The score, up top
The turnaround is real. Not "real" the way a press release means it — real the way a 10-K means it. When Lisa Su took over AMD in October 2014, the stock cost $3.28, the whole company was worth about $2.5 billion, and it had just finished a year in which it lost $364 million. Today the stock is $504.80, the company is worth roughly $824 billion, and it made $4.3 billion last year. That part is not a story. It's arithmetic, and you can check every line of it below.
The question worth asking is the second one: what does the stock have to do from here to justify $505? Because at this price you are not buying the turnaround. That already happened. You're buying the sequel — AMD taking a real bite out of Nvidia in AI — at roughly 130 times trailing earnings. Both things can be true at once: one of the great CEO runs in tech history, and a stock priced like the next twelve years go right too.
The patient: AMD in October 2014
AMD was the company Intel used to step on. It made the chips you bought when you couldn't afford the good ones. By the fall of 2014 the numbers looked like this:
- Stock price on October 8, 2014 — the day Su was appointed CEO: $3.28
- Market value: about $2.5 billion (roughly 775 million shares outstanding)
- Fiscal 2014 revenue: $5.51 billion
- Fiscal 2014 result: a net loss of $364 million
- The biggest customer-facing problem: its server CPU share was a rounding error. Intel held essentially the entire server market.
And it got worse before it got better. Revenue fell to $3.99 billion in 2015. The stock's monthly closes bottomed at $1.72 in September 2015. There were serious people writing serious articles about whether AMD would survive at all.
Meanwhile Intel — the company AMD was supposedly competing with — did $55.9 billion in revenue in 2014 and was worth about $162 billion, roughly 64 AMDs. Keep that ratio in your head. It flips before this post ends.
Who's behind it: an engineer, not a turnaround artist
This is the part of every review where we check the people, because it's not just what you're investing in — it's who. Su's file is unusually easy to check, because AMD laid it out in an SEC filing the day she got the job.
She has three electrical engineering degrees from MIT, including a PhD. She spent 13 years at IBM, ending as vice president of its Semiconductor Research and Development Center — the group responsible for the direction of IBM's silicon technology. In 2007 she joined Freescale Semiconductor as chief technology officer, then ran its networking and multimedia business. She arrived at AMD in January 2012, became chief operating officer in July 2014, and got the top job that October, replacing Rory Read.
Notice what she is not: not a finance person, not a marketing hire, not a "visionary" brought in to rebrand a dying company. She's a chip engineer who had spent twenty years inside the exact industry AMD was losing. That matters, because the bet she made next was a technical bet, and she was qualified to know whether it was working before the revenue showed it.
The ownership check, because we run it on everyone: the 2026 proxy lists Su as beneficially owning about 4.3 million shares — around $2.2 billion worth at today's price, and under 1% of the company. The float belongs to the index funds: Vanguard at about 8.8% and BlackRock at about 7.7% are the two biggest holders. Insiders as a group own roughly half a percent.
Her recent trades are the standard CEO pattern, and we'll describe them plainly rather than dress them up: she sells on a schedule. Her Form 4 filings show planned sales under a Rule 10b5-1 plan set up in advance — 125,000 shares in May 2026, 15,000 in August, and 95,000 shares on September 10, 2026 at an average of $506.80 (about $48 million), plus a 35,000-share gift the same day. She still holds millions of shares. That's "diversifying on a timetable," not "running for the exit" — but it is selling, not buying, and nobody at AMD has been buying shares on the open market lately. File that under: the people closest to the machine are happy to hold a lot of it, and in no rush to add at $500.
The bet: Zen, and renting the best factory instead of owning one
Su's turnaround had two moves, and both were bets the old AMD couldn't have made.
Move one: one architecture, everywhere. AMD scrapped its failing chip designs and built a new CPU core called Zen, designed to scale from laptops to servers. Ryzen brought it to PCs in March 2017. EPYC took it into servers — Intel's profit vault — the same year. Then AMD kept shipping: Zen 2, Zen 3, Zen 4, each on time, each faster. Boring word, "execution." It's the whole story.
