How Customer Concentration Undermines a Wide Moat

Ninety-one percent. That's the share of Cirrus Logic's net sales that traced back to one end customer in fiscal year 2026, according to the company's own 10-K. Not a typo. The year before, it was 89%. The year before that, 87% — three straight years, one direction, the same company on the other end of nearly every dollar of revenue the business brought in.

By the standard tests moat analysis usually applies, Cirrus Logic looks like a genuinely strong business. Its mixed-signal audio and power-management chips are embedded deep inside a product architecture that takes years to requalify a new supplier into, and the engineering relationship runs far past a simple purchase order. Ask the question this beat always starts with — what would it take to displace them — and the honest answer, from a rival chipmaker's seat, is: quite a lot. That's a real durable competitive advantage, not a marketing claim.

But there's a version of this risk that moat analysis, on its own, doesn't answer. A moat is built to survive a competitor trying to take the business. It says nothing about what happens when the one customer on the other side of that switching cost decides, for its own reasons, to stop needing you at all. I've come around to thinking that's one of the more underappreciated blind spots in how investors apply the framework — we spend most of our energy asking whether rivals can take the business, and comparatively little asking whether the customer can just take it back. Cirrus Logic is the live example. A company called Dialog Semiconductor is the cautionary one, and it already happened once, in public, in 2018.

What a Moat Is Actually Built to Withstand

Moat durability, as this framework defines it, is a claim about competitive replication. Switching costs, network effects, a license a regulator won't hand out twice, a cost structure nobody else can match — each one answers a version of the same question: can a rival realistically take this business away? When the answer is no, or not without years of effort and capital a competitor probably won't commit, that's a durable competitive position worth paying attention to.

Customer concentration is a different axis entirely, and it doesn't show up on that scorecard at all. A company can score well on every conventional moat test — high switching costs, genuine intellectual property, a multi-year design-in cycle a competitor can't shortcut — and still be structurally fragile, because the protection those switching costs provide only works in one direction. They keep competitors out. They do nothing to stop the customer itself from walking through a door that was never locked against them in the first place.

Cirrus Logic's Moat Is Real — That's What Makes the Concentration Worth a Second Look

It's worth being precise about what kind of company this is, because the point here isn't that Cirrus Logic is a weak business dressed up as a strong one. The audio and power-management chips it designs for smartphones require years of joint engineering work with the device maker before a single unit ships — firmware, package design, and board layout that get built around Cirrus Logic's specific parts, not swapped out like a commodity component. That's exactly the kind of switching cost this beat looks for.

Why a Competitor Can't Just Walk In

A rival chipmaker — Qualcomm, Texas Instruments, a scrappier mixed-signal shop — would need to win a design slot two to three product cycles in advance, match performance and power specs that took Cirrus Logic years to tune, and convince a customer to absorb the qualification risk of ripping out an incumbent supplier mid-architecture. That's not a theoretical barrier. It's the reason Cirrus Logic has held its position inside the same customer's products for well over a decade, through several full silicon redesigns. Against a competitor, in other words, the moat holds.

Three Years, Three Numbers

The concentration itself isn't a secret — it's disclosed, plainly, in the risk-factors section of the 10-K every year, because it has to be. Apple represented approximately 87% of Cirrus Logic's total net sales in fiscal 2024, 89% in fiscal 2025, and 91% in fiscal 2026. The trend line matters as much as any single year: this isn't a concentration that's easing as the business diversifies. It's tightening. Every year, not just this one.

I'm not fully sure how much of that increase reflects Cirrus Logic winning a larger share of content per device versus the rest of its business simply growing more slowly — the disclosures don't break the mix down that cleanly, and I'd rather flag the ambiguity than pretend the filings answer it. What the numbers do make unambiguous is the dependency itself, regardless of which force is driving it.

When the Customer Becomes the Competitor

Cirrus Logic isn't the first chip company to look unassailable against competitors while carrying this exact structural risk. Dialog Semiconductor ran the same profile a decade ago, on the same customer, in a different chip category — and the way it played out is the clearest illustration available of why the two risks need to be judged separately.

Dialog Semiconductor's Warning

Dialog built its business around power-management chips for the iPhone, with Apple accounting for nearly three-quarters of its revenue by 2016. Through most of 2018, Dialog's market value was cut roughly in half as investors priced in a growing fear: that Apple was quietly developing its own power-management silicon and would eventually stop needing an outside supplier for the part Dialog had spent a decade perfecting.

The fear turned out to be accurate, just not in the form the market had braced for. In October 2018, Apple and Dialog announced a $600 million transaction — roughly half paid in cash, half as a prepayment for future parts — under which Apple licensed a large portion of Dialog's power-management intellectual property and hired around 300 of its engineers directly. Dialog's own guidance afterward pointed to a real, multi-year decline in the revenue tied to that relationship, even as the stock itself jumped as much as a third in a single trading session on relief that the terms weren't worse.

