Every AMD stock price prediction published this month looks like it disagrees with the last one, and almost none of them actually do. Three FinanceFeeds pages since 5 September carried bear cases of $310, $385 and $500, a spread wide enough to look like an argument. Normalise each one against the price on the day it was written and the argument mostly evaporates: they sit at 35.1%, 31.2% and 19.8% below their own spot. What changed was not the view of AMD’s downside. What changed was the denominator. AMD closed Friday 25 September at $630.63, the highest close of the trailing year, up 291.0% over twelve months on stockanalysis.com daily closes. The interesting question is not whether the bulls have been right. It is that the published range has been quietly narrowing on both sides at exactly the moment the stock became most extended.
Here is the part that has gone almost unremarked, and it is buried in a filing rather than a price chart. When AMD issued Meta a performance warrant over 160 million shares in February, vesting was made subject to “specified AMD stock price thresholds that escalate to $600 per share for the final tranche”. Friday’s close cleared that top threshold. Combined with the 160 million-share warrant issued to OpenAI in October 2025, AMD has now promised 320 million shares at an exercise price of one cent to its two largest AI customers, against 1,632,475,042 shares outstanding. Every dollar the share price rises transfers roughly $320m of value to those two holders. At Friday’s close the pair is worth about $202bn. The $1trn headline that ran last week was calculated on a share count that AMD has already contracted to expand by 19.6%.
Key facts
- AMD closed at $630.63 on Friday 25 September 2026, its highest close of the trailing year, after touching $639 intraday — stockanalysis.com, as of 4:00pm ET, 25 Sep 2026
- That is +291.0% over twelve months and +182.2% year to date, from a 52-week low of $154.78 — FinanceFeeds calculation from stockanalysis.com daily closes, 25 Sep 2026
- Market value of $1.03trn on 1,632,475,042 shares outstanding, meaning AMD crossed $1trn at $612.57 a share — AMD Q2 FY26 10-Q cover page, 29 Jul 2026
- Q2 2026 revenue was a record $11.536bn, up 50% year on year, with Data Center revenue of $6.7bn up 107% and 58% of the company — AMD Q2 2026 results, 4 Aug 2026
- Q3 2026 guidance is approximately $13bn plus or minus $300m, about 41% year-on-year growth, at roughly 56% non-GAAP gross margin — AMD Q2 2026 results, 4 Aug 2026
- Warrants over 320 million shares at $0.01 are outstanding: 160m to OpenAI OpCo and 160m to Meta, each fully vesting only on 6 gigawatts of Instinct purchases — AMD 8-K, 6 Oct 2025 and AMD 8-K, 24 Feb 2026
- AMD closed $4.75bn of senior notes on 17 August 2026 across four tranches at coupons from 4.600% to 5.500% — AMD 8-K, 17 Aug 2026
What is actually happening, and why the tape moved this fast
The move that produced Friday’s record is compressed into about three weeks. AMD closed at $477.57 on 4 September. By 21 September it had put on 9.95% in a single session to close at $615.52, crossed $1trn in market value on the way, and then spent four sessions consolidating in a tight band to finish at $630.63. That is a 32.0% gain in fifteen trading sessions, on a company that already carried a twelve-figure valuation when the run began.
The fundamental case underneath it is not speculative, which is what makes this rally different in character from AMD’s previous ones. Second-quarter revenue of $11.536bn was a record, up 50% year on year, and the mix is what matters: Data Center did $6.7bn, up 107%, and now accounts for 58% of the company. Client revenue grew 23% to $3.1bn. Gaming fell 31% to $779m on weaker semi-custom volumes, which is the one segment genuinely shrinking. Non-GAAP gross margin reached 56%, up 13 percentage points in a year, and non-GAAP operating margin reached 27% against 12% a year earlier. Operating leverage of that magnitude is the actual engine of the re-rating.
Guidance extended it. AMD told the market to expect roughly $13bn in the third quarter, about 41% growth, with non-GAAP gross margin holding near 56%. The product news landed alongside: AMD launched the Helios rackscale system and the Instinct MI400 series, including the MI455X. Having tracked AMD’s accelerator roadmap since the MI300 launch, the notable change is not the silicon specification. It is that the customer list in AMD’s own results release now reads Anthropic, Cirrascale, HUMAIN, Meta, Microsoft, OpenAI, Oracle, Tensorwave and Vultr. For an eighteen-month stretch AMD’s problem was never the chip; it was that the largest buyers of AI compute would not design it in. That has changed. The read-across runs through the whole supply chain, and it is visible in the way Intel repriced on CPU demand in the same window.
Lisa Su, chair and chief executive at AMD, framed the second half in the results release: “We enter the second half with strong momentum as EPYC demand accelerates, Instinct deployments scale and Helios begins to ramp.”
What the named players are actually doing
The response worth tracking is not analyst commentary. It is what AMD’s customers and AMD’s treasury have committed to in writing.
