Arista’s Biggest Customers Are Also Its Biggest Risk

I like companies with important customers. I become considerably less relaxed when two of those customers account for nearly half the business.

That is the strange position I find myself in whenever I look at Arista Networks. The company has become one of the most important suppliers of high-performance networking equipment for hyperscale data centers and artificial-intelligence infrastructure. Its technology sits in the digital engine rooms of some of the richest and most technically demanding companies on Earth.

That sounds wonderful because it is wonderful.

It is also the problem.

According to Arista’s 2025 annual filing, two customers individually accounted for 26% and 16% of annual sales. Together, those two relationships represented 42% of the company’s revenue. The customers were widely identified as Microsoft and Meta, two technology giants spending aggressively to expand cloud and AI infrastructure.

If I were trying to invent the perfect customer list for a networking company, Microsoft and Meta would be near the top. They have enormous budgets, massive data centers, difficult technical requirements, and an urgent need to move increasingly ridiculous quantities of data at increasingly ridiculous speeds.

I would gladly accept their purchase orders.

Then I would wake up at three in the morning remembering that two customers were responsible for 42 cents of every revenue dollar.

That is the Arista paradox. Its biggest customers validate the company’s technology, accelerate its growth, and help demonstrate that Ethernet can compete inside advanced AI networks. At the same time, those customers possess enough purchasing power to influence Arista’s growth rate, product priorities, margins, and investor narrative.

Arista does not merely sell to giants. It depends on decisions made inside them.

Investors should celebrate the opportunity. They should also stop pretending concentration risk becomes harmless simply because the concentrated customers have excellent credit ratings.

A golden leash is still a leash.

Arista Earned Its Place in the Data Center

Before I spend the next several thousand words worrying about Arista’s customers, I should acknowledge why those customers buy from the company in the first place.

Arista is not some interchangeable box vendor that wandered into Microsoft’s procurement department carrying a glossy brochure. It earned its position by building reliable, high-performance networking systems around a strong software foundation.

The company’s Extensible Operating System, or EOS, gives customers a consistent software environment across its networking platforms. That consistency matters in enormous data centers where thousands of devices must be monitored, configured, automated, and upgraded without turning a routine change into a weekend-long emergency.

Hardware gets attention because people can photograph it. Software creates endurance because customers build operations around it.

Arista’s architecture appealed to cloud companies that needed scalability, automation, visibility, and reliability. Those customers were not interested in buying expensive networking equipment merely because a legacy vendor had enjoyed a long and comfortable history. They were building a different kind of infrastructure and wanted suppliers that understood it.

Arista did.

The company now describes itself as a leader in data-driven, client-to-cloud networking for large data center, AI, campus, and routing environments. That is corporate language, but the underlying position is real. Arista has expanded beyond its original cloud-networking stronghold into enterprise switching, campus networking, routing, observability, security, and AI fabrics.

Its financial performance shows that customers are not merely complimenting the products during conference presentations. They are buying them.

Arista reported first-quarter 2026 revenue of $2.709 billion, representing 35.1% year-over-year growth. It followed that with second-quarter revenue of approximately $3.036 billion, up 37.7% from the comparable period a year earlier. Those are not the growth rates of a company being politely tolerated by the market. They suggest Arista is participating directly in one of the largest infrastructure investment cycles in modern technology.

The company also raised its 2026 expectations after entering the year with an already ambitious outlook. Management projected roughly $11.25 billion in annual revenue and lifted its estimate for AI networking revenue to approximately $3.25 billion.

Naturally, this is when investors begin speaking as though demand will rise in a straight line until every available acre of land contains a data center.

I prefer to admire the numbers without becoming spiritually dependent on them.

Microsoft and Meta Are Not Ordinary Customers

Microsoft and Meta are attractive customers precisely because they are not ordinary.

Microsoft is expanding Azure, building infrastructure for its own AI products, and supporting an ecosystem of outside developers and businesses. Meta is investing heavily in the computing capacity needed for recommendation systems, advertising, generative AI, and its broader ambitions in advanced machine intelligence.

These companies consume networking capacity on a scale most enterprises cannot approach.

