Der neueste Bericht von Huawei schlägt zehn Schlüsselrichtungen für die „Intelligent World 2035" vor

Huawei veröffentlichte seine Berichte „Intelligent World 2035“ und den „Global Digitalization and Intelligence Index 2026“ und prognostiziert einen 100.000-fachen Anstieg des jährlichen globalen KI-Token-Verbrauchs bis 2035. Laut dem gemeinsam mit der Tsinghua University entwickelten Index könnte KI in den kommenden fünf Jahren kumuliert mehr als 27 Bio. US-Dollar an wirtschaftlichem Wert schaffen; Investitionen in digitale und intelligente Infrastruktur sollen bis 2030 mit 19,14% CAGR auf über 4 Bio. US-Dollar steigen. Die Mitteilung ist primär strategisch und positioniert Huawei bei agentischer KI, Recheninfrastruktur, intelligenten Netzen, autonomen Fahrzeugen sowie Sicherheits- und Datenschutzlösungen, ohne konkrete Finanzkennzahlen oder kurzfristige Unternehmensprognosen zu liefern.

Analysis

This is primarily a capex-intensity signal rather than a near-term earnings catalyst. If enterprise AI shifts toward persistent, multi-step agent workloads, bottlenecks migrate from model training GPUs to inference networking, memory bandwidth, power delivery and cooling. That favors Broadcom (AVGO), Arista (ANET), Vertiv (VRT), Eaton (ETN), Micron (MU) and Corning (GLW) more directly than software vendors whose valuations already assume rapid agent monetization.

The non-obvious risk is that higher token volumes do not necessarily translate into proportional hardware demand: model compression, routing to smaller models, on-device inference and utilization gains can sharply lower cost per task. Over the next 1-3 months, hyperscaler capex commentary and AI-cluster orders remain the relevant confirmation points; over 6-18 months, grid interconnection queues and data-center power availability are likely to constrain deployment more than chip availability. This increases the strategic value of power-management and thermal vendors, but also raises risk of delayed revenue recognition for GPU and networking supply chains.

Huawei's framing should be treated as ecosystem positioning, not an independently verifiable demand forecast. A bifurcated global AI infrastructure stack is structurally supportive of Chinese domestic substitutes—SMIC (0981.HK), Cambricon (688256.SS), and domestic network-equipment suppliers—but export-control tightening could simultaneously cap their leading-edge capability and force inefficient duplicated capex. Consensus remains too focused on accelerator unit growth; the more durable spread may be between constrained physical infrastructure suppliers and commoditizing AI software layers.

Key Decisions for Investors

• Maintain a 6-12 month long basket in VRT/ETN versus a short position in a broad SaaS proxy (IGV) only if the relative spread retraces 10%: power and cooling have contracted supply and project-critical pricing, while agent-software revenue conversion remains unproven. Exit if hyperscaler capex guidance falls below 10% year-on-year or VRT backlog conversion weakens for two consecutive quarters.
• Prefer AVGO over ANET for the next 1-3 months: custom silicon and high-speed switching provide exposure to AI-cluster scaling without requiring a single network architecture winner. Use a 12-15% stop on the relative trade; falsification is evidence that hyperscalers defer scale-up clusters or materially reduce custom-ASIC programs.
• Add MU on pullbacks ahead of the next two earnings cycles, sized modestly: agentic inference increases memory-intensity, but the trade requires confirmation through HBM pricing and supply-allocation commentary. Do not add if HBM supply expansion drives pricing down faster than bit-volume growth.
• No direct trade on Huawei or Chinese AI infrastructure from this release. Create an alert around new US export-control actions and China telecom/cloud procurement data; a policy-driven domestic-substitution order cycle would be the required catalyst for exposure to 0981.HK or 688256.SS.