I’ve made three mistakes with AMD, and each one taught me something about how I read this company.
The first was 2019. I thought the EPYC server story was overhyped and that enterprise IT was too conservative to move off Intel. I passed. The stock rose about 150% over the next two years. The second was late 2021: I bought around $155, watched the semiconductor correction take it to $56, and convinced myself the downturn would last longer than it did. It didn’t. The third was early 2025, when I looked at the valuation, decided it was priced for perfection, and didn’t add. It has roughly tripled since.
I’m putting that on the table first because it shapes how you should read the rest. My record on this stock is humbling. AMD has done things I didn’t think it could do, and the reasons matter, because this isn’t luck and it isn’t just the AI wave everyone assumes it’s riding.
The Q1 2026 number that stopped me cold
For the quarter ending March 28, AMD posted $10.25 billion in revenue, up 38% year over year. Data center alone brought in $5.8 billion, up 57%, now 55.9% of the company. That’s a business that has remade its own mix in about four years.
The figure I can’t stop thinking about isn’t revenue. It’s market share. Mercury Research put AMD’s EPYC line at 46.2% of server CPU revenue in Q1 2026, up from 41.3% just one quarter earlier. Revenue, not units: AMD’s unit share was 27.4%, which means EPYC is winning the expensive end of the market, not just cheap volume deals. In 2019, AMD’s server CPU revenue share was under 10%. Intel had run that market for two decades.
One more data point: Intel’s share of the discrete GPU market fell from 8.2% in March to 3.7% in April 2026. That’s not a trend line. That’s a collapse.
What Lisa Su actually built
The standard story is that AMD is the scrappy underdog that got lucky when Intel tripped on its manufacturing roadmap. That’s incomplete in a way that leads investors to misjudge how durable AMD’s position is.
Intel wasn’t unlucky. It spent years optimizing near-term margins instead of investing in its foundries, got caught behind TSMC’s process technology, and is now attempting a manufacturing turnaround that takes years even when it goes well. AMD saw the window and engineered EPYC around it, not only on raw performance but on the whole package enterprise buyers actually evaluate: software compatibility, platform consistency, power efficiency, and a predictable roadmap.
The Turin generation, on 3nm and 4nm nodes, delivers performance-per-watt that has made cloud providers far more willing to switch. When power is the binding constraint on an AI cluster, a chip that does more work per watt saves real money on electricity and cooling. That’s not a spec-sheet talking point. It’s a line in a hyperscaler’s financial model.
So there are really two prizes here. One is server CPU share, where AMD already sits at 46.2% of revenue and climbing. The other, larger one is the data-center AI accelerator market, which Lisa Su has described as a hundreds-of-billions opportunity by the end of the decade, and where AMD is still the challenger.
The AI GPU story: real, but not the one people tell
AMD’s Instinct GPU line, the MI300X, the MI350, and the new MI450 series, is the most misread part of the thesis.
The 2024 bull case was basically “AMD is Nvidia but cheaper, so hyperscalers will switch.” That was always too easy. CUDA has a 15-year head start. Models are written for it, frameworks are optimized for it, and nobody spending $100 million on a training run wants to be the one that took a 15% performance hit at scale to save on hardware.
Where AMD’s chips have gained real traction is inference, serving already-trained models to users. Inference leans less on CUDA’s training optimizations, the performance gap narrows, and the price-performance math tips toward AMD in some configurations. As AI compute shifts from a training-heavy buildout toward inference at scale, that’s the part of the market growing fastest.
The commitments behind this aren’t pilots. OpenAI signed a multi-year deal for 6 gigawatts of AMD GPUs, an estimated $60 billion of business, with a warrant for OpenAI to buy up to 160 million AMD shares if both sides hit their milestones. Meta has committed to as much as 6 gigawatts of its own. AMD doesn’t need to unseat Nvidia at the top of training. It needs a meaningful slice of inference while that market expands toward hundreds of millions of daily users. The MI455X, shipping in the second half of 2026 in the Helios rack system, offers 40 petaflops of FP4 compute and 432GB of HBM4 memory, numbers that can challenge Nvidia on inference in ways the prior generation couldn’t.
The part that makes me uncomfortable
AMD trades at roughly 58 times forward earnings. Nvidia is around 25 times. Broadcom is near 38 times.
