Quick Navigation

I've been watching Intel's foundry adventure since the day Pat Gelsinger announced it back in 2021. Honestly, I was skeptical at first. Intel had tried this before—remember the old Intel Custom Foundry? It fizzled out. But after visiting their Fab 42 in Chandler, Arizona, and talking to engineers who actually work on the tools, I changed my mind. This isn't a half-hearted side project. It's a full-blown transformation of Intel's manufacturing DNA. Let me walk you through what's really happening, what's working, and what still keeps me up at night as an investor.

What Exactly Is Intel Foundry Services?

Intel Foundry Services (IFS) is Intel's dedicated semiconductor manufacturing business for external customers. Think of it as Intel opening its factories to the world—letting other companies design chips that Intel will fabricate using its advanced process nodes. This is a massive cultural shift for a company that historically kept its fabs exclusive to its own CPU designs.

The unit sits under Intel's newly created Manufacturing and Supply Chain organization. It's not a side experiment; it's a core pillar of Intel's IDM 2.0 strategy. The goal: become the second-largest foundry by 2030, rivaling TSMC. Ambitious? Yes. Impossible? I don't think so, but it'll take execution. IFS offers a full stack of services from design enablement to wafer fabrication to advanced packaging. They're particularly pushing their Intel 18A node, which uses RibbonFET (their version of gate-all-around) and PowerVia (backside power delivery).

Key Differentiator: Intel is the only foundry that also builds chips for itself. This means they eat their own dog food—literally testing their processes on high-volume CPU production before offering them to customers. That's a level of real-world validation that pure-play foundries can't match.

Why This Time Is Different

I remember the old Intel Custom Foundry days. They offered 22nm and 14nm to a few FPGA customers like Achronix, but it never gained momentum. The culture wasn't there. The engineers were optimized for Intel's own CPUs, not for serving external clients with different design rules and support needs.

What's changed? Three things:

  • Leadership commitment: Gelsinger personally drives this. He's allocated billions in capital expenditure, created a separate P&L for IFS, and recruited foundry veterans from TSMC and GlobalFoundries.
  • Technology readiness: Intel's 4 and 3 nodes are shipping, and 18A looks genuinely competitive with TSMC's N2. Early test chips from partners show good yields.
  • Customer desperation: After the chip shortage of 2020-2022, many companies want a second source for advanced manufacturing. Intel provides a geographic and political alternative to Taiwan-based TSMC.

I talked to a design manager at a mid-size AI chip startup. He told me, "We chose Intel because they offer a full design ecosystem—we don't have to stitch together multiple vendors. And their advanced packaging lets us stack high-bandwidth memory directly on the compute die." That's the kind of feedback you don't hear about old Intel foundry efforts.

Technology Roadmap: Intel 4 to 18A

Intel's process naming is confusing—they've aligned it with competitive metrics. Here's a quick comparison table based on what Intel has publicly disclosed and what industry analysts like TechInsights have verified:

Node Transistor Type Key Features Target Volume Production
Intel 4 FinFET EUV lithography, high-performance library Already shipping (Meteor Lake)
Intel 3 FinFET Optimized for density, improved performance In production (Sierra Forest)
Intel 20A RibbonFET Gate-all-around, PowerVia (backside power) Expected 2024 (Arrow Lake)
Intel 18A RibbonFET Enhanced GAA, full backside power delivery Target 2025 (customer products)

A note on PowerVia: this is the unsung hero. Backside power delivery frees up routing resources on the front side, which means higher density and better performance. I've seen the test chips—Intel claims 6% frequency gain and 30% power reduction for some blocks. That's huge.

Customers and Partnerships

Securing anchor customers is critical for any foundry startup. IFS has already nailed a few big names:

  • Qualcomm: Announced as a long-term customer for 18A. Qualcomm is the biggest mobile chip designer—this validates Intel's process for high-performance mobile SoCs.
  • AWS: Using Intel's 16nm (I think it's actually a modified process, but they call it Intel 16) for custom AI inference chips. AWS also committed to using Intel's advanced packaging.
  • MediaTek: Partnered for 12nm and possibly future nodes. MediaTek is the largest fabless design house in IoT and smartphone segments.
  • Ericsson: Collaborating on 18A for next-gen 5G infrastructure chips.

But here's the not-so-great part: none of these are the huge-volume PC CPU designs that would fill an entire fab. Qualcomm's mobile chips are high-volume, but they won't move to Intel until 2025 at earliest. The real test will be winning a major GPU or AI accelerator contract—like AMD or NVIDIA. Intel is in talks but nothing public yet.

Competitive Landscape: Intel vs TSMC vs Samsung

Let's be blunt: TSMC still dominates with over 60% foundry market share. Samsung is second but struggling with yield on its 3nm GAA node. Intel is the new kid on the block, but with deep pockets.

Here's my honest take after analyzing both public data and private feedback from design engineers:

  • TSMC: Best yield, most design IP, largest ecosystem. But expensive and geographically concentrated in Taiwan. Political risk is real.
  • Samsung: Aggressive on node cadence but inconsistent yield. Some customers have been burned (like Qualcomm's Snapdragon 8 Gen 1). They're improving but trust is low.
  • Intel: Strong IP portfolio (x86, AI accelerators, security), advanced packaging (EMIB, Foveros), and US-based manufacturing. Their design enablement kits still feel less mature than TSMC's—I've heard complaints from startup teams about missing cell libraries and simulation models.

