Unstoppable Surge in AI Infrastructure Orders: Global Computing Infrastructure Enters an All-Out Boom Cycle
From hundreds of billions of dollars in capital expenditure by North American cloud vendors to the mass launch of domestic intelligent computing centers; from AI server production lines running three shifts at full capacity to liquid cooling and optical interconnection orders booked through next year, the global AI infrastructure industrial chain is witnessing an unprecedented order boom in 2026. Driven by three converging forces—the iteration of large language models, commercialization of AI Agents, and national computing sovereignty strategies—the computing foundation has shifted from conceptual expectations to large-scale delivery. Robust order growth across the entire industrial chain has become an irreversible trend, ushering in an industrial expansion wave spanning chips, servers, networking, thermal control and storage.
I. Full-Scale Demand Explosion: Three Core Logics Underpin Sustained Massive Long-Term Orders
The skyrocketing orders for AI infrastructure stem not from short-term speculative hype, but from three irreversible long-term demand drivers that continuously generate incremental demand. The widening global computing gap compels enterprises to ramp up procurement relentlessly.
1. Global Tech Giants Launch a "Computing Arms Race", Pushing Capital Expenditure to Record Highs
Overseas hyperscale cloud providers stand as the primary catalyst for order growth. The total annual capital expenditure of the four major North American cloud vendors is projected to exceed USD 710 billion in 2026, a year-on-year surge of over 70%. Meta, Microsoft, Google and Amazon each allocate over USD 100 billion annually to AI infrastructure. OpenAI has partnered with Oracle on a USD 500 billion Stargate computing expansion initiative, releasing massive long-term framework orders on an ongoing basis.
Dell secured USD 24.4 billion in AI server orders in a single quarter, with backlog orders exceeding USD 51.3 billion. Its full-year revenue forecast for AI hardware has been raised to USD 60 billion, while hardware delivery lead times have stretched to 3–4 months.
China’s computing market is expanding in tandem. Eight national computing hubs and the East Data, West Computing project continue to build high-density intelligent computing centers. The rollout of large models in government, finance and manufacturing sectors has triggered a surge in inference computing demand, forming a dual-track procurement model covering both training and inference workloads.
Lenovo’s backlog of AI infrastructure orders has surpassed USD 21 billion. Leading full-system manufacturers hold over USD 100 billion in annual reserved orders, with production line utilization consistently above 95%. Lead times for new orders have extended from four weeks to sixteen weeks.
2. Closed-Loop Commercialization of AI Applications Extends Computing Demand from Labs to All Industries
Large-scale deployment of AI Agents, multimodal large models, industrial intelligence and autonomous driving has completely unlocked the long-term ceiling for computing demand. Previously, computing resources were concentrated solely on model training for top tech firms. Today, small and medium-sized enterprises alongside traditional industries are undergoing full-scale AI transformation, driving parallel spikes in orders for inference servers, edge computing and private intelligent computing clusters.
Industry estimates indicate the global market related to large models will surpass USD 2000 billion in 2026. Commercial revenues feed back into capital expenditure, creating a virtuous cycle: rising application revenue → increased computing procurement → iteration of superior AI products.
The sharp rise in unit value further amplifies order volumes: a high-end AI server delivers hardware value ten times that of a conventional general-purpose server. When paired with liquid cooling systems, 800G/1.6T optical modules and high-speed storage, the total procurement cost per rack computing cluster rises substantially, doubling overall order value for equivalent rack deployments.
3. Implementation of National Computing Sovereignty Policies Drives Infrastructure Procurement via Mandatory Regulation
Countries worldwide classify AI infrastructure as core digital strategic assets and introduce massive industrial support policies to stimulate computing construction orders from top to bottom. The EU’s InvestAI program plans to inject EUR 200 billion to build AI super factories; the U.S. Chips Act allocates hundreds of billions in subsidies to support domestic computing clusters; Japan, the UK and India have each launched special hundred-billion-scale programs for AI computing infrastructure.
Domestic policy mandates directly fuel order growth across niche segments. The Ministry of Industry and Information Technology stipulates that newly built intelligent computing centers in 2026 must achieve a PUE below 1.15 with liquid cooling penetration of no less than 60%. High-density data centers will fully phase out air cooling solutions, pushing liquid cooling equipment orders up 200%–300% year-on-year. All newly built ten-thousand-card clusters will be fully equipped with liquid cooling systems.
Combined with policy dividends and domestic substitution requirements, homegrown servers, optical modules and thermal control equipment have been included in bidding lists for government and enterprise computing projects, with incremental orders increasingly tilting toward local manufacturers.
II. Record Order Volumes Across the Entire Industrial Chain, With Booming Prosperity in All Subsectors
AI infrastructure encompasses five core segments: computing hardware, high-speed interconnection, thermal control & cooling, supporting storage and intelligent computing data centers. Every link currently faces order backlogs and supply shortages, with industrial value transmitted seamlessly up and down the chain without material bottlenecks.