They also force suppliers to perform at a very high level. If Arista’s switches and software can satisfy hyperscalers operating some of the world’s largest networks, that becomes a powerful form of technical validation. A smaller potential customer does not have to rely entirely on Arista’s marketing department. It can observe that Microsoft and Meta have already subjected the technology to demanding real-world conditions.

That credibility has value.

Large customers can also create operational efficiency. Winning one enormous deployment may generate more revenue than signing hundreds of smaller accounts. Sales teams do not have to negotiate thousands of separate transactions to produce meaningful growth. Product development can focus on large, clearly defined technical challenges.

However, scale changes the balance of power.

A customer responsible for 1% of revenue is a customer. A customer responsible for 26% of revenue is practically a weather system.

If that customer delays an order, changes an architectural plan, slows capital expenditures, negotiates lower pricing, or shifts a deployment into another quarter, Arista feels it. Investors feel it. Analysts hold conference calls about it. Financial television discovers the customer’s name and begins using the word “concerns” before breakfast.

Microsoft and Meta may value Arista’s products, but neither exists to protect Arista’s shareholders. They will purchase what they need, when they need it, under terms they consider acceptable.

That is exactly what intelligent customers should do.

I simply do not want to confuse a productive commercial relationship with permanent loyalty.

The Numbers Make the Risk Difficult to Ignore

In 2024, Microsoft represented roughly 20% of Arista’s sales, while Meta represented approximately 15%. A year later, Arista’s annual filing showed its two largest customers contributing 26% and 16%.

The combined concentration therefore moved from roughly 35% to 42%.

That increase occurred during a period of strong overall growth, which is important. Arista was not shrinking around two remaining accounts. Its business was expanding while its biggest customers were spending heavily.

Still, 42% is 42%.

If one of those customers reduced purchases by half, the effect would not be a rounding error. Assuming everything else remained equal—which it never does—the revenue impact could reach several percentage points or even move into the double digits, depending on which customer pulled back.

The consequences would extend beyond revenue.

Large volumes can support production planning, inventory decisions, supplier commitments, and operating leverage. When those volumes fall unexpectedly, expenses do not necessarily disappear with them. Arista may have already invested in engineering, components, capacity, and support resources tied to anticipated demand.

Margins could come under pressure just as investors begin questioning the growth story.

This is one reason customer concentration deserves more attention than a quick mention beneath an otherwise cheerful revenue chart. It does not create only sales risk. It creates forecasting risk.

A business with thousands of similarly sized customers can often rely on aggregate behavior. Some customers spend more, others spend less, and the portfolio smooths out the variation.

A business with two whales has a different experience.

When one whale changes direction, the ocean notices.

Hyperscaler Spending Is Powerful but Uneven

The current enthusiasm surrounding AI infrastructure makes it tempting to assume Microsoft and Meta will continue increasing capital expenditures indefinitely.

They might spend enormous amounts for years. That does not mean every supplier’s revenue will rise smoothly every quarter.

Data-center construction occurs in phases. Customers plan facilities, secure power, install computing equipment, build network fabrics, test systems, and bring capacity online according to complicated schedules. Components do not arrive in a perfectly synchronized parade designed to make quarterly earnings comparisons attractive.

Orders can be lumpy.

A customer may purchase aggressively in one period and pause in another while it absorbs installed capacity. It may accelerate one project while delaying another because of permitting, energy constraints, hardware availability, or changes in technical priorities.

Even when the long-term demand trend remains strong, Arista’s quarterly results can reflect these timing shifts.

Investors often claim to understand this until a major customer’s spending moves from one quarter into the next. Then the stock market reacts as though the internet has been permanently canceled.

I do not consider temporary order variability a reason to avoid Arista. I consider it a reason to remain skeptical of overly precise forecasts.

The larger the customer, the more visible its timing becomes.

Meta demonstrated this dynamic in 2024. Its contribution to Arista’s revenue declined from approximately 21% in 2023 to 15% in 2024, even as Arista’s total annual revenue grew around 20% to approximately $7 billion. Sales associated with Meta reportedly declined, but growth from Microsoft and other customers more than compensated.