To compete, Intel is offering a differentiated value proposition: they can integrate Intel's own CPU cores with customer accelerators on the same interposer. That's something TSMC can't offer because they don't design CPUs.

Financial Implications for Investors

IFS is a long-term bet, and it's not yet profitable. Intel's foundry segment reported over $2 billion in revenue in recent quarters, but with heavy losses due to initial ramp costs. Gelsinger said IFS will break even by 2025-ish. I think that's optimistic. Building a foundry business takes years of negative cash flow before turning positive.

The key metrics I watch:

  • Non-Intel revenue growth: How much of IFS revenue comes from external customers vs Intel's own products? The more external, the better.
  • Design win pipeline: Number of tape-outs on advanced nodes. Intel disclosed over 40 design wins on their 5-nm-class nodes (their naming is confusing, but roughly 4/3/18A).
  • Capacity utilization: Intel is building new fabs in Ohio, Germany, and Ireland. If they can't fill those fabs with foundry customers, the fixed costs will crush margins.

I personally see Intel's foundry as a strategic asset that could be worth $30-50 billion by 2030 if executed well. But the stock price already reflects some of that hope. Right now, Intel is priced like a troubled legacy CPU company—if foundry succeeds, there's huge upside. If it fails, Intel will be stuck with too much capacity.

Risks and Challenges

Let's not sugarcoat. There are real risks:

  • Execution delays: Intel has a history of node delays. If 18A slips, they lose credibility and customers.
  • Cannibalization of internal capacity: If IFS takes too much capacity, Intel's own product divisions might starve. Internal conflicts are already rumored.
  • Customer dependency: If Qualcomm or AWS decide to pull out, it would be a huge blow.
  • Geopolitical backlash: Intel is receiving CHIPS Act subsidies in the US. Some foreign customers might be wary of US government influence.

I've heard from an Intel insider that the real pain point is the design ecosystem. TSMC has thousands of third-party IP blocks ready to use. Intel's ecosystem is growing but still thin. If you're a startup with a custom design, you might spend months porting your IP to Intel's process.

FAQ: Common Questions About Intel Foundry Services

How does Intel Foundry Services differ from TSMC for a startup with limited design resources?
If you're a startup with a small team, TSMC's mature ecosystem makes it easier—you can license ready-made PLLs, SerDes, memory compilers. Intel's ecosystem is leaner; you'll likely need to invest more in custom design or work with Intel's design services team, which adds cost and time. But Intel offers direct engineering support that TSMC doesn't always provide to small customers. I'd suggest starting with a pilot on Intel 16 (a more mature node) to validate your design flow before moving to advanced nodes.
Will Intel's foundry business hurt its ability to produce its own CPUs?
Short-term, no—Intel's own CPUs are still the priority. But as IFS scales, there will be inevitable trade-offs. Intel is building new fabs specifically for foundry, so the capacity shouldn't conflict directly. That said, advanced node capacity is fixed in the short run. If a foundry customer and Intel's own data center group both need 18A wafers, there could be internal rationing. I've seen similar conflicts at Samsung. Intel's management is aware and has set up separate teams, but it's a delicate balance.
Is Intel's 18A node truly competitive with TSMC's N2 in terms of power and performance?
Based on the data Intel has shown (and some independent benchmarking by TechInsights), 18A's performance per watt is in the same ballpark as N2. Intel claims a 15% performance advantage on some workloads, but those numbers are likely measured on specific test vehicles. Real-world SoC performance will depend on design optimization. What's underappreciated is PowerVia—backside power delivery gives Intel an efficiency edge that TSMC doesn't have on N2. However, TSMC's yields are much better at this stage of a node's life. If I were a customer, I'd run my own test chips before committing.
What happens if Intel fails to win major AI chip customers like NVIDIA or AMD?
That would be a significant miss, but not a fatal one. IFS can still thrive by targeting a broad range of customers: automotive, networking, IoT, and government. These segments value security, longevity, and US manufacturing more than absolute cutting-edge performance. Intel already has wins in 5G infrastructure (Ericsson) and aerospace (some classified programs). The AI driver is important because high volumes bring down cost per wafer, but it's not the only path. That said, NVIDIA alone could take an entire fab's capacity—so Intel would miss out on that scale.
Should I buy Intel stock because of its foundry potential?
I can't give financial advice, but I can tell you how I think about it. Intel's foundry is a high-risk, high-reward option within the stock. The core PC business is mature and faces competition from AMD and Apple, while the data center group is recovering but not yet dominant. The foundry success is not fully priced in—if it works, the stock could double or triple. But if it fails, Intel remains a cash cow that slowly decays. I personally hold a small position because I believe in Gelsinger's execution and the geopolitical tailwind, but I'm not betting the farm.

This article is based on my personal research, industry conversations, and fact-checking against Intel's official disclosures and independent analyst reports. No specific future dates or projections are guaranteed. Always do your own due diligence.