A Foundry Business Still Bleeding Red
Intel’s foundry division is losing money at a pace that would unsettle even the most patient investor. The unit, which Intel has been building out as a contract chipmaking business to rival Taiwan Semiconductor Manufacturing Company, posted an operating loss of $2.3 billion in the first quarter of 2025 alone – continuing a streak of quarterly losses that stretch back years and show no clear floor yet. The ambition behind Intel Foundry is genuine: become the Western world’s answer to TSMC, reduce reliance on Asian chip supply chains, and capture a slice of the most lucrative manufacturing contracts in technology. The reality has been far more painful.
The core problem is not simply that Intel is spending heavily to build capacity. It is that the clients who would make that capacity worthwhile are not moving. TSMC’s major customers – Apple, Nvidia, AMD, Qualcomm, Broadcom – remain firmly planted in Taiwan, and there is no public signal from any of them that Intel Foundry is about to become their next manufacturing home. For Intel CEO Pat Gelsinger’s replacement, Lip-Bu Tan, who took over in early 2025, the inherited challenge is less about cutting costs and more about convincing a skeptical industry that Intel can actually deliver.
Why TSMC Customers Are Not Moving
Switching foundries is not a simple procurement decision. For a company like Apple or Nvidia, moving even a portion of production requires years of qualification work, new tooling, updated design rules, and intensive testing. TSMC has spent decades fine-tuning its manufacturing processes, and its customers have built their chip architectures around those specific capabilities. The switching cost is enormous – not just financially, but in engineering hours and competitive risk. A chip that underperforms because of a manufacturing hiccup can cost hundreds of millions in lost product cycles.
Intel Foundry has struggled to demonstrate that it can match TSMC’s process consistency. Early versions of Intel’s 18A node – its most advanced manufacturing process and the one it hoped would attract premium customers – faced yield concerns that delayed timelines and raised questions about whether the technology was ready for high-volume production. Yield, the percentage of chips on a wafer that actually work correctly, is the metric foundry customers watch above almost everything else. A process with inconsistent yields is not just more expensive; it is unpredictable, and unpredictability is the one thing chip designers cannot tolerate in high-stakes product launches.
There is also the matter of trust built over time. TSMC has never competed with its customers by designing its own chips. Intel, by contrast, still operates one of the world’s largest chip design businesses. For a company like Qualcomm or MediaTek – direct Intel competitors in certain markets – handing over their most sensitive chip designs to an Intel-run fab requires a level of trust in operational separation that Intel has not yet convincingly established. Intel has moved to create structural separation between its foundry and product businesses, but skepticism remains.
The customers who have publicly committed to Intel Foundry are notable but limited in scale. Microsoft has signed on for 18A production, and Amazon Web Services has discussed using Intel manufacturing for certain chips. These are real relationships, and Intel has highlighted them heavily in investor communications. But they do not yet represent the volume or the margin profile that Intel Foundry needs to approach breakeven, let alone profitability.
The Financial Drain and What It Means
The losses inside Intel Foundry are structural rather than incidental. Building leading-edge semiconductor fabs requires investments measured in tens of billions of dollars, and those costs hit the income statement long before revenue arrives to offset them. Intel has been spending at that scale for several years while simultaneously trying to catch up to TSMC on process technology – a gap that existed before the current buildout even began.
What makes Intel’s position particularly complicated is that its legacy chip business – the PC and server processors that have funded everything for decades – is also under pressure. AMD has taken meaningful market share in the data center processor market, and while Intel still holds a large portion of the PC chip market, that segment does not generate the growth needed to subsidize an open-ended foundry buildout. The company is effectively fighting a two-front battle: defend the core business while building an entirely new one from scratch.
What Has to Change for This to Work
Intel Foundry’s path to relevance runs almost entirely through 18A. If that process node can be shown to perform reliably at volume – and if the chips it produces are competitive with what TSMC’s N2 node delivers – there is a real argument that some customers will diversify. Geopolitical pressure to build non-Taiwan chipmaking capacity is real, and U.S. government support through the CHIPS Act has already directed billions toward Intel’s American fab expansion. That funding does not solve the technology problem, but it meaningfully reduces the capital burden.
Lip-Bu Tan has signaled a more focused approach than his predecessor, with reports indicating he wants to streamline the business and concentrate engineering resources on the processes most likely to attract external customers. Whether that involves narrowing the number of nodes Intel pursues simultaneously, or cutting headcount further to reduce overhead while fabs ramp up, the direction is toward discipline over scale. The question the industry is watching is whether any of that repositioning changes the calculation for TSMC’s largest clients – the companies whose contracts would actually move the needle.
The Longer Game Intel Is Playing
There is a version of this story where Intel’s bet eventually pays off – where geopolitical tensions between Washington and Beijing accelerate the push to diversify chip supply chains away from Taiwan, where 18A proves itself in volume production, and where a handful of major customers begin splitting orders between TSMC and Intel Foundry as a hedge. That version is not impossible. It simply requires things to go right on multiple fronts simultaneously, and Intel has not had many of those stretches recently.
The CHIPS Act subsidies give Intel financial runway that a purely private-sector competitor would not have. But government support is not a substitute for customer confidence. The fab in Ohio that Intel has been building in phases, and the Arizona expansions, represent real physical infrastructure that will exist regardless of near-term business performance. The harder asset to build is a track record – consistent yields, on-time delivery, and the kind of process reliability that convinces an Apple or Nvidia supply chain team to put a next-generation product on Intel silicon.
Right now, none of TSMC’s biggest customers have given Intel Foundry that kind of vote of confidence. Microsoft’s commitment is meaningful, but Microsoft is not designing the chips that define where semiconductor manufacturing trends go. Apple’s A-series and M-series chips, Nvidia’s Blackwell and whatever comes after it – those are the contracts that would signal Intel Foundry has arrived. Until one of those names appears on Intel’s customer roster in a serious production context, the losses will keep accumulating against a promise that has yet to be redeemed.
