Huawei navrhuje 10 klíčových směrů pro inteligentní svět 2035

Huawei’s Intelligent World 2035 report forecasts global annual AI-token consumption will increase 100,000-fold by 2035, with autonomous agents generating more than 90% of traffic. The company calls for major investment in AI compute, SuperPoD clusters, storage, connectivity, agent-native cloud platforms, energy-efficient data centers and agent security. Huawei and Tsinghua University also project AI will generate more than $27 trillion in cumulative economic value over the next five years, while digital and intelligent-infrastructure investment grows at a 19.14% CAGR to exceed $4 trillion by 2030.

Analysis

This is not an investable demand signal by itself; it is a vendor roadmap designed to shape enterprise and sovereign procurement. The relevant read-through is that AI infrastructure spending is broadening from accelerators toward the bottlenecks created by inference-heavy, always-on workloads: networking, memory, power delivery, cooling, storage and security. That mix is incrementally favorable to ANET, AVGO, MRVL, VRT, ETN and MU versus a market still disproportionately valuing GPU capacity additions.

Huawei's strategic incentive is to promote an end-to-end domestic alternative stack where its own compute, networking, cloud and device layers are tightly integrated. If Chinese enterprises and state-linked buyers increasingly standardize on that stack, the medium-term risk is not an immediate revenue hit to NVDA but reduced addressable-market optionality for US semiconductor and networking vendors in China, plus faster local substitution for optical, server and switching components. Watch Chinese AI-server tender awards and Huawei ecosystem adoption rather than treating long-dated token-demand forecasts as independently validated.

Over the next 1-3 months, the likely market effect is negligible absent accompanying capex commitments, product launches, or customer deployments. Over 6-18 months, a sustained shift toward agentic inference would favor infrastructure suppliers with recurring content per deployed workload, while creating margin pressure for commodity server assemblers; the critical swing factor is whether inference economics improve through model/software efficiency faster than physical infrastructure demand expands. Consensus may be overestimating a linear compute boom: lower cost per task can unlock volume, but it can also defer hardware purchases if utilization and model compression improve faster than enterprise monetization.

Key Decisions for Investors

• No directional trade solely on this release; treat it as a watch item until it is paired with disclosed capex, hyperscaler orders, or Chinese procurement data.
• For a 6-18 month AI-infrastructure basket, prefer a diversified long in VRT and ETN alongside ANET over pure accelerator exposure: power and thermal constraints are less exposed to a single-chip architecture outcome. Reassess if data-center order growth or backlog conversion decelerates for two consecutive reporting periods.
• Maintain a relative-value watch: long ANET / short a broad commodity server proxy only after evidence that inference clusters are driving high-speed network port growth. Falsify the setup if Ethernet switch revenue guidance fails to accelerate despite announced AI-cluster deployments.
• Monitor NVDA and AVGO China commentary at earnings for evidence of domestic-stack substitution. A material reduction in China revenue expectations, rather than aspirational ecosystem language, would justify reassessing semiconductor geographic exposure.