L'ultimo rapporto di Huawei individua 10 direttrici strategiche verso il mondo intelligente del 2035

Huawei’s Intelligent World 2035 report forecasts that annual global AI token consumption will rise 100,000-fold by 2035, with intelligent agents generating more than 90% of traffic. Its companion 2026 digitalization index projects AI will create over $27 trillion of 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. The report advocates major investment in compute clusters, cloud, connectivity, chips, storage, autonomous-driving systems, and agent-specific security infrastructure.

Analysis

This is strategic positioning rather than a verifiable demand event, so it does not warrant a directional trade by itself. The investable implication is that agentic workloads shift infrastructure bottlenecks away from headline GPU capacity toward networking, memory/storage, power delivery and thermal management; these layers monetize recurring traffic and utilization even if model-training spend normalizes. Near-term beneficiaries of confirmed hyperscaler and sovereign-AI capex would likely include AVGO/ANET (high-speed fabric), VRT/ETN (power and cooling), and MU/WDC (memory and data persistence), rather than broad software exposure.

Huawei's framing also highlights a less appreciated competitive risk: export restrictions can redirect Chinese AI build-outs toward domestically qualified stacks rather than eliminate the spend. That is structurally constructive for China-listed infrastructure suppliers but is not automatically incremental for US semiconductor leaders; it raises the probability of localized alternatives in networking, storage, edge inference and automotive compute over the next 6-18 months. For US names with meaningful China revenue, the relevant monitor is not AI capex rhetoric but China-specific order mix, inventory turns and gross-margin commentary.

Consensus remains concentrated on compute accelerators. If agent deployment becomes economically viable, inference token volume could make interconnect bandwidth, electrical gear and data-center availability the binding constraint, extending the capex cycle for VRT and ETN after GPU lead times ease. Conversely, if enterprise agents fail to demonstrate measurable labor or revenue ROI within the next 1-3 quarters, infrastructure orders—not long-dated vision statements—will be cut first; elevated AI-infrastructure multiples leave limited tolerance for a guide-down.

Key Decisions for Investors

• No immediate trade on this release; require corroboration from hyperscaler capex guidance, optical/networking order commentary, or China AI-server shipment data before adding risk.
• On a 1-3 month capex-confirmation catalyst, express the infrastructure bottleneck through long VRT / short an equal-dollar basket of AI application software (IGV): target 10-15% relative upside if data-center power/cooling backlog remains intact; exit if VRT backlog conversion or margins weaken on the next earnings report.
• Accumulate a measured long ANET versus short SMH only after ANET reports sustained AI-backend demand and raises guidance. The pair isolates fabric-content growth from broad accelerator-cycle de-rating; falsify on a material decline in cloud customer spending or an ANET guide below consensus.
• Maintain a 6-18 month watchlist for MU and WDC rather than chase: agentic inference increases the value of high-bandwidth memory and persistent data layers, but entry should wait for pricing discipline and inventory normalization. A renewed memory ASP decline is the stop condition.
• For China-exposed semis, monitor AVGO, QCOM and MU for incremental export-control or localization disclosures; reduce exposure if China revenue declines faster than management can offset through non-China AI demand, since the margin impact can exceed the revenue impact.