Intel

Intel is the only large chip company that both designs leading processors and manufactures them, and both halves are being rebuilt at once.

Intel Products designs and sells chips. The Client Computing Group sells Core and Core Ultra processors for desktops and laptops, plus chipsets, Wi-Fi, Thunderbolt and Arc graphics; it is the larger part of revenue. Data Center and AI sells Xeon server processors, Ethernet controllers and infrastructure processors, and is developing data center GPUs for AI inference after its Gaudi accelerators failed to sell. Both run on x86, the instruction set Intel created, which most PC and server software is built for.

Intel Foundry develops the manufacturing processes and runs the fabs. For years Intel led the industry in process technology, then fell behind TSMC with delays starting at 10nm. Its recovery plan was five nodes in four years, ending with Intel 18A, which entered high-volume production in 2025 in Core Ultra Series 3. It also offers its fabs and its advanced packaging to other companies, to win the volume that leading-edge fabs need to cover their cost.

Intel's newer products are built from tiles: several dies made on different processes, Intel's or TSMC's, joined in one package. That let Intel use TSMC for parts of recent chips while its own processes caught up.

Mobileye, which makes driver-assistance chips and software for carmakers, is the main business outside the three segments; Intel sold a majority of Altera, its FPGA business, in 2025.

In 2025 Intel cut staff and costs under a new CEO, Lip-Bu Tan, and agreed with NVIDIA to co-develop x86 products for PCs and data centers.

The x86 software base is the main reason. Windows, business applications, games, drivers and decades of company software run on x86, and Intel's chips run all of it unchanged. That keeps customers on x86, but it does not keep them on Intel: AMD's chips run the same software, so a PC maker or a cloud company can move between them with each new generation, and many have moved toward AMD in servers.

What ties customers to Intel specifically is the platform around the processor. Companies manage their laptop fleets through vPro, and their IT tools, security software and images are certified on it. Server buyers have validated their applications, firmware and operations on Xeon platforms over years, and some workloads use Xeon-only accelerators such as AMX and QAT. PC makers build most of their designs around Intel first, and Intel helps fund their marketing.

What they would move to: AMD for x86 in both PCs and servers, Qualcomm's Snapdragon X or Apple's Macs for laptops, and Arm chips from the clouds themselves. For Intel Foundry the question runs the other way: it has to earn customers away from TSMC before any of them can stay.

Owning the fabs shows in every line. Revenue was $52.9B in fiscal 2025, down 0.5%, after $79.0B at its peak in fiscal 2021; over the five years from fiscal 2020 it fell about 7.5% a year, as Intel lost share to AMD and Arm chips and AI budgets went to GPUs.

Gross margin was 34.8% in fiscal 2025, against 55.4% in fiscal 2021 and over 60% in fiscal 2016 to 2018. A fab's costs are largely fixed, so lower volume, new processes that are expensive while yields improve, and wafers bought from TSMC all pushed margin down. Operating margin was -4.2%, an operating loss of $2.2B, after a loss of $11.7B (-22.0%) in fiscal 2024 that included large charges for impairments and restructuring.

The capital needs are what set Intel apart from fabless rivals. Capital spending was $14.6B in fiscal 2025, down from $23.9B in fiscal 2024 and $25.8B in fiscal 2023, as Intel slowed fab construction. Operating cash flow was $9.7B, so free cash flow was -$4.9B; after $2.4B of equity pay it was -$7.4B. Free cash flow has been negative every year since fiscal 2022. Intel ended fiscal 2025 with $46.6B of debt and $14.3B of cash, about $9.2B more debt than cash and short-term investments.

Everything turns on Intel 18A and on finding foundry customers. Core Ultra Series 3 (Panther Lake) is the first 18A product, Xeons on 18A follow, and 18A is planned for several generations of Intel's own CPUs. If its yields and performance hold, Intel makes more of its own chips again instead of buying them from TSMC, which improves margins.

Intel 14A is the next node, designed for outside customers from the start. Intel has said it may pause 14A and later nodes if no significant outside customer commits, and would then rely more on TSMC for its leading products, a large change for a company built on making its own chips. It has slowed its Ohio fab and cancelled projects in Germany and Poland.

In products, Intel is pairing x86 CPUs with accelerators: data center GPUs for AI inference on a yearly cadence (Crescent Island first), custom chips designed for customers, and products co-developed with NVIDIA that combine Intel CPUs with NVIDIA's GPUs and NVLink. In PCs, Core Ultra chips with larger NPUs aim at on-device AI.

The company is also smaller and simpler than two years ago: fewer staff, fewer management layers, and Altera and other businesses partly sold.

• Winning foundry customers. Leading-edge fabs need more volume than Intel's own products provide. Without major outside customers, Intel may have to stop developing leading-edge processes, and the 10-K says so.
• Executing 18A on time and at good yields, after a decade in which process delays cost Intel its lead.
• AMD in x86. EPYC and Ryzen have taken a large part of server and PC sales, and AMD keeps releasing on TSMC's newest processes.
• AI spending going to GPUs. Data center budgets favour NVIDIA's and AMD's GPUs and custom chips; Intel's own AI accelerators have not sold, and its next ones are still in development.
• Arm in PCs and servers. Apple's Macs, Qualcomm's Snapdragon X and the clouds' own Arm chips each take work that once ran on Intel.
• The cost of fabs with revenue that has shrunk: years of capital spending above operating cash flow have added debt.