//php echo do_shortcode('[responsivevoice_button voice="US English Male" buttontext="Listen to Post"]') ?>

Intel’s second-quarter results offered the clearest evidence yet that its manufacturing recovery is gaining traction. Intel Foundry revenue reached $5.8 billion, up from $4.4 billion a year earlier, while its operating loss narrowed from $3.2 billion to $2.1 billion. The segment’s operating margin improved from negative 71.7% to negative 36.2%. Intel attributed the progress to better yields, shorter cycle times, and increased fab scale.

Yet the quarter also highlighted the central uncertainty surrounding Intel Foundry: Most of the improvement came from manufacturing Intel’s own products. Only $293 million of Foundry’s quarterly revenue came from external customers, according to CFO David Zinsner.

Intel is certainly demonstrating that it can run its fabs more effectively. But it has yet to prove that it can turn those fabs into a large merchant-foundry business serving outside chip companies.

The newly announced collaboration with Fortinet provides an early test. Intel confirmed to EE Times that the agreement is more substantial than a preliminary technology evaluation, although its commercial scale remains undisclosed.

Fortinet becomes an Intel Foundry customer

Fortinet plans to use the Intel 4 process for SP6, its next-generation security processor, an Intel spokesperson told EE Times in an exclusive exchange. Fortinet is leading the architecture and front-end design, while Intel will provide back-end design services and manufacture the chip.

“Intel is the exclusive manufacturing partner for SP6,” the spokesperson said, while declining to identify the fab, packaging facility, or geographic location that will support the product. Intel was also not willing to disclose the packaging technology, production timetable, committed volumes, contract duration, or financial terms.

The spokesperson nevertheless described Fortinet as a commercial Intel Foundry customer rather than a development partner merely exploring the process, telling EE Times: “Fortinet is an Intel Foundry customer. They are using Intel’s custom ASIC capabilities built for Intel 4 to develop and produce a solution tailored to their requirements.”

That makes Fortinet a genuine design-and-manufacturing win. But because Intel has not disclosed when SP6 will tape out, enter qualification, or reach volume production, the agreement cannot yet be measured against Intel Foundry’s financial performance.

It also does not directly validate Intel’s latest external-foundry roadmap. Intel 4 is already an established production process used for Intel products. The more consequential test will be whether major external customers commit leading-edge designs to Intel 18A-P and, eventually, Intel 14A.

Intel CEO Lip-Bu Tan said during the earnings call that 18A-P had entered risk production and that Intel was seeing positive engagement around 14A. He cited progress in defect density, transistor performance, process-design kits, and intellectual-property development as reasons for committing to 14A high-volume manufacturing in 2028.

When analysts asked when that confidence would translate into named customers, however, Tan focused on technical milestones and customer discussions rather than announcing a major external production commitment.

A different model from TSMC

Fortinet may reveal the type of external customer Intel can attract first. It is not simply a fabless semiconductor company handing Intel a completed design and buying wafers. Fortinet controls the architecture and front-end design, while Intel provides back-end design and manufacturing. The relationship combines foundry services with custom-ASIC development.

Tan positioned Fortinet that way during the call, describing it as “a strong step forward” in Intel’s ASIC strategy. He argued that Intel can combine processor intellectual property, design capabilities, wafer manufacturing, and advanced packaging to build purpose-specific products for customers.

Zinsner said Intel’s purpose-built silicon business was approaching a $2 billion annual run rate and could reach approximately $4 billion in the “not too distant future.”

That suggests Intel may initially compete as much with Broadcom, Marvell, and other custom-silicon providers as it does with TSMC. Systems companies in networking, cybersecurity, cloud infrastructure, automotive, and industrial markets may want custom chips but lack the internal design resources of Apple, Nvidia, or AMD.

Intel can offer those companies a broader service than wafer fabrication alone.

Stephen Sopko, an analyst-in-residence at HyperFRAME Research, said Fortinet may become a test of whether Intel can adapt its organization around an outside customer.