Move two: don't own the factory. AMD had already spun off its fabs years earlier, so it designed chips and had TSMC — the best manufacturer on the planet — build them. Intel did the opposite: it insisted on making its own, then spent the Zen years tripping over delay after delay in its own factories. AMD's design team only had to beat Intel's design team. Intel's had to beat AMD's design team and its own fab problems.
The scoreboard for that bet, from Mercury Research, the firm that tracks x86 shipments: AMD's share of server CPU units was 2.9% in early 2019. By the second quarter of 2026 it was 34.5% — and because AMD sells the expensive chips, its share of server CPU revenue hit a record 46.2% in the first quarter of 2026. Nearly half the money spent on x86 server processors now goes to the company that had 3% of the units seven years ago. Intel still ships more boxes. It just ships the cheaper ones.
The AMD mark. Editorial use only — no affiliation.
Then vs. now: the whole turnaround in one table
| October 2014 | October 30, 2026 | |
|---|---|---|
| Share price | $3.28 | $504.80 |
| Market value | ~$2.5 billion | ~$824 billion |
| Revenue (fiscal year) | $5.51B (2014) | $34.64B (2025) |
| Net income (fiscal year) | −$364M (2014) | +$4.34B (2025) |
| Intel's revenue (fiscal year) | $55.9B (2014) | $52.9B (2025) |
Read that last row twice. Over the twelve years Su has run AMD, its revenue went up more than six times — and Intel's went down. AMD is now worth about 2.4 Intels. In 2014, Intel was worth about 64 AMDs.
The last twelve months alone would be a career for most CEOs. Revenue for the four reported quarters through June 2026 totals $41.3 billion. The stock hit a record close of $649.42 on October 6, 2026, which put AMD over a $1 trillion market value for the first time. It has since fallen about 22% in a rough few weeks for chip stocks — which is its own kind of information about how much hope is baked into the price.
AMD monthly closing price, October 2014 – October 2026. Source: Yahoo Finance data, checked Oct 30, 2026. The flat line along the bottom is roughly a decade under $200. Everything after is the AI repricing.
The new exam: AI, where Nvidia is not a rival — it's the weather
Here is where the puff pieces stop and the math starts. AMD's data center business is genuinely booming: $6.7 billion in the second quarter of 2026, up 107% year over year, 58% of the whole company's revenue, with third-quarter company revenue guided to about $13 billion. EPYC CPUs and Instinct GPUs are both selling. By any normal standard, this is a triumph.
Now the comparison nobody at AMD puts on a slide. Nvidia reported its second quarter of fiscal 2027 in August: revenue of $96.2 billion, of which data center was $89.0 billion — in one quarter. Nvidia's data center business alone did roughly 13 times AMD's data center business in the same season, and Nvidia's quarterly profit ($54.0 billion) was bigger than AMD's entire revenue for the last twelve months.
That's the CUDA moat in dollar form. Nvidia doesn't just sell faster chips; it sells the software layer a decade of AI code is written on. AMD's counter is ROCm, its open alternative, plus a simple pitch to the giant buyers: you do not want one supplier for the most important input in your business. And to be fair, that pitch is landing — which brings us to the deals, and the part of them that gets left out of the headlines.
The deals — and who actually pays for them
The announcements, as reported and as disclosed in AMD's own filings:
- OpenAI (October 2025): up to 6 gigawatts of AMD GPUs across generations, starting with 1 gigawatt of MI450 systems in the second half of 2026.
- Meta (February 2026): up to another 6 gigawatts, built around a custom MI450 chip.
- Anthropic (July 2026): up to 2 gigawatts of MI450 systems in AMD's Helios racks — with AMD committing to invest up to $5 billion in Anthropic.
Fourteen gigawatts of commitments from the three most serious AI buyers on earth. That's real validation. Now the fine print, which is our whole beat here: attached to the OpenAI and Meta deals are warrants — the right for each customer to buy up to 160 million AMD shares at $0.01 apiece. One cent. The warrants vest as the gigawatts get delivered and as the stock hits price targets — the final OpenAI tranche requires a $600 share price. As of the June 2026 quarterly report, none had vested.
