Intel Goes Custom Chip Foundry, Nvidia Named as Key Customer

Intel has taken two decisive steps to establish itself as a serious contender in the custom chip design arena, appointing a dedicated executive to lead its custom silicon business and securing a major multi-year agreement to produce bespoke Xeon CPUs for Nvidia’s AI platforms. This move signals a significant evolution in the company’s strategy, positioning it as a direct competitor to established contract chip designers.

For more than ten years, Intel has provided semi-custom Xeon processors to select clients. These chips are typically modified versions of standard Xeon models, adjusted for specific frequency, power, packaging, or feature requirements. Hyperscalers and large data center operators often request these tailored processors to better handle their unique computational workloads. Back in 2021, Intel outlined ambitions for even deeper customization involving core architecture and proprietary intellectual property, though until recently, its only publicly acknowledged custom design was a Xeon processor developed for Amazon Web Services.

The level of customization for these projects remains largely undisclosed. We do know, for instance, that AWS’s custom Xeon 6 CPU operates at a higher all-core turbo frequency and supports faster memory than its off-the-shelf counterpart. Still, this falls short of the extensive modifications many hyperscalers implement for their in-house data center processors. To genuinely compete in a market filled with specialists like Alchip, AMD, and Marvell, Intel needed to strengthen its custom silicon operations. The recent appointment of Srini Iyengar to lead the Central Engineering Group addresses this need directly.

Iyengar brings over twenty years of experience at Intel, with a strong focus on custom architecture for infrastructure platforms. His background includes key roles in developing Arm-based Infrastructure Processing Unit systems-on-chip and architecting special-purpose accelerator subsystems for server CPUs. His expertise in performance, power, and area optimization, along with his experience collaborating across IP vendors and manufacturing teams, makes him well-suited to grow Intel’s custom design services.

The announcement of a partnership with Nvidia represents Intel’s most notable custom silicon achievement to date. Given Nvidia’s commanding position in the AI hardware market, this contract carries substantial weight both in volume and strategic importance. It demonstrates Intel’s capability to secure high-profile clients for fully customized processor solutions.

Demand for custom silicon is expanding rapidly across multiple sectors. A decade ago, designing proprietary chips was feasible only for the largest corporations. Today, thanks to mature development tools, robust IP ecosystems, and improved foundry processes, companies of various sizes are exploring custom processors to gain competitive advantages.

Apple’s success with custom smartphone processors inspired many in consumer electronics, leading companies like Google, Huawei, and Xiaomi to develop their own SoCs. In the data center, hyperscalers such as Amazon and Google design their own AI accelerators and CPUs to tightly integrate hardware and software, achieving greater efficiency and cost savings. The trend is also strong in automotive, where manufacturers are investing in proprietary chips for software-defined vehicles, prioritizing long-term availability and performance.

Advanced electronic design automation tools from companies like Cadence, Synopsys, and Siemens have made custom chip development more accessible. When the potential return on investment is compelling—through reduced unit costs, lower total ownership expenses, or dramatic performance gains—companies are increasingly willing to consider in-house chip initiatives. However, the process remains risky and resource-intensive, leading many firms to seek out contract design partners instead.

Clients looking for custom silicon solutions have clear expectations from their design partners. Proven experience in delivering complex SoCs and ASICs is a top priority, especially within specific industry verticals like automotive, AI, or networking. Customers also expect access to a broad IP portfolio—including interfaces like PCIe, DDR, and Ethernet—without having to license each block individually. A mature, automated design flow is essential for meeting timelines and budgets, as are established relationships with foundries and assembly and test providers to ensure smooth mass production. Long-term support for yield improvement, validation, and lifecycle management is another critical factor.

Intel brings several strengths to the custom chip design table. The company possesses deep expertise in designing high-performance x86 processors for consumer and data center applications, along with GPUs, FPGAs, and AI accelerators. Its recent experience with multi-node, multi-foundry chiplet integration is a notable industry advantage. For clients needing high-performance x86 solutions, Intel’s architectural knowledge is formidable.

The company also maintains a substantial IP library, including x86 cores, GPU technologies, AI accelerators, and high-speed I/O controllers. For projects centered on x86 architecture, Intel offers a strong foundation. Additionally, Intel Foundry provides manufacturing and advanced packaging options like EMIB and Foveros, giving it an edge over design-only firms that rely on external partners.

There are, however, limitations. Intel has less experience integrating non-x86 architectures like Arm or RISC-V into customer designs compared to rivals such as Broadcom or MediaTek. While Intel can license third-party IP, its own IP for cutting-edge process nodes is still developing, which may require additional partnerships. Some customers may also perceive a potential conflict of interest, fearing that Intel could prioritize its own IP or manufacturing nodes over client preferences. Furthermore, Intel’s historical focus on high-volume production may need adjustment to meet the agile, lower-volume demands of contract design work.

Intel’s initial foray into custom silicon appears best suited for hyperscalers and infrastructure clients seeking high-performance x86 solutions manufactured at Intel Foundry. The Nvidia deal proves Intel can secure orders for highly customized CPUs, marking positive momentum. The company has also been building relationships with TSMC and EDA tool developers like Cadence and Synopsys, while partnering with third-party IP providers to broaden its offerings.

While inexperience with Arm and RISC-V remains a competitive drawback relative to general-purpose ASIC firms, it may not hinder Intel’s early growth in custom silicon. How far Intel’s ambitions in this space will ultimately reach remains to be seen, but these recent developments indicate a committed and structured approach to capturing a share of the expanding custom chip market.

(Source: Tom’s Hardware)