important missing layers in the hub because it turns AI from a software story into a power, cooling and networking buildout. Vertiv is the infrastructure anchor; Modine is the cooling/thermal-management beneficiary; Celestica is the AI infrastructure manufacturing anchor; Credo and Coherent represent the high-speed data movement layer.
Current working conclusion: prioritise power, cooling, networking and manufacturing bottlenecks over generic AI exposure. The theme is strong, but valuation and capex-cycle discipline are essential.