Santa Clara, California
When your banking app freezes at lunchtime or a retail website crashes on Black Friday, the problem usually starts in a massive data center. The servers running these services are often underpowered, inefficient, or both. On June 2, 2026, at Computex in Taipei, Intel and Foxconn announced a partnership to tackle this issue. The new architecture they introduced could quietly change how fast the internet feels for every American with a smartphone.
The Intel Silicon Brain Behind the Announcement
Intel’s new Xeon 6+ Processor, called Clearwater Forest, is fundamental to this story. The top model, the Xeon 6990E+, has 288 specialized efficiency cores, known as Intel’s Darkmont E-cores, in a single socket. For comparison, that’s 50 percent more cores than AMD’s 192-core EPYC 9965 chip. Intel also says its chip uses less power, with a 450-watt limit compared to AMD’s 500-watt rating.
This chip gets its high core count through a method called Core Stacking. It combines 12 compute tiles made with Intel’s new 18A process, 3 base tiles, and 2 I/O tiles, all joined using Intel’s Foveros Direct3D packaging technology. This approach not only adds more cores but also changes how the cores connect with memory, cache, and each other.
Intel’s benchmarks show that the Xeon 6990E+ delivers 2.26 times higher average performance than the previous Xeon 6780E while using less power. Ericsson’s independent tests found the chip cut rack-level power use by 38 percent and boosted overall throughput by 30 percent compared to a dual-socket Sierra Forest setup with the same number of cores. These improvements are important for data center executives, since power costs can top $10 million per facility per year.
What Foxconn Rackscale Infrastructure Actually Means
The Foxconn partnership turns Intel’s chip into a ready-to-use data center product. Foxconn Rackscale Infrastructure means pre-built server racks that cloud providers or firms can order, set up, and run without spending months on custom engineering. One liquid-cooled rack with the Xeon 6+ can provide 36,864 processing cores in just 32 rack units.
Foxconn, the world’s largest contract electronics maker, brings supply chain expertise and manufacturing scale that Intel can’t match on its own. Their partnership covers everything from chip design and system integration to global delivery, including data center setups and large-scale builds. Foxconn also plans a rack version focused on CPUs for tasks that don’t need extra AI accelerator cards, aiming at cost-effective Inference Execution and standard data processing.
For customers needing AI acceleration, SambaNova’s SN-50 Reconfigurable Dataflow Units work alongside Xeon processors in production-ready racks built for large-scale Inference Execution.
Why Inference Execution — Not Training — Is the Next Infrastructure Battleground
Over the past three years, data centers have focused on AI model training, which needs the huge parallel power of graphics processing units. GPUs became the center of attention because training a massive language model can take thousands of them running for weeks.
However, Inference Execution, which means running a trained model to answer questions, process transactions, or give product recommendations, is a different kind of task. It needs fast, concurrent processing for thousands of users at once, not just raw computing power. In many business situations, modern CPUs with many cores manage this better than large GPU arrays and use much less energy.
Intel is betting its market recovery on this structural shift. As more AI features are embedded into everyday applications search autocomplete, fraud detection, personalized content feeds the volume of Inference Execution requests grows by orders of magnitude while training jobs remain relatively rare. This is where the Intel Xeon 6-plus data center processor infrastructure performance benchmarks start to look genuinely competitive with GPU-centric alternatives.
What Core Stacking Solves That More Chips Cannot
You can’t keep adding servers to a data center forever. Space, power, and cooling all have strict limits. Intel’s Core Stacking method in Clearwater Forest tackles these limits head-on.
Intel fits 12 compute tiles, each with 24 Darkmont cores, into a single processor package. This stacks computing power vertically instead of spreading it across more servers. The Xeon 6990E+ has 576 megabytes of L3 cache, and using two sockets doubles that to 1,152 megabytes, for a total of 576 cores. Memory bandwidth in this setup reaches 1.3 terabytes per second with DDR5-8000.
Intel says its new chips can consolidate servers at up to a 9:1 ratio compared to older Xeon models. For example, a company with 450 old servers could move that work to just 50 new Xeon 6+ nodes, cutting costs for leases, power, maintenance, and cooling all at once. That’s a strong argument for IT executives watching their budgets.
What the Intel Xeon 6 Plus Data Center Processor Infrastructure Performance Benchmarks Mean for Everyday Users
This brings us back to someone trying to log into their bank at noon on a Tuesday. Application latency, or the time between tapping a button and receiving a response, depends on how well the server handles multiple requests simultaneously. If a checkout page handles 50,000 shoppers at once, even tiny delays add up for everyone.
The Intel Silicon Brain architecture inside Xeon 6+ solves this with strong concurrency. Its efficient cores use less power per thread, so a single server can handle more sessions simultaneously without delays. With Foxconn Rackscale Infrastructure, cloud providers can quickly add dense computing power at lower cost, which means users get faster, more responsive apps. percent over competing AMD architecture in Intel’s own benchmarks — translate directly into applications that respond faster, fail less frequently under load, and require less aggressive horizontal scaling to maintain service quality.
The Shift Away From GPU-Only Thinking
The tech industry has long assumed that modern AI infrastructure means using GPU racks, but that idea is starting to change. Intel’s Computex 2026 announcement and the Xeon 6 Plus data center processor infrastructure performance benchmarks that support it make a credible case that CPU-centric systems can handle the heavy, concurrent workloads of enterprise AI more cost-effectively than GPU arrays alone.
The Foxconn partnership turns this idea into reality, moving from specs to actual products. With the world’s largest contract manufacturer on board, getting these systems takes just weeks, down from months. U.S. cloud providers, internet companies, and IT teams will start seeing Xeon 6+ as a real alternative to the GPU-first setups that now dominate data center spending.
So the next time your shopping cart checks out in less than a second, or you get a banking alert before the transaction finishes, it might be thanks to 288 small, fast, and highly efficient cores inside an Intel Silicon Brain working behind the scenes.
Source: Intel Newsroom
