Intel and Tencent Cloud Launch New Xeon 6+-Based Cloud Instances Optimized for High-Density AI Agent Deployment

At the 2026 Tencent Global Digital Ecosystem Conference, Intel and Tencent Cloud jointly launched a new generation of cloud server instances based on Xeon 6+ processors. The collaboration goes beyond simple core count increases, specifically targeting the memory resource constraints and task startup latency issues encountered when scaling AI agents in real-world business applications. According to both parties, the comprehensive performance of these instances has improved by over 30% compared to the previous generation.

As AI applications shift from simple Q&A to multi-step agent tasks, CPU loads exhibit distinct burstiness: agents require short peaks of computation when executing code or calling tools but have low utilization while waiting for feedback. However, each agent's runtime state continues to occupy memory even during idle periods. Intel technical experts noted that as cloud providers attempt to increase agent deployment density, memory capacity often becomes a limiting factor earlier than CPU compute power. For instance, in an example involving a 256-core, 12-channel memory Xeon 6+ processor, the average physical memory per vCPU is only about 6GB, which struggles to support high-concurrency agent workloads.

To address this bottleneck, Tencent Cloud introduced the In-Memory Analytics Accelerator (IAA), a dedicated hardware unit built into Intel Xeon CPUs designed to handle data compression and decompression tasks. This offloads work from the CPU and reduces latency. Test results showed that for decompressing 2MB memory huge pages, the IAA solution improved average latency by approximately 16 times and maximum latency by 9 times compared to traditional software solutions like LZ4, maintaining microsecond-level stability under high concurrency. This allows cloud providers to accommodate more idle agent states within limited memory space.