That was encouraging because it showed Arista could continue expanding while one large customer’s contribution fell.

It also demonstrated why investors monitor these accounts so obsessively. One hyperscaler’s purchasing pattern can meaningfully change the composition of an entire year.

The Customers Can Become Competitors

The risk is not limited to spending cycles.

Microsoft and Meta possess extraordinary engineering resources. They are not passive buyers waiting for suppliers to tell them what networking should look like. Hyperscalers help shape technical standards, design infrastructure architectures, develop software, and sometimes build custom hardware.

They have the scale to consider internal alternatives that ordinary enterprises could never justify.

If a networking component becomes strategically important or expensive enough, a hyperscaler may decide to design more of it internally. It may work directly with semiconductor suppliers and manufacturing partners. It may separate software from hardware, use white-box systems, or create custom equipment optimized for its specific environment.

This does not mean Microsoft or Meta will suddenly replace Arista across their networks. Designing hardware is not the same as operating a dependable networking platform at scale. Arista’s software, reliability, support, and accumulated expertise create genuine value.

Still, the possibility matters.

Arista is helping some of the most capable technology companies in the world build critical infrastructure. Those customers are learning constantly. They know their own requirements better than any outside supplier, and they have both the money and talent to explore alternatives.

The better the opportunity becomes, the more incentive they have to control it.

This is the awkward reality of selling strategic technology to hyperscalers. A major customer can be a partner, validator, negotiator, architect, and potential competitor at the same time.

That is a crowded relationship.

Bargaining Power Matters Even When Demand Is Strong

Customer concentration can limit pricing power.

When a buyer represents more than a quarter of annual revenue, contract negotiations are unlikely to resemble a casual conversation between equals. The customer knows its importance. Arista knows the customer knows.

Large buyers can demand volume discounts, customized features, favorable payment terms, dedicated engineering support, and aggressive delivery schedules. They may also encourage competition among suppliers to reduce dependence and improve negotiating leverage.

Arista can resist unreasonable demands because its products provide real value. Switching networking platforms is not always easy or desirable. Reliability matters, and hyperscalers do not casually replace technology supporting critical operations.

But the relationship still carries asymmetry.

If Arista loses Microsoft, Arista has a major financial event.

If Microsoft loses Arista, Microsoft has a procurement and engineering problem.

Both outcomes matter, but they do not matter equally.

This difference becomes especially important as competition intensifies. Cisco remains a formidable networking vendor. Nvidia is pushing deeper into AI networking with InfiniBand and Ethernet-based offerings. Broadcom supplies critical switching silicon. Other equipment manufacturers and custom-system providers want their share of hyperscale spending.

Large customers benefit from keeping multiple options available. Suppliers would prefer deep commitment. The resulting tension is perfectly normal, but it rarely appears in investor presentations decorated with arrows pointing upward.

AI Networking Creates a Bigger Opportunity—and Bigger Expectations

Arista’s growth opportunity is not merely traditional cloud networking with a fashionable label attached. AI clusters create demanding networking requirements because enormous numbers of accelerators must exchange data quickly and efficiently.

A powerful collection of GPUs becomes less useful if the network connecting them creates delays, congestion, or unpredictable performance.

This has created a major debate between InfiniBand, historically favored in high-performance computing and strongly associated with Nvidia, and Ethernet, the broadly adopted standard Arista knows exceptionally well.

Arista is betting that improved Ethernet technologies can capture a significant portion of AI back-end networking. The company participates in industry efforts such as the Ultra Ethernet Consortium and offers products designed for large AI fabrics.

If Ethernet gains share inside AI clusters, Arista could benefit from a market expanding in both size and strategic importance.

That is the optimistic case, and I find it credible.

The problem is that investors have already noticed.

Once a company becomes identified as a primary beneficiary of AI infrastructure spending, expectations can rise faster than the underlying business. Strong growth stops being impressive and becomes mandatory. Revenue beats are celebrated for several minutes before attention moves to next quarter’s guidance.

A concentrated customer base magnifies this pressure.