1. AI Servers: Core Carriers of Computing With Sustained Full Order Books
As the core hardware enabling computing deployment, AI servers sit at the heart of the current order boom. China’s AI server market size is expected to exceed CNY 3500 billion in 2026. Top manufacturers including Inspur, Unis and Lenovo delivered mass batches of liquid cooling integrated racks in Q2, while overseas ODM foundries operate round-the-clock three-shift production schedules.
New high-power platforms NVIDIA GB300 and Rubin are ramping into mass production, with single-rack power output exceeding 100 kilowatts and drastically lifting full-system unit value. Domestic computing clusters built on Ascend and Cambricon chips have won bulk bids for government and energy projects, driving rapid growth in orders for information technology innovation computing infrastructure.
2. High-Speed Optical Interconnection: Explosive Orders for 800G, 1.6T, CPO and OCS Products
Clusters with ten thousand or one hundred thousand GPUs generate rigid demand for high-speed networking, making optical interconnection the breakthrough solution to eliminate computing expansion bottlenecks. 800G optical modules are in mass delivery, while 1.6T products have entered large-scale commercial deployment. Leading optical component manufacturers have order backlogs stretching through year-end.
OCS optical switches excel at ultra-large-scale computing clusters thanks to low latency and low power consumption, projected to register a compound annual growth rate of 58% from 2026 to 2029. Each CPO optical engine carries a supporting value of CNY 50,000–80,000, growing in tandem with full-system shipments. Overseas cloud vendors continuously place multi-year framework procurement orders, creating persistent shortages of optical chips and high-speed connectors.
3. Liquid Cooling Thermal Control: The Highest-Growth Segment With Unmatched Certainty
Rising chip power draw has pushed past the physical limits of air cooling technology. Liquid cooling has evolved from an optional accessory to a mandatory standard for high-end computing, ranking as the fastest-growing subsector in terms of order expansion.
China’s liquid cooling market size is set to double in 2026. Leading manufacturers reported year-on-year order growth exceeding 200% in Q1, with both immersion and cold-plate liquid cooling solutions expanding in parallel. NVIDIA has fully opened its new rack architecture to domestic liquid cooling suppliers, enabling local manufacturers to directly enter global top-tier computing supply chains.
Full localization has been achieved for CDU units, cold plates and insulating coolant, breaking overseas monopolies and driving cost declines. Retrofit orders for small and medium-sized computing data centers have surged, with core component manufacturers locking in full production capacity through 2027.
4. High-Speed Storage, Power Distribution & Data Center Auxiliaries: Structural Shortages Spur Supporting Equipment Orders
Massive read-write workloads for large AI models create robust demand for premium storage. Manufacturers of AI server SSDs and memory modules expect up to tenfold half-year earnings growth. Distributed storage equipment enjoys sustained volume growth fueled by computing leasing and IT innovation projects.
High-power computing racks drive demand for high-voltage DC power supplies, intelligent power distribution cabinets and high-density transformers. Data center projects across the globe face delays due to insufficient power supporting equipment capacity, leaving upstream power device producers swamped with orders.
Civil engineering for intelligent computing facilities, modular racks and integrated cabling projects keep landing continuously. Computing leasing operators keep constructing new racks, unlocking steady medium-to-long-term engineering orders.
III. Long-Term Sustainability of Order Prosperity: Three Pillars Mitigate Cyclical Volatility
Market participants have raised concerns over potential short-term bubbles amid computing capacity expansion. Nevertheless, multiple tangible factors confirm that the current AI infrastructure order boom is not a temporary spike, and high growth will persist for at least 3–5 years.
First, rigid barriers restrict capacity expansion on the supply side. Manufacturing high-end AI servers, premium optical chips and precision liquid cooling equipment requires stringent thresholds covering precision manufacturing, specialized materials and proprietary processes. Capacity expansion typically takes 1–2 years, making it impossible to match explosive near-term demand and prolonging order backlogs. Leading manufacturers secure predictable future revenue via long-term supply agreements.
Second, continuous technological iteration generates fresh incremental demand. Constant upgrades from GB200 to GB300, custom ASICs and next-gen TPUs require regular hardware refreshes for computing clusters. MoE large models and distributed training architectures keep raising per-rack computing density. Every round of hardware iteration triggers new equipment replacement orders, opening vast market space for retrofitting existing data centers.
Third, domestic substitution unlocks new growth avenues. High-end computing hardware and core components once relied heavily on foreign imports. Years of technological breakthroughs have drastically lifted localization rates for full servers, liquid cooling systems, mid-to-low speed optical modules and storage devices. Government and enterprise computing projects prioritize domestic products, while global vendors diversify supply chains by onboarding Chinese suppliers, widening order channels across both domestic and overseas markets.
Industry risks have also been largely absorbed. The industrial chain has transitioned from pure order expectations to tangible earnings delivery. Top liquid cooling, optical module and server firms maintain robust gross profit margins, while extensive long-term framework orders stabilize cash flow. This greatly strengthens earnings predictability and weakens cyclical swings across the sector.
