Najnowszy raport Huawei wskazuje 10 kluczowych kierunków rozwoju prowadzących do Inteligentnego Świata 2035

Huawei prognozuje, że do 2035 r. globalne zużycie tokenów AI wzrośnie 100 000-krotnie, a agenty AI będą generować ponad 90% ruchu, co ma wymagać znaczącej rozbudowy infrastruktury obliczeniowej, sieciowej, chmurowej i energetycznej. Wspólny raport Huawei i Uniwersytetu Tsinghua szacuje, że AI wygeneruje ponad 27 bln USD wartości gospodarczej w ciągu pięciu lat, podczas gdy inwestycje w infrastrukturę cyfrową i inteligentną mają rosnąć w tempie 19,14% CAGR i przekroczyć 4 bln USD do 2030 r. Publikacja ma charakter strategicznej prognozy i promuje inwestycje w klastry AI, półprzewodniki, łączność, bezpieczeństwo agentów oraz inteligentną mobilność.

Analysis

This is not a near-term demand signal; it is a vendor-authored framework that reinforces the market’s existing AI-infrastructure narrative. The investable implication is a gradual mix shift from training-centric GPU spend toward inference economics: networking bandwidth, memory capacity, storage provenance, power delivery and thermal management become binding constraints as autonomous workflows move from episodic queries to persistent machine-to-machine traffic. That favors diversified picks-and-shovels suppliers such as AVGO, ANET, VRT, ETN, CEG and WDC more than application-layer AI beneficiaries whose monetization remains less proven.

The second-order risk is that higher token volumes do not translate one-for-one into infrastructure revenue. Model efficiency, inference-specific ASICs and edge processing can sharply reduce centralized compute intensity, creating a relative headwind for merchant GPU expectations while expanding the addressable market for custom silicon and optical/networking. Huawei’s self-interested positioning also matters: export controls could accelerate a parallel Chinese stack, benefiting domestic Chinese component substitution but limiting the accessible revenue pool for US semiconductor equipment and accelerator suppliers.

Over 1-3 months, this should be treated as confirmation of capex expectations rather than a catalyst. The relevant validation points are hyperscaler 2027 capex guidance, AI-networking backlog conversion, power-availability commentary and evidence that inference revenue is growing faster than training spend; absent those, infrastructure multiples remain vulnerable to a crowded-capex unwind. Over 6-18 months, grid interconnection delays and data-center power density—not chip availability—are the most likely bottlenecks, supporting electrical equipment and generation exposures.

Key Decisions for Investors

• No standalone trade on this release; use it only as a thematic confirmation. Add exposure only after hyperscaler earnings validate 2027 AI capex growth and suppliers confirm backlog-to-revenue conversion.
• Construct a 6-12 month basket long VRT, ETN and CEG versus short an equal-dollar AI software proxy (IGV): power and cooling monetization is nearer-term and less dependent on enterprise agent ROI. Exit if data-center order growth decelerates for two consecutive quarters or power-project lead times normalize materially.
• Prefer AVGO over a broad merchant-accelerator expression for 6-18 months: custom AI silicon plus networking provides a hedge against inference shifting toward lower-cost, workload-specific compute. Falsifier: custom-silicon revenue growth misses management’s next two quarterly targets or hyperscalers revert decisively to merchant GPU architectures.
• Watch ANET and optical suppliers for a networking re-acceleration trade after next earnings. Initiate only if AI Ethernet revenue/backlog shows sequential acceleration; the key risk is proprietary fabric adoption or lower token intensity from model-efficiency gains.