HUAWEI CONNECT 2026 kicked off in Shanghai, launching with a keynote from David Wang, Huawei’s Deputy Chairman of the Board and Rotating Chairman. Titled ‘Advancing the Agentic World, Building a Solid Silicon Foundation’, Wang set the tone for the event by focusing on the work the company is doing to support the next phase of AI development.
AI is spreading across the world at a pace that outstrips every technological shift before it. Foundation models are already closing in on 10 trillion parameters and that figure is expected to climb past 100 trillion by the end of the decade. The staying power of AI agents is growing just as fast – they can now sustain focus on a single task for hours without interruption and by 2030 they’re expected to manage tasks that unfold over months rather than hours.
In China alone, daily inference token consumption has climbed to roughly 500 trillion, a figure on track to reach the quintillions by 2030.
Progress is moving just as quickly for on-device AI. Smartphone AI models have grown from 3 billion parameters in 2024 to 30 billion today, with hundreds of billions expected within the next few years.
These developments are raising the stakes considerably, demanding underlying technical systems that can deliver at a far higher level of scale, performance and reliability than before. Meeting that bar is the only way the industry can build the powerful AI infrastructure needed to ground the future intelligent world.
The intelligent world is arriving faster than anticipated. To position itself at the front of this shift, Huawei has put its weight behind AI infrastructure, treating it as the silicon bedrock on which the future will be built.
At the core of Huawei’s AI strategy sits computing power, with hardware monetisation as a key priority. Alongside this, the company is pushing its competitive advantage further through innovation in systems and architecture. SuperPoDs and SuperClusters sit at the centre of this push, forming the basis for a solid computing foundation and giving the world a new option to choose from.
Huawei has positioned itself as a significant contributor to open computing ecosystems and plans to keep supporting native training for mainstream foundation models on its own systems, while backing a broad spectrum of models and applications more generally.
On the customer side, Huawei offers flexible compute options – both on-premises and cloud-based – designed to help organisations accelerate intelligent transformation across their industries.
Huawei’s compute portfolio spans the full spectrum – from micro- and low-tier through to mid-range and massive-scale power – driving forward the expansion of on-device and in-vehicle AI and pushing intelligence closer to being genuinely everywhere.
Alongside this, the company is investing in next-generation communications networks, aimed at making compute and intelligence readily accessible to every person, home and organisation.
Introducing SuperPoD innovations
A SuperPoD is defined as a computing system in which multiple computing nodes are tightly coupled through high-speed interconnect protocols, featuring unified memory addressing across physical nodes – functioning like a single logical computer.
SuperPoDs have become key for building AI infrastructure, accepted across industry, academia and research – over 1,000 of these systems have been deployed, with a newer generation now in large-scale commercial use.
UnifiedBus is Huawei’s interconnect protocol, consolidating multiple interconnect protocols into one and significantly reducing protocol conversion overhead. It powers SuperPoDs and SuperClusters and underpins an 11-chip portfolio covering computing, interconnect, storage and management, of which Ascend is the most critical component.
Development of Ascend 960 has outperformed the company’s own expectations. Wang confirmed Huawei is committed to evolving its Ascend chip series on a one-generation-a-year cycle, rolling out the Ascend 970 and 980 chips in 2028 and 2029. Each is expected to continue to double in power while making significant gains in memory and connectivity – improvements attributed to Huawei’s Tau Scaling Law.
