News

IBM and Cisco announced in December a collaboration to develop a large-scale fault-tolerant quantum computer network, with a target deployment in the early 2030s.  Combining IBM's quantum computing capabilities with Cisco's network infrastructure, this partnership will drive innovation in quantum communication and data center interconnectivity.

Dell establishes Asia Pacific AI Innovation Center in Singapore: In December 2025, Dell launched its Asia Pacific AI Innovation Center in Singapore to drive enterprise AI transformation; simultaneously, it upgraded its Dell AI data platform, adding NVIDIA/Elastic integration capabilities to meet the needs of enterprise-level AI workloads.

Recently, the prices and supply and demand of various NVIDIA computing GPUs have shown a clear divergence, which is a key focus for current enterprise procurement: the price of supercomputing-grade H100/A100 GPUs is continuously falling due to the iteration of new products based on the next-generation Blackwell architecture, with H100 cloud rental prices dropping below 50,000 RMB per month; the NVIDIA L40S ADA computing accelerator card, however, maintains a stable price and is experiencing a surge in demand.  With its 48GB GDDR6 large memory, full-speed PCIe 4.0 interface, and passive cooling compatibility, it has become a mainstream choice for AI inference, edge computing, and enterprise-level deep learning. Cloud rental prices are approximately $1.57 per hour, and it is the standard GPU for flagship GPU servers such as the Supermicro 4125GS-TNRT.  There is ample stock available, and no price increase is expected for the time being.

On December 18th, the Hainan Free Trade Port officially launched its island-wide customs closure, expanding the core zero-tariff list to over 6,600 items (covering 74% of goods). This allows for duty-free circulation among eligible entities within the island, and goods with a value added of 30% through processing can be sold domestically without tariffs. This initiative establishes a hub for both domestic and international circulation, providing zero-tariff access and transit channels for industries such as IT hardware and consumer electronics, reducing cross-border logistics and tax costs, and helping companies expand into Southeast Asian and global markets.

The Nokia 7220 IXR-H6 high-performance AI data center switch, featuring an ultra-high throughput of 102.4 Tb/s and 800GE/1.6TE high-speed interfaces, is compliant with the Ultra Ethernet Consortium specifications and can support the interconnection of millions of XPU-scale AI clusters. The product is available in both liquid-cooled and air-cooled versions, compatible with both SR Linux and open-source SONiC operating systems, and includes AI-enhanced AIOps management tools to significantly reduce network downtime. Designed specifically for ultra-large-scale AI training and inference scenarios, it is expected to be mass-produced and launched in Q1 2026.

On December 5th, Lenovo launched its first high-end AI server for large model training, the Wentian WA8080a G5, at the Heterogeneous Intelligent Computing Summit Forum.  The company also unveiled its "AI Factory" solution and the WanQian Heterogeneous Intelligent Computing Platform 4.0. The Wentian WA8080a G5 features the latest Intel Xeon 6 processor, supports mainstream OAM2.0 GPUs, and is compatible with both air and liquid cooling, meeting the high-performance computing needs of thousand-GPU-scale large model training. The WanQian Heterogeneous Intelligent Computing Platform 4.0 can shorten large model pre-training time by 35% and improve inference performance by 30%.  Furthermore, in its collaboration with the "East Data West Computing" intelligent computing hub, it has increased the model computing utilization rate in thousand-GPU training scenarios from 30% to 60%.

Due to tight chip supply and rising prices, Dell plans to increase the prices of its commercial computers across the board starting December 17th, with increases ranging from 10% to 30%. As commercial computers are one of the core devices in enterprise network terminals, this price increase may impact the cost of large-scale deployment of enterprise terminal networks.  This also reflects the pressure Dell is facing in adjusting its supply chain for network terminal hardware.

US lifts restrictions on H200, boosting interest in domestically produced data center switches: On December 9th, Trump announced that Nvidia would be allowed to deliver H200 chips to approved customers in China, and this policy also applies to companies such as AMD and Intel. Huatai Securities' communications analysis points out that this move is expected to drive increased computing power procurement by leading domestic internet companies, directly benefiting data center switches as a standard product.  New types of switches, such as Switch Trays, may also emerge by 2026.

HPE Expands AI-Native Networking Portfolio, Advancing Autonomous Networking Strategy
On December 8th, at a conference in Barcelona, ​​HPE announced the expansion of its security-enabled, AI-native networking portfolio. This included the introduction of the HPE Juniper Networking QFX5250 switch, featuring Broadcom Tomahawk 6 chips and a bandwidth of 102.4 Tbps, and the MX301 multi-service edge router with a performance of 1.6 Tbps. The company also integrated the AIOps capabilities of two major operations platforms and expanded its partnerships with NVIDIA and AMD to launch networking solutions adapted for AI training and inference, thereby strengthening its AI infrastructure support capabilities.

Samsung completes HBM4 development and sends samples to Nvidia for testing, planning for rapid mass production: On December 3rd, Samsung Electronics announced the completion of the development of its sixth-generation high-bandwidth memory (HBM4). Prototype samples have been sent to Nvidia for quality verification, and the corresponding production line is in the final debugging stage. Mass production will begin immediately after testing is completed.  As a core memory component for high-end AI chips, this product offers significantly superior performance compared to existing products. Its mass production will challenge SK Hynix's market position and also put competitive pressure on domestic companies such as GigaDevice that are developing HBM products. Currently, domestic HBM3 products only offer half the bandwidth of Samsung's HBM4.