OpenAI signed first. The 8-K filed on 6 October 2025 discloses a warrant over 160 million shares at $0.01, issued alongside a product purchase agreement, with a binding commitment to buy the initial one gigawatt of MI450 series product and full vesting contingent on six gigawatts. Meta followed on 23 February 2026 with a structurally near-identical instrument: 160 million shares at $0.01, an amendment to a Master Purchase Agreement dating to May 2023, a binding commitment on the first gigawatt equivalent, full vesting at six gigawatts, and the stock-price ladder topping out at $600. Both warrants run to 2030 and 2031 respectively and both carry registration rights, including, in Meta’s case, the right to sell through underwritten block trades.
Read those two documents together and the commercial logic is explicit. AMD is buying design-in commitments with equity rather than discounting the hardware, and it is doing so on terms that only pay out if the volume actually ships. The vesting is milestone-linked, not time-linked. A sceptic reads that as circular financing. A more precise reading is that AMD has converted a customer-acquisition problem into a contingent equity cost, and pushed the cost into the share count rather than the gross margin line, which is exactly why gross margin has been expanding through the ramp.
The treasury moved too. On 17 August AMD closed $4.75bn of senior notes: $1.25bn at 4.600% due 2029, $1.50bn at 5.000% due 2031, $1.00bn at 5.250% due 2033 and $1.00bn at 5.500% due 2036, underwritten by Barclays, BofA, Citigroup, J.P. Morgan, Morgan Stanley and Wells Fargo. Against total debt of $3.226bn and $13.1bn of cash and short-term investments at the end of Q2, that roughly triples the debt stack on a company that has run conservatively for years. Capital intensity is arriving. The same dynamic is visible at Micron ahead of its 30 September print, where the memory side of the same build-out is being financed.
Jean Hu, executive vice president, chief financial officer and treasurer at AMD, was direct about the shape of the second half: “We expect Data Center sales to accelerate in the second half of 2026, driving stronger overall revenue growth and continued earnings expansion.”
The data: a range that keeps narrowing
Setting the last three FinanceFeeds AMD scenarios beside each other, normalised to the closing price available when each was written, is the clearest way to see what has actually been happening to the published range.
| Published | Spot then | Bull | Bear | Bull above spot | Bear below spot |
|---|---|---|---|---|---|
| 5 September 2026 | $477.57 | $700 | $310 | +46.6% | -35.1% |
| 19 September 2026 | $559.82 | $725 | $385 | +29.5% | -31.2% |
| 22 September 2026 | $623.77 | $780 | $500 | +25.0% | -19.8% |
| This page, 25 September 2026 | $630.63 | $780 | $430 | +23.7% | -31.8% |
The bear column is the tell. The first two pages sat in a consistent 31% to 35% band. The 22 September page cut that to 19.8%, and it did so in the week the stock went vertical. That is the classic shape of a downside estimate being dragged upward by price rather than by evidence. Nothing in AMD’s disclosure between 19 and 22 September justified halving the implied drawdown; the company filed no 8-K in that window. This page restores the earlier band at 31.8%, which is the single substantive disagreement it has with its immediate predecessor.
On the upside, the compression is real but defensible. A $700 target was +46.6% away on 5 September. The same class of target is +23.7% away now because the stock has done most of the work. Put the levels in valuation terms rather than percentage terms and they are legible. Q2 delivered $1.66 of non-GAAP diluted earnings on about 1,663m diluted shares. Annualising the Q3 guide at a stable non-GAAP net margin points to roughly $1.85 to $1.90 for the September quarter. A full-year 2026 figure near $7 is consistent with the guidance AMD has actually given. At $630.63 that is about 90 times this year’s earnings, and somewhere near the mid-50s on a 2027 number in the $11 to $13 range. The comparison that matters is a sector where Nvidia carries its own bull and bear spread on a far larger revenue base, and where TSMC’s monthly revenue prints now function as the industry’s highest-frequency demand signal.
The structural tension nobody is pricing cleanly
The warrants are the unresolved question, and the market is currently treating them as a footnote.
On the current count, 320 million new shares is a 19.6% increase. Measured against the enlarged base, OpenAI and Meta would together own 16.4% of AMD. Recompute the headline on that basis and AMD passed $1trn of fully-diluted value at $512.17 a share, not $612.57. The two numbers describe the same company; only one of them is the number that gets written down.
There is a legitimate counter. Neither warrant vests on price alone. Both require six gigawatts of Instinct purchases, and six gigawatts from each of two customers is an enormous revenue event that would arrive long before the shares did. Dilution that is paid for by the revenue that triggers it is not the same category of problem as dilution raised to fund losses. The honest framing is that AMD has sold a call option on its own equity to the two buyers best placed to determine whether it gets exercised, and the strike is one cent.
