INTC: Turnaround – Server Market Momentum and Potential Major Foundry Wins

In the near term, Intel’s growth is underpinned by: 1) a booming general server market, 2) Intel 18A yields (~60%) exceeding market expectations, improving the cost structure, and 3) initial revenue contributions from advanced packaging. We estimate 2025/26/27 EPS at $0.38/$0.64/$0.94, respectively. Long-term, the 14A process—optimized for foundry and ARM architectures—is positioned to attract major clients, potentially including Apple (entry-level M/A-series), Nvidia-Intel co-developed server CPUs, and Nvidia entry-level gaming GPUs.

Datacenter: Capitalizing on AI Tailwinds

General servers were previously viewed as insufficient for AI due to compute limitations. However, two structural shifts are driving a demand resurgence:

• AMX Integration: Sapphire Rapids and Emerald Rapids feature Advanced Matrix Extensions (AMX), a dedicated hardware block specifically for matrix operations. Per Intel, AMX delivers 8x INT8 throughput vs. legacy AVX-512 VNNI. Real-world applications (e.g., BERT inference) show 3-6x performance gains, enabling these CPUs to run Small Language Models like Llama 3 or Mistral without expensive discrete GPUs.

• Shift to SLMs: Demand is bifurcating from LLMs toward SLMs. For occasional inference, dedicated GPU servers are cost-prohibitive. Enterprises in retail, manufacturing, and finance prioritize data privacy and low latency, favoring on-premise general servers over cloud-based solutions.

Beyond AI inference driven demand, 800G network upgrades are forcing a server refresh. 800G requires ~100 GB/s throughput. Legacy platforms like Ice Lake (PCIe 4.0 x16) max out at 31.5 GB/s, only one-third of the required bandwidth. To unlock 800G network performance, enterprises must migrate to next-gen PCIe 5.0 platforms.