That gap — a business fundamentally getting smaller against the same customer that had made it look unassailable, alongside a stock market reaction that looked like good news — is exactly the distinction this piece is arguing for. Dialog's switching-cost moat, the one built from years of embedded engineering, never had to face a competitor to fail. So it didn't fail the way a moat usually fails. It failed because the party the moat was supposedly protecting Dialog from turned out to be the one holding all the leverage the whole time. Dialog no longer exists as an independent company: Renesas Electronics acquired it in August 2021, a little less than three years after the Apple transaction had already reshaped its largest customer relationship.

A Wide Moat Without the Concentration Risk

The contrast worth holding onto is a business with real switching costs and genuine diversification at the same time — proof the two aren't in tension with each other. ADP's payroll and HR outsourcing business is a clean example: the operational cost of switching a company's payroll, tax withholding, and compliance infrastructure to a new provider is high enough that clients rarely do it once they're set up, the same switching-cost mechanism that protects Cirrus Logic from a rival chipmaker. ADP, though, serves more than 740,000 clients, and no single client or group of affiliated clients accounts for more than 2% of its annual revenue, according to its own 10-K disclosure.

That's the version of a wide moat worth wanting: real protection against competitors, and no single counterparty capable of unwinding the business on a decision made in someone else's boardroom. Cirrus Logic's protection against competitors is every bit as real as ADP's. Its exposure to one customer's own strategic choices just isn't offset by anything comparable — and that's the piece a moat rating alone won't show you.

Reading and Weighing the Concentration Risk

How to Read It in the 10-K

The disclosure itself is usually easy to find. Item 1A risk factors will typically include language naming a customer's share of revenue directly, or describing dependence on a limited number of customers when the company would rather not name names. Item 7's MD&A section often repeats the concentration figure alongside the year-over-year sales bridge, which is the more useful place to watch the trend rather than a single year's snapshot.

A few questions are worth running through whenever that disclosure shows up:

• Is the concentration disclosed as a percentage, and is that percentage rising, falling, or flat across the last three fiscal years?
• Does the filing name the customer directly, or use vaguer language like "a limited number of customers" — vaguer language is sometimes a sign there's more than one relationship at risk, not fewer?
• Is the switching cost that protects the company from competitors the same mechanism that could let the customer walk away, or a different one entirely?
• Has the customer shown any history of insourcing similar capabilities elsewhere in its supply chain?

That last question is doing more work than it looks like. Read enough concentration footnotes and the language starts to look nearly interchangeable company to company — but a customer with a track record of insourcing critical suppliers, the way Apple has more than once, is a materially different risk than a customer that has never shown any inclination to build the capability itself.

Where It Belongs in the Moat Judgment

None of this is an argument for treating customer concentration as a reason to write off an otherwise strong competitive position. Genuine switching-cost moats are rare and worth paying for; Cirrus Logic's is real, and Dialog's was real too, right up until it wasn't relevant anymore. A moat judgment that only asks whether competitors can take the business, though, is answering half the question. The other half — whether the specific customer behind the concentration has ever shown a taste for insourcing, and whether the switching cost would actually survive that customer's own decision to leave — belongs in the same analysis, much like the ongoing warning signs worth tracking every year, not filed away once and forgotten. This isn't a quirk unique to chip suppliers, either. The same dynamic shows up anywhere a switching-cost story runs through one large customer instead of thousands of smaller ones — contract manufacturers tied to a single device maker, software vendors embedded deep in one enterprise's stack, franchisees dependent on a single franchisor's renewal decision. The mechanism that protects a business from competitors and the mechanism that exposes it to one counterparty's own choices can coexist inside the same company, and usually do, whenever concentration runs high enough to show up in a 10-K at all.

Key Takeaways

• A moat measures resistance to competitors, not resistance to a customer's own strategic choices — the two risks require separate judgments.
• Cirrus Logic's switching-cost moat against rival chipmakers is real: Apple represented 87%, 89%, and 91% of its net sales in fiscal 2024, 2025, and 2026, respectively, and the relationship has held through years of design cycles.
• Dialog Semiconductor shows the downside case: a similarly strong engineering relationship with Apple, roughly three-quarters of revenue in 2016, didn't prevent Apple from licensing its core IP and hiring its engineers directly in a 2018 transaction that permanently shrank Dialog's business with its largest customer.
• ADP shows the version without this risk: real switching costs, more than 740,000 clients, and no single one above 2% of revenue.
• When reading a concentration disclosure, weigh not just the percentage but whether the customer has any history of insourcing — that history is a better predictor of this specific risk than the switching cost itself.