To The Moon

The AI Toll Collector Nobody Expected

Most investors still think $Advanced Micro Devices(AMD)$ is fighting Nvidia for the AI throne. I increasingly think AMD may profit even if it never wins the crown.

That is what makes the stock so dangerous at today’s valuation.

At nearly $700 billion in market value and more than 140 times trailing earnings, AMD is no longer priced like a challenger. It is priced like a future AI superpower. Yet unlike Nvidia — whose software ecosystem behaves less like a product suite and more like a fully electrified railway network that developers are already locked into — AMD is still racing to prove its moat can widen fast enough to justify the premium investors have assigned it.

The paradox is brutal. AMD may not need to beat Nvidia to become a vastly larger business. But it may need to beat $NVIDIA(NVDA)$ enough to justify the stock price investors are paying today.

That distinction is where the real analytical depth lives.

AMD may profit most from roads Nvidia accidentally built

The Ratio Revolution Hiding Beneath AI

Most AI commentary still treats the industry as a GPU arms race. I think the more important shift is happening quietly beside the GPUs themselves.

During the first wave of generative AI, infrastructure was typically built around roughly one CPU supporting four to eight GPUs. But agentic AI workloads are changing that balance. Multi-step reasoning systems require more orchestration, memory handling, retrieval, routing, and coordination — work that increasingly falls back onto high-core-count CPUs. In some deployments, the ratio is beginning to move far closer toward 1:1.

That architectural shift matters enormously for AMD.

Counterintuitively, every new rack of Nvidia accelerators may create automatic demand for AMD EPYC CPUs regardless of who wins the GPU war. The market still talks about EPYC as though it is the supporting actor in AMD’s AI story. I increasingly suspect it is the structural revenue floor underneath the entire company.

And the numbers are beginning to validate that thesis.

AMD’s data-centre revenue surged 57% year-on-year in the latest quarter, while server CPU revenue is expected to grow roughly 70% in Q2. EPYC server market share also climbed to a record 27.4%, which is remarkable considering Intel once treated the server market as hereditary property.

This is the hidden asymmetry in the AMD story: Nvidia’s dominance may actually strengthen AMD’s most durable business line rather than weaken it.

The company does not necessarily need to dethrone Nvidia to keep compounding revenue aggressively. It simply needs AI infrastructure spending to keep expanding. In effect, AMD may have positioned itself as the toll collector on the roads leading to the GPU gold rush.

Financially, the machine is already substantial. Revenue has reached $37.45 billion, operating cash flow sits near $9.72 billion, and levered free cash flow exceeds $7 billion. Those are not speculative figures. They are the numbers of a company already embedded inside the plumbing of global computing infrastructure.

Conviction built quickly. Fragility often hides beneath crowded positioning

The problem is that the stock no longer trades on resilience. It trades on near-perfection.

ROCm and the Narrowing Software Window

The most important AI battle is not really happening in silicon. It is happening in developer ecosystems.

This is where AMD becomes simultaneously compelling and deeply risky.

The Nod.ai acquisition transformed ROCm from a fringe software layer into a credible CUDA alternative far faster than many expected. Downloads reportedly surged tenfold within a year, helping convince hyperscalers to commit serious workloads to AMD accelerators.

Yet Wall Street increasingly talks about ROCm as though success is merely a matter of time.

I think that framing misses the actual danger.

The question is not whether ROCm eventually becomes 'good enough'. The question is whether it becomes good enough before Nvidia’s Blackwell ecosystem locks in the next generation of developers.

That timing issue is everything.

CUDA’s dominance is not simply about performance benchmarks. It is accumulated habit, tooling, optimisation libraries, university training, and workflow dependency. Engineers build careers around CUDA. AI start-ups optimise entire architectures around it. Once ecosystems mature, inertia becomes frighteningly powerful.

AMD is therefore racing against a moving target. Every quarter Nvidia deepens developer dependence, the migration cost rises further. ROCm does not merely need improvement; it needs improvement fast enough to interrupt a lock-in cycle that may otherwise harden for another decade.

That is why AMD’s valuation feels more fragile than its operational performance suggests.

The stock is not truly trading on chips. It is trading on whether AMD can squeeze through a narrowing software window before Nvidia seals it shut.

And software windows have an irritating habit of slamming shut all at once.

Perfection trades beautifully—until volatility remembers mathematics

The Challenger’s Paradox

The most under-discussed risk in the AMD story is not technological. It is mathematical.

AMD’s path toward becoming a trillion-dollar company may simultaneously erode the per-share returns investors expect from getting there.

That contradiction sits at the centre of the investment case.

AMD recently approved a 65 million-share expansion to its equity participation programme. Pair that with the company’s large shelf registration and the realistic possibility of future acquisitions, stock-based compensation, or capital raises, and dilution of 50–80 million shares by 2027 becomes entirely plausible.

Normally, investors would shrug at that. Growth companies dilute. Silicon Valley practically treats share issuance as a love language.

But AMD’s situation is structurally different because it is trying to challenge one of the most financially dominant technology ecosystems ever created.

The more aggressively $Advanced Micro Devices(AMD)$ chases $NVIDIA(NVDA)$, the more capital it must consume. Software ecosystems require relentless investment. AI partnerships require incentives. Elite engineers require extraordinary compensation. Supply-chain guarantees require long-term commitments.

In other words, the better AMD executes its strategy, the more likely dilution becomes.

That is the paradox most investors are still missing.

A trillion-dollar market cap is entirely achievable if AI infrastructure spending continues exploding. But a trillion-dollar valuation spread across materially more shares changes the per-share compounding equation dramatically.

Investors may eventually discover they correctly predicted AMD’s strategic success while simultaneously overestimating how much of that success they would personally own.

It is a little like backing a brilliant Formula One team only to realise your seat in the car keeps getting fractionally smaller every lap.

Winning the race means little if ownership quietly keeps shrinking

Verdict: Brilliant Company, Delicate Stock

I admire AMD enormously as a business.

Lisa Su has engineered one of the most impressive corporate recoveries in modern technology. AMD now has meaningful positioning across CPUs, GPUs, adaptive computing, and AI infrastructure, while maintaining a healthy balance sheet and relatively modest debt levels.

But the stock has become far less forgiving than the company itself.

The hidden strength in the AMD story is that EPYC may thrive regardless of who dominates AI accelerators. That structural CPU exposure gives AMD a resilience many investors still underestimate.

The hidden danger is that the valuation assumes AMD solves the software gap quickly enough, scales AI revenue aggressively enough, and manages dilution carefully enough to preserve exceptional per-share outcomes.

That is an extraordinarily narrow path to walk.

I think AMD is increasingly likely to become indispensable to the AI economy. I am simply less convinced that shareholders buying at today’s valuation will collect enough of those tolls to justify the price of admission.

Because in the end, AMD may not need to own the AI kingdom to prosper.

It merely needs to keep collecting tolls on the road to it.

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