Huawei's latest report proposes 10 key directions to Intelligent World 2035

Huawei’s Intelligent World 2035 report forecasts global annual AI-token consumption will increase 100,000-fold by 2035, with autonomous-agent traffic accounting for more than 90% of total traffic. The company outlined 10 infrastructure priorities, including 100x larger computing clusters, 1,000x lower agent-task costs, AI-native connectivity, agent operating systems, and data-center power and thermal advances. Separately, Huawei and Tsinghua University project AI will create more than $27 trillion in cumulative economic value over the next five years, while global digital and intelligent-infrastructure investment grows at a 19.14% CAGR to exceed $4 trillion by 2030.

Analysis

This is not an earnings-relevant event for public equities; the long-dated infrastructure figures and token-growth assumptions are vendor forecasts rather than independently auditable demand commitments. The investable read-through is nevertheless that AI spending is broadening from training accelerators toward network fabric, power delivery, cooling, storage, and inference orchestration. That mix favors AVGO, ANET, VRT, ETN and DLR/EQIX more than a further indiscriminate extension of the NVDA trade, particularly if hyperscaler capex shifts toward lower-cost inference capacity over the next 6-18 months.

The more important competitive implication is regional bifurcation. China’s push for an indigenous end-to-end AI stack should support domestic compute, optical networking and data-center supply chains while reducing the addressable Chinese market for NVDA, AMD and INTC; sanctions can accelerate substitution even where domestic silicon remains less efficient. Conversely, non-Chinese cloud and enterprise customers will prioritize the highest token-per-watt systems, preserving pricing power for leading networking, HBM and power-management vendors rather than for commodity server assemblers.

Near term, this should not be chased as a catalyst: there is no disclosed order, capex plan, or customer deployment attached to the announcement. Over 1-3 months, the relevant confirmation is whether hyperscaler and telecom guidance shows inference-driven network/power spend growing faster than GPU unit spend; over 6-18 months, sustained AI data-center load growth would tighten grid interconnection and cooling capacity, creating a second-order bottleneck for VRT, ETN and infrastructure REITs. The thesis is falsified by a material deceleration in 2027 cloud capex guidance, falling AI-cluster utilization, or evidence that token efficiency reduces total compute demand faster than usage expands.

Key Decisions for Investors

• No event-driven position in response to this release; treat it as a thematic watch item until Huawei or customers disclose funded deployments, shipment volumes, or procurement commitments.
• On broad AI-infrastructure pullbacks, prefer a 6-12 month basket long AVGO, ANET, VRT and ETN over a concentrated GPU exposure. The risk/reward improves if these names underperform NVDA by 10%+ without a reduction in hyperscaler capex estimates; exit or hedge if cloud capex guidance is cut by more than 10%.
• Use a relative-value expression: long VRT / short a diversified commodity server OEM basket where available, over 6-12 months. Higher rack power density raises cooling and electrical-content dollars per deployment, while server OEMs face more competitive pricing; invalidate if liquid-cooling attach rates and backlog conversion fail to accelerate through the next two earnings cycles.
• Monitor NVDA and AMD China revenue disclosure, SMIC capacity commentary, and Chinese cloud procurement. A faster-than-expected domestic-stack transition is a medium-term headwind to incremental China revenue for US accelerators, but not automatically a short signal absent evidence that displaced demand is not redirected to other regions.
• For cybersecurity exposure, retain PANW or CRWD only as a secondary beneficiary rather than an AI-infrastructure proxy. Autonomous-agent adoption expands identity, data-governance and machine-to-machine attack surfaces, but monetization requires enterprise deployment budgets; wait for billings or platform-module attach-rate evidence before adding.