“What you’re seeing with Fortinet is a proof point of an Intel that is willing to listen to a customer and then use a node to deliver something that customer wants,” Sopko told EE Times in an exclusive interview.

For Intel, that customer responsiveness may be as important as transistor performance. Its manufacturing organization was historically optimized around Intel’s own designs, schedules, and priorities. A merchant foundry must repeatedly adapt its processes, tools, and support organization to the needs of customers that Intel does not control.

Higher spending, but for which customers?

Intel is also increasing capital expenditure. The company now expects to spend more than $20 billion in 2026, with investment rising further in 2027. Much of that spending will go toward manufacturing tools, front-end capacity, advanced packaging, substrates, and memory supply.

Management repeatedly said it was aligning spending with customer demand and avoiding large investments ahead of commitments. Zinsner said the increase should be read as a sign of Intel’s confidence in future demand.

“We’re going to be very careful around making bets ahead of customer commitments,” he said. “And you can read that inversely. Given our confidence around next year, we must have pretty significant confidence in our customers, or we wouldn’t be putting POs in place today.”

But Intel did not separate spending driven by external foundry customers from spending required to meet demand for Intel’s own processors.

That distinction matters because Intel described severe internal supply constraints. The company is expanding Intel 3 capacity to produce more Granite Rapids server processors, while 18A output is growing to support new client and server products. Zinsner said 18A output was more than 50% higher sequentially and approximately 25% above target. Intel also reduced the cost of its main Panther Lake product by roughly 50% during the first half of the year.

Those achievements support Intel Products as much as they support Intel Foundry. Higher internal demand can fill fabs, improve utilization, raise yields, and reduce wafer costs without proving that external customers are choosing Intel over other foundries.

Optionality before displacement

Sopko cautioned against viewing Intel Foundry as an immediate threat to TSMC.

“I don’t see Intel necessarily as a threat to TSMC,” he said. “TSMC is so overcommitted. They’re running flat out to meet the demand. Somebody else being able to produce chips and package them is a net positive. It’s not necessarily a competitive story between TSMC and Intel yet.”

Intel’s early opportunity may therefore come from giving customers supply chain optionality rather than persuading them to abandon TSMC entirely.

A customer could qualify Intel as a second source, place selected products on an Intel process, or use Intel’s advanced packaging while continuing to manufacture other dies at TSMC. Intel said customer interest in EMIB-T packaging remained high and that it was preparing for customer ramps in 2027.

That approach could be especially attractive as leading-edge wafer and advanced packaging capacity remain constrained. It also allows customers to diversify production geographically without transferring an entire flagship product family.

But optionality is only an intermediate milestone. Intel Foundry will have achieved something more significant when an outside customer trusts Intel enough to make it the primary manufacturing source.

Sopko said that trust has two dimensions. Customers must trust Intel to protect their intellectual property, and they must believe Intel will honor external commitments when capacity and manufacturing inputs are tight.

“Customers have to feel like their products won’t go to the back of the line when Intel is being pressured,” he said.

That concern goes to the heart of Intel’s structure. Unlike TSMC, Intel owns product businesses that compete for the same fabs, engineering resources, and capital. Intel Foundry now reports its results separately, but it remains part of Intel, with Intel Products as its dominant customer.

For now, the second-quarter results show that Intel’s manufacturing organization is executing better. Fortinet shows that Intel can attract at least one commercial external customer to Intel 4 and combine design services with production.

The larger breakthrough will come when customers choose Intel not merely as an alternative, but as their first manufacturing choice.

“What I need to see,” Sopko said, “is somebody choosing Intel as the primary, with another fab providing optionality.”

---

Read also:

CEA-Leti Looks Beyond SRAM and DRAM as AI Reshapes the Memory Roadmap

AI Data Centers Push Silicon Photonics Toward 300-mm Scale

After Magdeburg, Intel Builds on Ireland’s Existing Strength

CHIPLETS, SEMICONDUCTORS, SOC

INTEL FOUNDRY, TSMC