If Microsoft and Meta drive a large part of Arista’s AI growth, investors may value the company based on spending plans Arista does not control. Any suggestion that one hyperscaler is adjusting its architecture or moderating deployment growth could change the stock’s narrative long before the long-term market opportunity has disappeared.

Stocks do not react only to business performance. They react to the distance between performance and expectation.

Arista can remain an excellent company while its stock becomes vulnerable to a single sentence spoken during a customer’s earnings call.

That is not necessarily fair. Markets are under no obligation to be fair.

A Great Customer Can Hide a Diversification Problem

Rapid growth from major customers can make diversification look less urgent.

Why spend years building a broader customer base when Microsoft is increasing purchases by hundreds of millions of dollars? Why celebrate a collection of smaller enterprise wins when one hyperscaler deployment can move the entire income statement?

Because large customers do not stay predictable forever.

Diversification often looks inefficient during a boom. It becomes essential when the boom changes shape.

Arista has been expanding its enterprise business, campus products, routing capabilities, network observability, and security offerings. Its acquisition of VeloCloud added software-defined wide-area networking capabilities and created another path into enterprise infrastructure.

I see these efforts as more than optional side projects. They are part of the company’s long-term risk management.

Enterprise customers generally cannot replicate hyperscaler volumes individually, but a broad enterprise base can produce resilience. Thousands of organizations upgrading campus networks, data centers, routing systems, and network-management platforms create multiple sources of demand.

The margins, sales cycles, and competitive landscape may differ. Enterprise customers often require channel relationships, broader support, and different product packaging. Cisco is deeply entrenched in many of these accounts.

Diversification is not free.

But neither is dependence.

Arista does not need Microsoft and Meta to become smaller customers in absolute terms. Ideally, those relationships continue growing while the rest of the business grows faster. That would allow the company to preserve the benefits of hyperscale leadership while reducing the percentage of revenue tied to any single buyer.

I would consider that one of the healthiest possible developments, even if it made quarterly growth appear slightly less spectacular.

A company should not have to lose a major customer to become less dependent on it.

Customer Concentration Also Creates Strategic Focus

To be fair, concentration is not purely a weakness.

Working closely with advanced customers can accelerate innovation. Microsoft and Meta encounter networking challenges earlier than much of the broader market. Their requirements can push Arista to develop products and capabilities that later become useful to other customers.

In that sense, hyperscalers can function as demanding development partners.

They force Arista to solve problems involving scale, automation, power efficiency, congestion management, telemetry, and reliability. The knowledge gained from those deployments can strengthen the company’s overall product portfolio.

Large volumes can also justify research-and-development investments that benefit smaller accounts. Features originally demanded by massive cloud customers may eventually improve enterprise products.

This is one reason I would not want Arista to avoid customer concentration at any cost. Turning away business from Microsoft because Microsoft is becoming too important would be a strange form of risk management.

The goal is not to reject success.

The goal is to understand what kind of success the company has achieved and what vulnerabilities arrived with it.

Every business model contains trade-offs. Arista’s close relationships with hyperscalers provide scale, credibility, technical insight, and enormous revenue potential. In exchange, the company accepts purchasing concentration, negotiating pressure, demand volatility, and strategic exposure to a small number of corporate decisions.

That may be an excellent trade.

It is still a trade.

What Would Actually Worry Me?

I would not panic simply because Microsoft and Meta remain large customers. Their continued spending may support years of growth.

I would become more concerned if several warning signs appeared together.

First, I would watch for declining revenue from a major customer that could not be explained by deployment timing. A temporary pause is different from a structural shift toward internal designs or competing platforms.

Second, I would watch gross margins. Strong revenue growth accompanied by persistent margin deterioration could suggest that large customers are gaining too much pricing power or that competitive intensity is rising.

Third, I would monitor whether Arista’s AI networking growth remains concentrated in a limited number of deployments. A broadening customer base would validate the technology beyond its earliest hyperscale adopters.

Fourth, I would examine enterprise growth. If enterprise initiatives repeatedly fail to produce meaningful diversification, Arista may remain more dependent on cloud titans than its expanding product catalog suggests.