Customer concentration is the second-order version of the same tension. AMD has solved the design-in problem by making its two largest prospective buyers shareholders. That aligns them through the ramp. It also means a material share of the Data Center growth now depends on the capital expenditure plans of a small number of firms whose own AI spending is under increasing scrutiny, and the political economy around that spending is hardening. The regulatory overhang is not antitrust here; it is export control. AMD’s own Q2 comparative still carries the $800m inventory charge taken in Q2 2025 on the MI308 export restriction, a reminder that a single administrative decision can strand a product line.
The call: base $660, bull $780, bear $430
This AMD stock price prediction anchors on the regular-session close of $630.63 on Friday 25 September 2026, not the after-hours print of $630.99.
Base case, $660, about 4.7% above spot, roughly 45% probability. AMD delivers the $13bn Q3 guide within the range, holds 56% non-GAAP gross margin, and guides Q4 to sequential growth without a step-change. The multiple compresses gently as earnings catch up rather than as price falls. This is the path where the stock does very little for two quarters and the valuation repairs itself through the denominator. It is the least discussed outcome and, on the current setup, the most likely one.
Bull case, $780, about 23.7% above spot, roughly 30% probability. The MI450 and Helios ramp lands on schedule, the first gigawatt tranches for both OpenAI and Meta ship, and AMD guides 2027 above the $11 to $13 consensus band. A 2027 figure near $14 at a mid-50s multiple supports $780, which is about $1.27trn of market value. The causal chain is specific: first-tranche shipment confirmation is what converts the warrants from a dilution story into a revenue-visibility story, and that reframing is worth more to the multiple than the incremental earnings.
Bear case, $430, about 31.8% below spot, roughly 25% probability. Hyperscaler capital expenditure enters a digestion phase, the MI450 ramp slips a quarter, or an export decision strands volume. On a 2027 figure of $12 and a de-rate to 35 times, $430 is arithmetic rather than panic. It is worth saying plainly that a 31.8% drawdown in a stock that has risen 291% in a year is an ordinary event, not a catastrophe.
What would change my mind. On the bull side: confirmation in an 8-K or a results release that a first gigawatt tranche has actually shipped and a warrant tranche has vested. That is a filing event, not a press release, and it is checkable. On the bear side: a Q3 print at or below the low end of the $12.7bn guidance floor, or a Q4 guide that implies Data Center growth decelerating below 60% year on year. Either would invalidate the base case immediately. A move above $639, the intraday high set on 25 September, on expanding volume would put the bull path in control; a weekly close below $500 would confirm the bear path.
Frequently asked questions
What is the AMD stock price prediction for 2026?
This analysis sets a base case of $660, a bull case of $780 and a bear case of $430, measured against the $630.63 close on 25 September 2026. The base case assumes AMD delivers its guided $13bn third quarter at roughly 56% non-GAAP gross margin without a step-change in the Q4 outlook, leaving valuation to repair through earnings growth rather than price movement.
Has AMD passed a $1 trillion market capitalisation?
Yes, on the current share count. With 1,632,475,042 shares outstanding per the Q2 FY26 10-Q cover page, AMD crosses $1trn at $612.57 a share and was worth about $1.03trn at Friday’s close. Including the 320 million warrant shares promised to OpenAI and Meta, the fully diluted threshold falls to $512.17, which AMD passed considerably earlier.
How much dilution do the OpenAI and Meta warrants represent?
Each warrant covers 160 million shares at a $0.01 exercise price, so 320 million in total, a 19.6% increase on the current count. The holders would own 16.4% of the enlarged company. Neither vests on price alone: full vesting requires six gigawatts of Instinct purchases from each customer, with Meta’s final tranche also requiring a $600 share price, a threshold now cleared.
Why did AMD raise $4.75bn of debt in August 2026?
AMD closed four tranches of senior notes on 17 August 2026 at coupons between 4.600% and 5.500%, stating the proceeds were for general corporate purposes which may include repaying debt. Against $3.226bn of total debt and $13.1bn of cash and short-term investments at the end of the second quarter, the raise roughly triples the debt stack as the accelerator ramp increases capital intensity.
What is the biggest risk to the AMD bull case?
Timing on the Instinct MI450 and Helios ramp, and export control. A one-quarter slip in first-tranche shipments would delay the warrant vesting that converts the dilution narrative into a revenue-visibility narrative. Separately, AMD’s Q2 2025 comparative still carries an $800m inventory charge caused by the MI308 export restriction, showing how quickly an administrative decision can strand a product line.
Which AMD segment is actually driving the growth?
Data Center. It produced $6.7bn of the $11.536bn second-quarter total, grew 107% year on year and now represents 58% of company revenue. Client grew 23% to $3.1bn and Embedded grew 19% to $977m. Gaming is the one shrinking segment, down 31% to $779m on lower semi-custom volumes.
This article is analysis, not investment advice. Figures are sourced to AMD’s SEC filings and to stockanalysis.com closing data as of 25 September 2026. Scenario levels are the author’s estimates and are not forecasts of actual results. Capital is at risk and past performance does not indicate future returns.