AMD Stock Prediction: $757 Bull Case, $390 Bear Case

The myth is that AMD still has room to run up to the average price target. It does not. Advanced Micro Devices (Nasdaq: AMD) closed Monday, 5 October 2026, at $631.75, down 0.34% from Friday’s $633.91. This AMD stock prediction is a $757 bull case, a $623 base case and a $390 bear case. The bull is 19.8% above that close. The base is 1.4% under it, a dollar above the sell-side average. The bear is 38.3% below it.

The consensus is missing a simple point. The bear case from here is a multiple reset, not a missed quarter. S&P Global’s poll, via StockAnalysis on 5 October, puts the average target at $619.51 across 55 analysts, 1.9% under Monday’s close. AMD has already guided the September quarter to about $13 billion, plus or minus $300 million, and a non-GAAP gross margin of about 56%. Hitting that guide confirms a year the average target already prices. It does not open a new leg.

Key facts

• Monday 5 October 2026 regular-session close $631.75, down 0.34% from Friday’s $633.91. Session range $619.99 to $633.50. The 2 October intraday high was $645.46. Source: StockAnalysis daily history, retrieved 6 October 2026.
• Equity value about $1.031 trillion at Monday’s close on 1,632,475,042 shares outstanding as of 29 July 2026. Source: AMD Form 10-Q for the quarter ended 27 June 2026.
• Q2 revenue $11.536 billion, up 50% year over year. Data Center revenue $6.718 billion, up 107%. Non-GAAP gross margin 56% and non-GAAP diluted EPS $1.66. GAAP gross margin was 54% and GAAP diluted EPS was $1.38. Source: AMD Q2 2026 earnings slides, 4 August 2026.
• Q3 outlook from those slides: revenue about $13 billion, plus or minus $300 million; non-GAAP gross margin about 56%; non-GAAP operating expenses about $3.65 billion; other income about $55 million; non-GAAP tax rate 13%; diluted share count about 1.661 billion.
• OpenAI and Meta each hold a warrant for up to 160 million shares at $0.01. None had vested as of 27 June 2026, the same 10-Q says. Meta’s final stock-price threshold is $600, per AMD’s 8-K of 24 February 2026.
• Street average target $619.51, low $365, high $1,250, 55 analysts, Strong Buy. Fiscal 2026 non-GAAP EPS consensus $7.58. Fiscal 2027 consensus $15.58, on revenue of $50.94 billion and $88.23 billion. Source: StockAnalysis, citing S&P Global, 5 October 2026.
• Targets still above the price: Stifel’s Ruben Roy to $700 from $635 on 5 October; Wells Fargo’s Joseph Quatrochi reiterating $700 on 29 September; Rosenblatt’s Kevin Cassidy at $700 on 28 September; Bank of America’s Vivek Arya at $720 on 25 September. Same ratings table.
• World Labs: about $8.2 billion, all stock, expected to close by the end of 2026. Source: AMD press release, 28 September 2026.

What just happened, and why the obvious reading is wrong

Monday was a quiet session after a violent month. The stock opened at $630.61, traded down to $619.99, and closed at $631.75 on about 13.7 million shares, under Friday’s 19.3 million. From the 31 December 2025 close of $214.16, Monday is up 195.0%. From the 4 September close of $477.57 it is up 32.3%. The first close above $600 was $615.52 on 21 September. Friday’s $633.91 was the highest close in the series. Monday sat 2.1% under the $645.46 intraday high of 2 October.

The obvious reading is that Wall Street caught up. Stifel lifted Ruben Roy’s target to $700 from $635 and kept a Buy. TipRanks reported the change on X and said the analyst described AI infrastructure as still supply-gated, not demand-gated. Wells Fargo, Rosenblatt and Bank of America were already at $700 or $720.

The dates run the other way. Quatrochi reiterated $700 on 29 September, when AMD closed at $607.57. Cassidy maintained $700 on 28 September, when it closed at $607.87. Roy’s old $635 was already through the stock. The new $700 restores about $68. On 4 September that same target would have been 46% above the close.

The average is cleaner. At $619.51 it sits under Monday’s close, on a September mix of 39 strong buys, five buys, 11 holds and no sells. The poll low of $365 is 23.4 times 2027 earnings of $15.58. The high of $1,250 is 80.2 times that figure. Monday is 40.5 times 2027 earnings and 83.3 times the $7.58 consensus for 2026. World Labs, at about 13.0 million shares or 0.8% of the count, does not change that. The warrants do.