Finally, I would listen carefully to management’s language. Executives rarely announce a customer problem by saying, “We have a customer problem.” They discuss digestion periods, deployment transitions, architectural evaluations, normal variability, or revised timing.

Sometimes those phrases describe exactly what they claim to describe.

Sometimes they are corporate vocabulary for “please stop calculating the percentage.”

What Would Make Me More Confident?

The best outcome would be simultaneous strength in hyperscale, AI, and enterprise networking.

I want Microsoft and Meta to keep buying because their demand validates Arista’s performance at the highest level. I also want other cloud customers, AI providers, sovereign computing projects, research organizations, and large enterprises to adopt Arista’s platforms.

I would like to see customer concentration fall because the denominator is expanding, not because a major relationship is deteriorating.

Continued software development would also strengthen the case. Hardware products can face pricing pressure and architectural substitution. A deeply integrated software ecosystem increases switching costs and helps turn equipment sales into durable customer relationships.

EOS remains central to Arista’s appeal, but the company’s broader software capabilities—including automation, visibility, management, and security—could become increasingly important as networks grow more complicated.

The AI infrastructure market may be large enough to support multiple winners. Arista does not need to defeat every alternative. It needs to demonstrate that Ethernet-based AI networking can deliver competitive performance and that customers value Arista’s implementation.

If it accomplishes that across a widening collection of buyers, the concentration risk becomes more manageable.

It never disappears entirely. It simply stops dominating the conversation.

My View of Arista’s Risk-Reward Balance

I see Arista as a high-quality company with a genuine technological position in an attractive market.

Its growth is not imaginary. Its customers are not speculative startups promising to pay invoices after their next funding round. Microsoft and Meta have enormous financial resources and urgent infrastructure requirements. Arista has already demonstrated its ability to serve them at scale.

That deserves respect.

But I refuse to transform customer quality into an excuse for ignoring customer concentration.

The better the customer, the easier it becomes to rationalize dependence. Investors look at Microsoft and Meta and conclude that revenue tied to them is safer than revenue tied to weaker companies.

From a credit perspective, that is probably correct.

From a strategic perspective, it is incomplete.

Microsoft and Meta are unlikely to disappear. They are entirely capable of changing suppliers, changing architectures, shifting spending, delaying deployments, building internal alternatives, or extracting better terms.

Their financial strength makes them reliable payers. Their technical and commercial strength makes them powerful counterparties.

Arista benefits from standing close to the giants. It must be careful not to stand beneath them.

For investors, this means the company deserves neither blind enthusiasm nor automatic rejection. Customer concentration should affect the price I am willing to pay, the growth assumptions I consider reasonable, and the volatility I am prepared to tolerate.

A premium business can justify a premium valuation. A concentrated premium business still requires a margin of safety.

That distinction tends to disappear during periods of excitement.

The Bottom Line

Arista’s largest customers are evidence of its success.

They are also the clearest reminder that success can create new forms of dependence.

Microsoft and Meta helped Arista establish itself as a central supplier to the cloud and AI infrastructure buildout. Their spending supports revenue growth, product development, market credibility, and the broader case for Ethernet-based AI networking.

Yet those same relationships expose Arista to customer-specific spending cycles, negotiating leverage, architectural changes, internal hardware development, and investor expectations shaped by decisions made outside Arista’s headquarters.

I do not view that tension as a contradiction. I view it as the company’s defining investment question.

Can Arista continue benefiting from hyperscale demand while building enough breadth to prevent any two customers from determining its financial destiny?

So far, the answer is encouraging but unfinished.

The company is growing rapidly, expanding its product portfolio, pursuing enterprise opportunities, and positioning itself for the AI networking cycle. However, two customers representing 42% of annual sales is not a footnote I am willing to admire briefly and forget.

It is central to the story.

I want Arista to keep winning business from Microsoft and Meta. I simply want it to win even more business from everyone else.

Because when one customer accounts for 26% of sales, that customer is not merely sitting at the table. It has selected the restaurant, approved the menu, and knows perfectly well who will be paying if dinner gets canceled.

Arista has earned an enviable seat in the AI infrastructure economy.

Now it needs a larger table.