The hyperscalers — your Googles, your Amazons, your Microsofts and your Metas — are spending over $100 billion this year on datacentre infrastructure. NVIDIA's GB200 NVL72 system is a liquid-cooled rack with 72 GPUs that draws 120 kW. A decade ago, a typical rack drew 10 kW. Today's AI clusters routinely pull 100-300 kW per rack. That's a 10x jump in power density, and it creates cascading problems.
Power delivery: When AI workloads surge, voltage drops. Drop too low and it crashes training runs that cost hundreds of thousands in compute time.
Heat: That 120 kW generates enormous heat in the same footprint. Some AI clusters spend 40% of total power just on cooling. You're burning electricity to remove heat from burning electricity.
Interconnects: Moving data between tens of thousands of GPUs requires massive bandwidth. On-chip copper fights rising resistivity and electromigration. Denser chip packaging needs bonds that survive thermal cycling. Facility cabling adds weight and resistive losses.
The entire stack — from chip to cooling — hits physical limits simultaneously. Today we're mapping CNT opportunities across datacentre infrastructure.
Key findings: Thermal interfaces, battery additives, and structural composites are production-ready now. Liquid cooling infrastructure is the 2-4 year opportunity as AI shifts to direct liquid cooling. On-chip interconnects are 5+ years out on semiconductor timelines. Markets range from hundreds of millions (thermal, batteries) to potentially billions (liquid cooling) to niche plays worth tens of millions (sensors).
I. Heat Density and Thermal Reliability
The thermal crisis starts at the chip and spreads outward. As rack power stepped from 10 kW to 100-300 kW, thermal interface materials began failing. Greases pump out. Pads age and lose compliance. Eventually processors throttle to prevent damage. Some facilities spend nearly half their power budget removing heat rather than performing computation.
Die-level Thermal Interfaces
CNT thermal interface pads fit between chips and cooling systems, replacing degrading greases with solid materials that install dry and maintain stable heat transfer for years. CNT thermal interfaces have millions of tiny carbon nanotube "bristles" that fill microscopic gaps between surfaces. The tubes are small and flexible enough to conform to microscopic irregularities, dramatically increasing actual contact area between hot chip and cooling system. More contact means better heat transfer. Plus you eliminate repeated thermal paste maintenance.
Conventional thermal paste costs ~$2 per CPU socket but needs replacement every 18-24 months. CNT thermal pads cost ~$8 per socket but last the server lifetime with no performance degradation. Over five years, conventional approaches cost $6-8 per socket in material and labour while CNT costs $8 once. At 100,000 servers with two sockets each, that's $400K in direct savings.
Facility Structural Composites
Facility components like ducting, fan blades, cable trays, equipment racks, raised floor tiles, and seismic bracing benefit from CNT-reinforced polymers at low loadings. The value prop is weight reduction, vibration damping, and protection against static discharge. A datacentre contains kilometres of cable tray and thousands of floor tiles. Weight reduction translates to shipping costs, installation labour, and reduced structural load on buildings.
II. Power Delivery and Reliability
Facility Scale
Uninterruptible Power Supply (UPS) batteries with CNT additives are now standard in energy storage. Lower resistive losses and better mechanical integrity mean faster recharge and longer service intervals. Multiple major battery suppliers including LG Chem have adopted CNT additives. CNT-loaded polymer EMI/ESD shielding is also qualified and production-ready: stable shielding without brittleness and weight of metal flakes.
Component and Rack Scale
Thin-film CNT power storage devices placed close to processors release energy during sudden demand spikes, preventing voltage drops that would crash systems. This is pilot territory — safety documentation and lifetime testing matter as much as electrical performance.
Flexible CNT sensors offer monitoring for strain, humidity, airflow, and corrosive gases. These can be printed onto surfaces or integrated into cable runs. The value is early warning — detecting coolant leaks before equipment damage, monitoring airflow to optimize cooling, identifying corrosive conditions before failures.
III. Interconnect, Packaging, and Signal Integrity
Production-Ready
Static-safe handling and EMI shielding using CNT composites reduce intermittent errors and ease service through weight reduction. CNT-enhanced cable jackets improve fire resistance, meeting building code requirements for flame spread and smoke generation.
Pilot-Ready
Copper-plus-CNT on-chip interconnect using CNT liners or partial fills targets current-carrying capacity and electromigration resistance in power rails and upper metal layers. Benefits — lower power consumption, higher reliability, higher clock speeds — flow through to deployed hardware. Expect movement on semiconductor timelines: development work now appears in production three to five years later.
CNT fibres in cable harnesses present an honest trade-off. Mechanically superb — fatigue-proof, lightweight, high strength. Electrically, commercial fibres achieve ~13-18% of copper conductivity, though development programs show ~61% with promising results on the horizon. For static facility-scale cabling, the conductivity gap remains prohibitive until it reaches 70-80% of copper.
IV. Cooling Systems and Fluid Management
Air Cooling: Production-Ready Applications
CNT composites in air handling — fan blades, impellers, ducting — offer weight reduction and vibration damping. Lighter fan assemblies reduce bearing loads and motor power.
Liquid Cooling: The 2-4 Year Opportunity
Direct liquid cooling infrastructure is the next wave. The transition to liquid cooling for AI creates a chance to specify new materials during initial deployment rather than displacing entrenched solutions in mature systems. CNT coatings on cold plates and heat exchangers target improved heat transfer and corrosion resistance. Testing must include thermal performance versus pressure drop, long-term corrosion in representative coolants, and fouling resistance.
CNT-enhanced coolant fluids are complex. Nanofluids with CNT dispersions show improved thermal conductivity and heat transfer in labs. Barriers: long-term dispersion stability, pump and seal compatibility, fouling potential, and EHS qualification. This needs corporate EHS review and isn't a drop-in replacement — it's system-level requalification.
V. The CNT Opportunity Map
Large opportunities, production-ready (hundreds of millions annually): Thermal interface materials, battery additives, and structural composites have crossed the qualification threshold. These markets share characteristics: qualification complete, supply chains exist, procurement straightforward, growth from market expansion not technology displacement.
Large opportunities, 2-4 year horizon (potentially billions): Liquid cooling infrastructure represents the next major market as hyperscalers transition from air to liquid for high-density AI. Timing advantage is substantial: specifying CNT materials during initial deployment avoids friction of displacing working systems.
Large opportunities, 5+ year horizon (potentially billions): On-chip interconnects, advanced packaging, and CNT photonics represent enormous markets if fundamental barriers are cleared. CNT logic competes against a semiconductor industry with trillions in sunk capital. CNT photonics competes against silicon photonics platforms with decades of development. CNT cabling competes against centuries of copper manufacturing optimization.
Niche opportunities, various timelines (tens of millions): Specialized sensors, point-of-load power storage, specific EMI/ESD components, and CNT fibres in robotic systems address real problems but in markets worth tens of millions rather than hundreds of millions.
CNT adoption will follow opportunity size. The hundreds-of-millions-annually markets are expanding now because qualification is complete. The potentially-billions markets depend on infrastructure build-out timing. The tens-of-billions markets require either semiconductor industry adoption (5+ years) or fundamental breakthroughs (decade-plus if ever).
Yes, hyperscalers are spending $100 billion on datacentre infrastructure. A fraction — hundreds of millions now, potentially billions over the next decade — will flow to CNT materials solving these thermal, power, and interconnect challenges. This is steady 5-10 year growth, not a sudden inflection point.