Sit with that. AMD trades at more than double Nvidia’s forward multiple, and Nvidia has faster revenue growth, higher margins, a deeper software moat, and closer customer relationships. The market is pricing AMD as if it will grow much faster than Nvidia over the next two to three years. In Q1 2026, AMD grew revenue 38%. Nvidia grew 85%. Closing that multiple gap requires either AMD accelerating hard or Nvidia decelerating hard. Neither is a given.
| Metric | AMD (Q1 2026) | Nvidia (Q1 FY2027) |
|---|---|---|
| Total revenue | $10.3B | $81.6B |
| Revenue growth YoY | +38% | +85% |
| Data center revenue | $5.8B | $75.2B |
| Gross margin (GAAP) | 53% | 75% |
| Forward P/E | ~58x | ~25x |
| Market cap | ~$685B | ~$5.7T |
The number I keep returning to is that gross margin gap: 53% against 75%, a 22-point spread. It’s not noise. It’s the structural advantage of Nvidia’s software-based pricing power. Control the platform everything runs on and you can price like a monopoly. AMD competes on hardware merit, and hardware merit carries lower margins.
What gaming and embedded tell you
Client and Gaming revenue was $3.6 billion in Q1, up 23%. Gaming specifically was $720 million, up 11%, held back by lower semi-custom revenue, the chips AMD makes for PlayStation and Xbox, now late in the console cycle. In discrete gaming GPUs, Nvidia holds about 80% of the market, and AMD’s Radeon line hasn’t closed the gap at the high end where the money is. Gaming is around 7% of revenue and isn’t a driver of the stock. What it does tell you is that AMD has real, diversified revenue outside AI, which is a healthier setup than a company whose entire valuation rides one product cycle.
Embedded revenue was $873 million, up 6%. That’s the industrial, automotive and IoT business: slower growth, but sticky, because customers with long design cycles don’t swap chips once they’ve been designed in.
The Intel situation matters more than it looks
Intel’s data-center and AI revenue in Q1 2026 was $4.1 billion. Still a substantial business, but down from prior quarters and now behind AMD’s $5.8 billion. For the first time in a very long time, AMD’s data-center revenue is larger than Intel’s.
That is not what Intel’s management or most analysts expected two years ago. Intel’s Granite Rapids server chips are a genuine response, not a surrender. But the server CPU market now runs on AMD’s multi-year lead on TSMC’s most advanced node, and that shows up directly in procurement decisions at the major clouds. Enterprise IT moves slowly by design: once a cloud validates an architecture, builds software on it, and deploys it at scale, it doesn’t rip it out next quarter. Those installed deployments keep feeding AMD’s 46.2% share through replacement and expansion orders even if Intel’s products get competitive again.
Where this leaves the investment case
Before I buy anything, I make myself state the bull and bear case clearly enough that I know exactly what I’m betting on.
The bull case: server CPU share keeps climbing past 50% as Intel’s turnaround runs late; inference GPUs become a real second revenue engine beyond EPYC; and the MI400 series closes enough of the gap with Nvidia to give AMD 15–20% of the AI GPU market by 2027. If all three land, earnings grow fast enough to justify even a high-50s multiple.
The bear case: roughly 58 times forward earnings is too much for a company earning 22 points less gross margin than its main competitor, against a software moat that inference share gains can’t fully cross. In that world, multiple compression is severe, because the current price leaves almost no cushion for disappointment.
My own view: AMD is a better business than it has ever been, Lisa Su has built something real in server CPUs, and the inference angle is credible enough to justify a premium over ordinary semiconductor names. But 58 times forward earnings asks me to be very confident about a specific outcome at a moment when Nvidia is growing twice as fast with margins AMD hasn’t proven it can match. I can hold what I own. I can’t add aggressively here.
| Scenario | AMD revenue (2027E) | EPS | Justified P/E | Implied price |
|---|---|---|---|---|
| Bull: GPU share grab + EPYC holds | ~$50B | ~$10+ | 35x | $350+ |
| Base: steady execution, modest GPU share | ~$40B | ~$7.50 | 28x | ~$210 |
| Bear: Intel recovers + GPU disappoints | ~$32B | ~$5 | 22x | ~$110 |
| Current price (~$421) implies | — | — | ~58x fwd | Requires the bull case |
Read that table slowly. The current price assumes the bull scenario, both the GPU share gain and continued EPYC dominance. The base case, which would be a fine result for most semiconductor companies, implies the stock is materially overvalued today.
That’s as far as I’ll go. Not because AMD is a bad company, it’s one of the best-run businesses in semis, and the EPYC story has more runway than the share numbers alone suggest. But a great company and a great stock at this price are different questions, and the set of outcomes that makes $421 attractive isn’t one I’m confident enough in to ignore the others.
I’ll keep watching the MI450 ramp. If the Helios rack shipments in the second half of 2026 show real hyperscaler adoption at scale, the GPU thesis gets meaningfully stronger, and that’s the signal to revisit. I’ve been wrong on this stock three times already. If I buy it a fourth, I’ll check the rating data first.
Financial disclaimer: The content on StockVane is for educational and informational purposes only and should not be construed as professional financial advice. Stock market investing involves risk of loss.