Intel has announced three new data center products: the Xeon 6+ CPU, a GPU called Crescent Island, and an Ethernet controller called the E835. The launches are the first major data center releases under CEO Lip-Bu Tan, who took the role in March of last year, and they reflect a deliberate push of Intel's product philosophy away from peak performance and toward power efficiency and cost management.
AI infrastructure has created significant pressure on data center operators over the past two years. GPU clusters require dense power delivery, aggressive cooling, and expensive memory, and the cost of running inference workloads at scale has become a recurring concern for enterprise buyers. At the same time, the memory types dominant in high-performance GPUs, particularly high-bandwidth memory, are facing supply constraints and projected price increases. Intel's product decisions are a direct response to that environment.
The Xeon 6+ is designed to fill a gap that Intel argues has emerged in AI deployments: GPUs running underutilized because the CPUs orchestrating their workloads can't keep up. The chip runs 288 low-power cores and includes a feature called Application Energy Telemetry, which gives operators real-time visibility into per-application energy consumption.
The Crescent Island GPU targets inference for agentic AI workloads at 350 watts, which is low relative to competing products from Nvidia and AMD that consume significantly more power. Intel stripped out graphics and 3D rendering support entirely to free up silicon area for AI compute. It also chose LPDDR5X memory rather than HBM, citing cost and a broader supplier base.
HBM supply is concentrated among three vendors (SK Hynix, Samsung, and Micron) and prices are projected to increase sharply through 2027. LPDDR5X draws from a different supply pool, which Intel argues gives it a more stable cost structure.
The E835 Ethernet controller runs at 200Gbps, which is below the 400Gbps threshold that competing cards have reached. Intel's positioning instead emphasizes power consumption: the company claims the E835 uses 47% less power than Nvidia's ConnectX6 and 28% less than a comparable Broadcom card at full bidirectional load.
The consistent thread across all three products is that Intel is not trying to win on performance benchmarks. It is targeting buyers who are managing infrastructure costs rather than buyers who need maximum throughput at any price. That is a defensible strategy given current market conditions, but it comes with real constraints. Intel has lost meaningful ground in GPUs to Nvidia over the past several years and is under pressure in CPUs from ARM-based designs developed by hyperscalers and from AMD's gains in x86.
The efficiency-first approach gives Intel a differentiated message, but it does not resolve the structural challenge of competing across multiple product categories where it no longer holds a clear lead.