The bull case, with the maths

The bull case here is $757. It is not the $1,250 high, and it is not the $700 Stifel printed on Monday. Both steps use figures on the 5 October StockAnalysis page. The product is our estimate, not a company figure and not Roy’s target.

First, the multiple inside the $700. Divide $700 by the $15.58 consensus for 2027 non-GAAP earnings and the target is 44.9 times that year, against 40.5 times at Monday’s close. Bank of America’s $720, set on 25 September, is 46.2 times the same earnings. As reported by Investing.com and carried on Yahoo Finance, Vivek Arya, a Bank of America analyst, lifted that target from $620 on 30 times his calendar-2028 estimate, up from 27 times, after raising the 2027 and 2028 estimates by 2% to 3%. The firm put the server-CPU market at $61 billion in 2026 and $211 billion in 2030, with $180 billion from AI workloads.

“Agentic AI requires compute across the system, with CPU latency and task-completion speed increasingly important to GPU utilization and the end-user experience,” Arya wrote, in that 25 September account. The sentence is a demand argument. It does not say that 40 times 2027 earnings is cheap.

Second, the published 2026 earnings range. The average non-GAAP estimate is $7.58 and the high is $8.20. The ratio is 8.20 divided by 7.58, or 1.082. Apply it to Stifel’s $700 and the product is $757.26, rounded here to $757. The same ratio on the 2027 consensus is $16.85. At 44.9 times, that is the same $757. Against Monday’s $631.75 the gap is $125.25, or 19.8%. On the 29 July share count the equity value is about $1.24 trillion.

Three things have to be true together. The quarter lands inside the guide, 2026 earnings finish at the high of the range, and the market keeps paying about 45 times 2027 earnings. Data Center is the lever: $6.718 billion of the $11.536 billion June quarter, 58.2% of revenue, with segment operating income of $2.103 billion against a $155 million loss a year earlier. Client was $3.062 billion, Gaming $779 million and Embedded $977 million.

The $1,250 high is 80.2 times $15.58, close to the 83.3 times the market already pays for 2026 earnings, rolled forward a year. This bull case does not assume the multiple never compresses.

The $15.58 figure is per share on today’s count. The warrants had not vested by 27 June, so those shares sit outside the denominator. Full exercise adds 320 million shares, 19.6% more than 1,632,475,042 and 16.4% of the enlarged company. If 2027 sales of $88.23 billion already count the revenue and not the shares, the same earnings on the larger count cut each price here by about that fraction, unless the gigawatts earn more than the consensus assumes. OpenAI’s warrant runs through 5 October 2030 and Meta’s through 23 February 2031.

The bear case, with the maths

The bear case is $390. Twenty-five times $15.58 is $389.50, rounded to $390. That is $241.75 under Monday’s close, or 38.3%. On the 29 July count the equity value is about $637 billion. The stock would still trade at 51.5 times the $7.58 consensus for 2026, and at about 7.2 times 2027 sales of $88.23 billion, against about 11.7 times that sales figure at Monday’s $1.031 trillion. This is a de-rating, not a model of a broken year.

Nothing in $390 requires a missed quarter. Revenue can print inside $12.7 billion to $13.3 billion and the non-GAAP gross margin can stay near 56%. The only change is the multiple, from 40.5 times 2027 earnings to 25 times, a step above the 23.4 times inside the poll’s $365 low.

The level is a traded price. The last close under $390 in this series was $355.26 on 5 May 2026. The next session closed at $421.39. The lowest close in the file is $190.95 on 3 March 2026, with an intraday low of $188.22. A move to $390 gives back the run since early May. It does not retest March.

A clean quarter also fails to force the stock up, because the full-year consensus is already ahead of the guide. Six-month revenue through 27 June was $21.789 billion. Add the $13.0 billion Q3 midpoint and $34.789 billion is booked, with one quarter left. Reaching the $50.94 billion consensus means a December quarter of about $16.15 billion, 24% above the September midpoint, not the 13% step in the Q3 guide. The $619.51 average target already sits under the price on top of that year. Delivering it gets the holder back to a lower price. A Q4 of only about $14.7 billion, the 13% step, would put the year near $49.5 billion and bring the reset forward.

