Huawei says AI will burn 100,000 times more tokens by 2035

Huawei published its Intelligent World 2035 forecast and a second Global report days before the opening of Huawei Connect 2026. The article indicates the reports frame most of the company’s conference launches, but the provided excerpt contains no specific product, financial, or market-impact details.

Analysis

The investable signal is not the forecast itself but whether Huawei uses its 2026 product cycle to convert domestic Chinese AI demand into a more closed hardware-software stack. That would pressure NVIDIA's China-adjacent revenue opportunity and weaken the assumption that export controls permanently translate into share gains for U.S. vendors; the more material effect would be on the valuation multiple assigned to China-exposed AI growth rather than near-term consolidated earnings.

Over the next 1-3 months, watch for independent evidence of Ascend accelerator availability, utilization at Chinese cloud providers, and developer migration to Huawei's software ecosystem. A credible supply ramp would also benefit mainland memory, networking and server integrators, while creating a procurement headwind for foreign components with China exposure. Conversely, launches without disclosed customer deployments, benchmarked performance, or manufacturing yield data should be treated as marketing rather than a change in competitive position.

Consensus may overstate the immediacy of substitution: training workloads remain constrained by software maturity, advanced-memory access and interconnect reliability, so displacement of CUDA-based infrastructure is likely gradual even if Chinese buyers prioritize sovereignty. The 6-18 month risk is instead bifurcation: Chinese enterprises may accept lower performance-per-watt in exchange for supply assurance, reducing the total addressable market available to Western AI infrastructure companies even before Huawei reaches technological parity.

Key Decisions for Investors

• No directional trade on the publication alone; set an event-driven watch on Huawei Connect follow-through and only reassess NVIDIA (NVDA) China-risk estimates after independently verified customer deployments or quarterly disclosure indicating a China revenue mix change.
• Maintain a 1-3 month relative-value bias toward NVDA versus China AI-sovereignty proxies rather than initiating an outright NVDA short: the falsifier is evidence of meaningful Ascend supply plus production-scale deployments, which would warrant reducing NVDA's medium-term China optionality and reassessing the premium multiple.
• Monitor SK Hynix and Samsung Electronics as second-order read-throughs: confirmed domestic Chinese accelerator scaling could shift demand toward locally sourced memory over 6-18 months, but absent evidence on high-bandwidth-memory qualification and yields, this remains a watch item rather than a position.
• For broad technology exposure, avoid treating Huawei roadmap targets as a near-term capex catalyst. Require disclosed cloud orders, utilization metrics, or supplier revenue acceleration before adding China AI-infrastructure exposure; failure to produce these within the next two reporting cycles would falsify the commercialization thesis.