To The Moon

For most of the AI boom, investors focused obsessively on who makes the chips and who rents the cloud capacity. I think that framing is already becoming outdated. The real constraint inside modern AI infrastructure is no longer raw compute power alone; it is the speed and efficiency with which thousands of GPUs communicate with one another.

That shift matters because idle GPUs are financial vandalism. Hyperscalers are spending tens of billions building AI clusters, but if the networking layer cannot move data efficiently between processors, expensive compute hardware simply sits there underutilised. In practical terms, networking has evolved from a supporting technology into one of the central determinants of AI economics.

That is why Arista Networks has quietly become one of the most strategically important companies in the entire AI infrastructure cycle.

The real AI bottleneck is no longer compute power alone

The Revolt Against Vendor Lock-In

One of the least discussed developments inside hyperscale infrastructure is the gradual move away from proprietary networking architectures towards open Ethernet standards.

For years, Nvidia’s InfiniBand dominated high-performance AI networking because it delivered ultra-low latency communication between GPUs. But hyperscalers have become increasingly uncomfortable with the idea of one vendor controlling both the compute layer and the networking fabric beneath it.

That is where Arista’s opportunity becomes enormous.

The Ultra Ethernet Consortium is not merely a technical initiative. It is an economic and strategic pushback against vendor lock-in. $Meta Platforms, Inc.(META)$, $Microsoft(MSFT)$ and other hyperscalers want flexibility, bargaining leverage, and lower long-term infrastructure costs. Arista happens to be exceptionally well positioned to benefit from that transition.

Most investors still view Arista as a networking hardware company, which misses the bigger picture. As AI clusters scale, networking software, orchestration and traffic optimisation are becoming critical determinants of overall system efficiency.

If Nvidia is building the engines of AI, Arista is helping determine how efficiently the entire transport system functions.

Margins That Look More Like Software

The strongest evidence for this structural shift is hiding inside Arista’s margins.

Traditionally, hardware businesses face relentless margin compression because physical products eventually become commoditised. Arista is moving in the opposite direction. Its operating margin sits near 43%, while net margins exceed 38%, levels usually associated with elite software firms rather than networking equipment manufacturers.

I think this reveals something deeper about where value is migrating inside AI infrastructure.

The market still assumes compute hardware captures most of the economic upside, but AI clusters are becoming more dependent on orchestration, latency management and traffic optimisation. As clusters scale toward tens of thousands of GPUs, networking intelligence becomes a far larger determinant of overall system efficiency.

That dynamic gives Arista unusual pricing leverage.

Hyperscalers are not buying switches merely as hardware components; they are buying efficiency gains across multi-billion-dollar AI systems. If Arista improves utilisation rates even marginally, customers can justify enormous spending because the alternative is underused compute infrastructure costing far more.

This partly explains why Arista’s financial profile resembles software economics despite selling physical systems. The company generated $9.71 billion in trailing revenue and converted that into $3.72 billion in net income, while producing $4.36 billion in levered free cash flow.

Even more striking is the balance sheet. Arista holds $12.35 billion in cash and carries effectively no debt, meaning its enterprise value is actually lower than its market capitalisation. In a sector where many AI infrastructure firms are burning capital at industrial scale, Arista looks almost absurdly disciplined.

Investors may still be underestimating how durable that advantage could become.

The market keeps repricing Arista as infrastructure, not merely hardware

The Networking Power Struggle

The competitive landscape around AI networking is becoming unusually fascinating because each major player approaches the market from a completely different angle.

$Cisco(CSCO)$ still dominates traditional enterprise networking, but hyperscale AI infrastructure rewards programmability, automation and cloud-native architectures rather than legacy installed base advantages. Arista’s EOS software platform has become particularly attractive because it was designed around large-scale cloud environments from the beginning.

Nvidia remains the most formidable competitive threat through InfiniBand and its expanding full-stack AI ecosystem. Yet Nvidia’s strength may also accelerate demand for alternatives. Hyperscalers are unlikely to feel comfortable allowing one company to dominate chips, networking and AI software simultaneously.

That strategic tension could become one of Arista’s greatest advantages.

There is also an underappreciated cultural difference between the companies. $NVIDIA(NVDA)$ behaves like a vertically integrated AI empire, while Arista positions itself as a more flexible infrastructure partner. Large cloud operators tend to value optionality highly, especially when annual AI spending is measured in hundreds of billions rather than millions.

The Valuation Problem

The valuation debate is where the investment case becomes genuinely difficult.

After compounding more than 640% over five years, Arista now trades at roughly 53 times trailing earnings and more than 20 times sales. That combination tells investors something important: the market is no longer pricing $Arista Networks(ANET)$ as a fast-growing networking company. It is pricing the company as critical AI infrastructure that must continue executing almost flawlessly.

That creates very little room for disappointment.

Exceptional businesses rarely look cheap during structural infrastructure cycles

This is not an overlooked bargain hiding in plain sight. The market clearly recognises the quality of the business. The more important question is whether consensus expectations still underestimate the scale of the networking opportunity itself.

Historically, networking represented a relatively modest percentage of overall data centre spending. AI changes that equation because distributed computing efficiency becomes disproportionately more important as clusters grow larger. Networking is no longer peripheral plumbing; it increasingly determines whether hyperscalers achieve acceptable returns on massive AI capital expenditure.

Because once infrastructure becomes essential to utilisation efficiency, customers stop evaluating it as a cost centre and start treating it as a productivity multiplier.

If networking spend expands as a percentage of total AI infrastructure budgets over the next five years, Arista’s current valuation may prove less aggressive than it appears today.

However, investors should not ignore the operational risks embedded inside those expectations.

Revenue concentration remains significant, particularly with hyperscalers such as Meta and Microsoft. Any slowdown in AI capex could compress both growth rates and valuation multiples simultaneously. Nvidia also remains a formidable competitive threat through InfiniBand and its broader ecosystem strategy.

AI investors are famously enthusiastic right up until the moment they suddenly remember valuations exist. Markets can be terribly inconsiderate like that.

Infrastructure becomes priceless once efficiency determines profitability

Final Verdict

I think Arista Networks represents something unusual in this cycle: a company benefiting not simply from AI enthusiasm, but from a structural redesign of how hyperscale computing itself operates.

The market initially treated networking as a secondary beneficiary of the AI boom. Today, it looks more like one of the core control points determining whether hyperscalers can scale AI profitably at all.

Arista’s combination of software-like margins, a fortress balance sheet, and deep hyperscaler integration suggests the company may be evolving beyond traditional networking hardware into something far more strategically valuable.

The irony is that while investors spent years obsessing over the silicon, one of the most profitable positions in AI may ultimately belong to the company making sure the machines can actually talk to each other properly.

@TigerStars @Daily_Discussion @Tiger_comments @Tiger_SG @Tiger_Earnings @TigerClub @TigerWire