Intel continues to face challenges in the datacenter market, but signs of improvement are emerging. The company’s long-standing two-core server CPU strategy—dividing its lineup between energy-efficient E-cores and high-performance P-cores—remains central to its approach. E-cores, derived from laptop processor designs, are built for high throughput at low power, ideal for lightweight, scalable workloads. P-cores, designed for single-threaded performance, are better suited for traditional enterprise workloads like databases and virtualization. The upcoming Diamond Rapids Xeon 7, based on Intel’s 18A process (roughly equivalent to 2nm), marks a notable shift: it removes simultaneous multithreading (SMT), also known as HyperThreading. This decision aims to eliminate the performance and security trade-offs associated with running two threads per core, a design choice common in Arm server chips. However, the absence of SMT may hurt overall throughput, especially in multi-threaded environments. Intel has also made strategic changes to its product roadmap. It is now focusing exclusively on high-end 16-channel Diamond Rapids variants, cutting out the eight-channel versions. This shift targets large-scale workloads in high-performance computing, AI host nodes, and database servers. With 18A still in early production and supply limited, Intel must carefully allocate capacity. High-bin Diamond Rapids parts will feature four compute tiles and up to 192 cores. Despite these efforts, supply constraints remain a major hurdle. Intel’s CEO Lip-Bu Tan acknowledged that the company is struggling to meet demand across all its Xeon lines due to bottlenecks in Intel 7 and Intel 3 processes. Even as Intel uses its more advanced Intel 3 process for core tiles, the I/O and base tiles in some designs still rely on older nodes, complicating production. To address this, Intel is accelerating the development of its next-generation Coral Rapids Xeon 8, which is expected to reintroduce SMT—possibly using the 18A process instead of the originally planned 14A. This could speed up the timeline, though Intel has yet to confirm whether 18A will be used. The company is also pushing forward with 14A development, which requires external customers to justify the ramp. Intel claims its 14A process is on track, with a growing customer base and industry-standard PDKs, though external adoption is expected to begin in late 2026. Coral Rapids may also introduce key innovations, including NVLink Fusion ports for coherent GPU and memory fabric integration, and support for DDR6 with up to four memory sticks per channel—significantly boosting memory capacity. On the financial side, Intel’s Q4 2025 results showed a revenue decline of 5.2% to $13.67 billion, but operating income turned positive at $580 million, up from a $401 million loss the year before. The Data Center & AI group, now consolidated under a single unit, saw revenue rise 8.9% year-on-year to $4.74 billion and 15.1% sequentially. Operating profit jumped to $1.25 billion—3.3 times higher than the same quarter last year—highlighting improved margins despite supply issues. However, Intel Foundry continues to lose money, with a $2.51 billion operating loss driven by costs from scaling Intel 7, Intel 3, and 18A, plus development for 14A. Profits from in-house CPUs are still subsidizing the foundry, a pattern that has persisted since Intel’s 10nm struggles. Looking ahead, Intel’s datacenter business may stabilize around $6 billion in quarterly revenue and $2 billion in operating profit, assuming all goes well. The market is shifting, with Arm-based chips expected to capture 25% of server revenue. Intel and AMD will battle for the remaining 75%, likely splitting the bulk of the market. One key takeaway: AMD’s Epyc processors, which offer dual-core variants by reducing cache, provide a cleaner way to differentiate performance tiers. Intel might benefit from adopting a similar strategy.