Advanced Micro Devices (AMD)

Fifteen years ago AMD was the chip company nobody wanted. It sold processors at a discount to stay solvent while Intel took the profitable end of the market. Today it is the only serious challenger to Nvidia in the most valuable hardware category in the world, and the share price says the market has already decided how that contest ends.

In the past twelve months the shares have gone from roughly $140 to $538. The company is now worth about $878 billion. That is not a recovery story being priced in. That is a near certainty being priced in.

What follows is what the company actually sells, what it earns, and what the share price now requires to be true.

What AMD actually sells

AMD reports in four parts.

Data Center is the biggest and the one that matters most. It covers EPYC server processors, which are the main chips inside the computers that run company systems and cloud services, and Instinct accelerators, which are the specialised chips that train and run AI models. In the quarter to June 2026 this segment brought in $6.7 billion, or 58% of everything AMD sold.

Client is Ryzen chips for laptops and desktops. $3.1 billion in the June quarter, up 23%.

Gaming is custom silicon for the Xbox and PlayStation plus Radeon graphics cards. $779 million, down 31%, because console makers are late in their product cycle and ordering less.

Embedded is the Xilinx business AMD bought in 2022, selling programmable chips to industrial, aerospace and defence customers. $977 million, up 19%.

For the full financial year to December 2025, the group made $34.6 billion in revenue, up 34%. Data Center alone was $16.6 billion of that.

The numbers have moved a long way, fast

Growth has accelerated through 2026, not slowed.

• Quarter to March 2026: revenue $10.25 billion, up 38%. Data Center $5.78 billion, up 57%.
• Quarter to June 2026: revenue $11.5 billion, up 50%, a company record. Data Center $6.7 billion, up 107%.
• Guidance for the September quarter: $12.7 billion to $13.3 billion, which would be growth of about 41%.

Profitability is improving alongside it. In the June quarter AMD reported adjusted earnings of $1.66 per share and adjusted net income of $2.8 billion. On the stricter accounting basis that includes the cost of writing down acquisitions, earnings were $1.38 per share and net income $2.3 billion. The gap between those two numbers is mostly the Xilinx purchase being gradually written off against profit, which is an accounting charge rather than cash leaving the business.

The balance sheet is genuinely strong. At the end of December 2025 AMD held $10.6 billion in cash and short-term investments against $3.2 billion of debt. It owes less than it holds.

The deals that changed the story

In October 2026 AMD did something no chipmaker had done at this scale. It sold equity in itself to win orders.

OpenAI and Meta each agreed to buy up to 6 gigawatts of AMD Instinct accelerators, 12 gigawatts between them. Gigawatts measure the electricity a data centre draws rather than the number of chips, which is how this industry has started counting. Chief executive Lisa Su has said revenue per gigawatt runs into double-digit billions. On that basis 12 gigawatts is worth at least $120 billion, which is more than seven times AMD's entire Data Center revenue for 2025.

In exchange, AMD issued each customer a warrant over 160 million of its own shares, 320 million in total, at an exercise price of one cent each.

Those warrants are not free shares handed over on signature. They vest in stages as the chips are actually bought, with the full 160 million only vesting once a customer reaches the whole 6 gigawatts. Each stage also carries a share price condition, rising to $600 for the final one, and both customers have technical and commercial conditions to meet before they can exercise anything. OpenAI's warrant runs to October 2030 and Meta's to February 2031. As at the end of June 2026, none had vested. The first chip shipments were scheduled for the second half of 2026.

What the warrants cost existing shareholders

This is the part most coverage skips.

AMD had roughly 1.63 billion shares in issue at the end of July 2026. If all 320 million warrant shares eventually vest, that becomes about 1.95 billion. OpenAI and Meta would own somewhere near 16% of the company, and every existing shareholder's claim on future profits would shrink by roughly the same proportion.

Buybacks will not meaningfully offset that. AMD bought back $221 million of its own stock in the first half of 2026 and has $9.2 billion of authorisation left, which at current prices retires only about 15 million shares. A pending all-share purchase of World Labs, at roughly $8.2 billion, adds around 13 million more shares rather than removing them.

So the trade is explicit. AMD has bought itself a guaranteed position in the two largest AI buildouts outside Nvidia's customer base, and the price is about a sixth of the company. Whether that is a good trade depends entirely on whether the revenue arrives, because the dilution is tied to purchases. If the chips are not bought, the shares do not vest, but the revenue does not arrive either.

Where AMD actually stands against Nvidia

The common claim is that Nvidia holds about 90% of the AI accelerator market. That was closer to true two years ago than it is now.

Industry estimates put Nvidia's share of AI accelerator revenue at roughly 87% in 2024, around 81% in 2025, and near 75% for 2026. AMD is estimated at somewhere between 5% and 8%, making it the largest merchant competitor but still a small one. The rest is increasingly taken by hyperscalers designing their own chips: Google, Amazon, Microsoft and Meta. These are third-party estimates rather than reported figures, and they should be read as approximate.

The reason Nvidia's lead has held is software, not silicon. Nvidia's CUDA platform is the toolkit almost every AI engineer has learned on, and switching to AMD's equivalent, ROCm, means rewriting work that already runs. AMD has closed much of the hardware gap. The software gap is the one that still decides purchasing decisions, and the OpenAI and Meta agreements are partly an attempt to buy past it, by giving the two most capable AI engineering teams in the world a financial reason to do the porting work themselves.

The business nobody argues about

Lost in the accelerator debate is that AMD has quietly won the server processor market on value.

Mercury Research data for the first quarter of 2026 put AMD at 46.2% of x86 server processor revenue, a record, and 33.2% of units shipped, also a record. Unit share was 28.8% a quarter earlier and 27.2% a year before that. Intel still ships about two-thirds of server processors by volume, but sells them for less.

This matters because it is already in the numbers. It is a completed shift rather than a promise, and it is the part of the Data Center business that does not depend on the AI buildout continuing.

What the share price requires

At $538 a share, AMD trades at roughly 21 times its revenue over the past twelve months, about 48 times expected earnings, and 138 times earnings already reported.

A year ago, the same business was valued near 9 times sales. The company has grown fast. The share price has grown faster.

That valuation does not require AMD to beat Nvidia. It requires the 12 gigawatts to be delivered close to schedule, at the pricing Su has implied, while roughly 16% more shares come into existence, and while the server processor business keeps its gains. It also assumes the AI buildout itself continues at the pace currently planned by a handful of very large buyers. OpenAI and Meta are now simultaneously AMD's biggest customers and, potentially, among its largest shareholders. That alignment is useful while demand holds. It is a concentration risk if it does not.

What to watch from here

The September quarter results, due 3 November 2026. The first question is whether Instinct shipments to OpenAI and Meta have actually started, and at what volume, because warrant vesting and revenue both depend on it.

Any disclosure of warrant tranches vesting. AMD has not published the size of each stage. The first vesting event will show how much dilution each gigawatt actually costs.

Data Center gross margin. Accelerators have historically carried lower margins than server processors. If accelerator sales grow into a much larger share of revenue, group margins can fall even as revenue climbs.

Nvidia's share trend. If independent estimates show Nvidia holding above 75% through 2027, the case for AMD taking meaningful share is weaker than the current price implies.