Do the arithmetic the press releases don't do for you. If both warrants fully vest, AMD's share count rises from about 1.63 billion to about 1.95 billion — 19.6% more shares, handed to two customers for a penny each. Existing shareholders pay for those deals, in dilution, if the deals succeed. That's not a scandal; the terms were disclosed in SEC filings from day one, and if 12 gigawatts actually ship, the revenue dwarfs the giveaway. But when a customer is also a future 10% shareholder at $0.01, "demand" is doing some complicated work in that sentence. Check who pays. It's kind of the house rule.
Promises vs. delivery: the part of her record that's hard to argue with
One reason the market gives Su the benefit of the doubt: she's earned it in public, in numbers, repeatedly.
When the MI300 AI chip launched, AMD guided 2024 data-center GPU revenue to about $2 billion. Then it raised the number to $4 billion in April, $4.5 billion in July, and "exceed $5 billion" by October — and full-year 2024 Instinct revenue came in over $5 billion. CEOs sandbag sometimes. What they rarely do is raise a brand-new product line's forecast three times in a year and then clear the final bar.
The honest debits exist too. In the second quarter of 2025, AMD took an $800 million inventory charge when U.S. export controls blocked its MI308 chips bound for China — a reminder that this business runs through geopolitics whether it wants to or not. And the AI revenue, for all its growth, is still a single-digit share of the market Nvidia owns: $6.7 billion against $89 billion a quarter is a foothold, not a siege.
The price today: what has to be true at $505
So, the BogusAPY question. At $504.80, with trailing twelve-month earnings of $3.89 a share, AMD trades at roughly 130 times trailing earnings and about 20 times trailing sales. The turnaround — the thing that already happened — cannot justify that by itself. For this price to pay off, some version of the following has to come true:
- The gigawatts ship. A meaningful share of those 14 GW of commitments has to turn into delivered, paid-for systems on schedule, starting with the first OpenAI and Meta gigawatts in the second half of 2026.
- ROCm gets good enough that running AI on AMD stops being a science project for anyone outside the three anchor customers.
- Data center revenue keeps compounding at something like its current pace for years — from a $6.7 billion quarter toward numbers that make a dent in an $824 billion valuation.
- The dilution stays worth it — all of that has to happen on a share count that may grow by a fifth along the way.
None of that is fantasy. All of it is future. The next checkpoint is close: AMD reports its third quarter on Tuesday, November 3, after the close — the first report since the stock's trillion-dollar moment, with that ~$13 billion guidance on the line.
Bottom line
Lisa Su took a company worth $2.5 billion, losing money, with 3% of the market that mattered, and turned it into an $824 billion company that takes 46 cents of every x86 server CPU dollar and has OpenAI, Meta, and Anthropic signed up for 14 gigawatts of its AI chips. Checked against the filings, the legend is undersold if anything.
The stock is a different claim than the CEO. The CEO's twelve years are audited fact. The stock's next twelve years are a bet — on AI share, on software, on warrants and dilution math — priced at 130 times earnings. Respect the operator. Price the bet like a bet.
Check it yourself: links and sources
- AMD's appointment announcement (Oct 8, 2014) with Su's IBM/Freescale record: AMD investor relations
- Latest reported quarter (Q2 2026 results release, filed with the SEC): SEC EDGAR
- Fiscal 2025 Form 10-K and all AMD filings: SEC EDGAR — AMD
- 2026 proxy statement (Su's share ownership; Vanguard and BlackRock stakes): SEC EDGAR
- Nvidia Q2 fiscal 2027 results (the $89.0B data center quarter): SEC EDGAR
- Mercury Research x86 share data for Q2 2026, via KitGuru: kitguru.net
- Price history used for the chart and the then-vs.-now table: Yahoo Finance (AMD), checked Oct 30, 2026
All figures checked Oct 30, 2026. AMD's most recent reported quarter at publication is Q2 2026 (ended June 27, 2026); Q3 2026 results are due Nov 3, 2026.
Get posts like this by email: Subscribe to BogusAPY
Not financial advice.
Comments
Post a Comment