Margin does not bail the multiple out. Non-GAAP gross margin was 56% in June and 55% in March. A year earlier it was 43%, including $800 million of Instinct MI308 inventory charges. With those charges removed, the slides put the year-earlier non-GAAP margin at 54%. The clean gain is about two points, and the Q3 guide simply holds 56%.

Cash does not floor it. At 27 June, cash and short-term investments were $13.111 billion against $3.226 billion of debt. June-quarter free cash flow was $1.558 billion after $808 million of spending. Unused buyback authority of $9.2 billion, after $221 million spent in the half, retires about 14.6 million shares at Monday’s close, 4.6% of the warrant block.

What the tape and the filings actually show

The closes used here are Monday’s official print and the daily series saved from StockAnalysis, not a premarket indication and not Tuesday’s tape. The call is the table.

A bridge from AMD’s Q3 lines prices a midpoint quarter. Revenue of $13.0 billion at a 56% non-GAAP gross margin is $7.280 billion of gross profit. Subtract $3.65 billion of operating expense and operating income is $3.630 billion. Add $55 million and pretax profit is $3.685 billion. Tax at 13% leaves about $3.206 billion. On 1.661 billion diluted shares that is about $1.93. The figure is our estimate from the 4 August outlook, not an AMD earnings guide. A midpoint quarter is not a reason to move from 40 times next year’s earnings to 45 times.

Read the warrants from the filing. The 10-Q says AMD issued them in October 2025 and February 2026 to OpenAI OpCo, LLC and Meta Platforms, Inc., each for up to 160 million shares at $0.01. Vesting needs Instinct purchase milestones and stock-price targets. OpenAI also has stock-performance thresholds, and each tranche still needs technical and commercial conditions. None had vested by 27 June, and the warrants had no effect on the statements. Each customer intends to deploy up to 6 gigawatts, the first on Instinct MI450.

Meta’s February 8-K escalates the thresholds to $600 on the final tranche and ties the first tranche to shipment of the initial gigawatt. Monday’s close is above $600, so price no longer binds that last tranche. Shipment does, and a close above $600 issues nothing. At Monday’s price the 320 million shares would be worth $202.16 billion on paper, against $8.2 billion for World Labs. The 10-Q also records up to $4.1 billion of gross exposure at 27 June on partner data-center lease guarantees, on terms of up to 15 years.

The 4 August slides also record Anthropic taking up to 2 gigawatts of MI450 GPUs in Helios racks, and an expanded Microsoft collaboration on Helios and 6th Gen EPYC on Azure. Those are customers. On the pages used here they are not a third 160 million share warrant.

Nvidia’s late-September scenarios were already a fight about the multiple on earnings in hand. Broadcom’s case asks whether a guided AI revenue path gets paid in the share price. TSMC’s September revenue is the foundry tape under both, and Intel’s prediction is the other x86 name in Arya’s CPU argument.

What happens next

Three documents can move the multiple. None of them prints $757 or $390 on a named afternoon.

First is the September-quarter report. Wall Street Horizon lists Tuesday 3 November 2026, after the close, and marks the date unconfirmed. MarketBeat lists the same day as estimated. AMD’s news page, reviewed on 6 October, still led with the World Labs release and carried no Q3 date, so 3 November is the street’s calendar, not a company confirmation. Revenue inside $12.7 billion to $13.3 billion and a non-GAAP gross margin near 56% is the $623 base. The $757 bull needs the December-quarter guide: near $16.2 billion, the residual inside the $50.94 billion consensus, the street’s year is intact. Near $14.7 billion, only a 13% step, that consensus is high.

Second is the next 10-Q, which is the warrant check. The last official line is zero vested shares as of 27 June. If the Q3 note is still zero, a price above $600 was not enough. If a tranche has vested, the 19.6% shows up in the diluted share table. The $757 case survives full dilution only if earnings from those shipments exceed a consensus struck on today’s count.

Third, World Labs is supposed to close by the end of 2026, subject to regulatory approvals and customary conditions. Near Monday’s price the issuance is about 13 million shares. “Building the compute platforms for the next generation of AI requires a deep understanding of how models are evolving,” Lisa Su, AMD’s chair and chief executive, said in the 28 September release. Fei-Fei Li, then co-founder and chief executive of World Labs, said: “Joining AMD will give our team the resources and engineering depth to accelerate our research and help define the infrastructure needed for the next era of AI.” Both sentences can be true at $623. The research does not set the multiple.

Frequently asked questions

What is this AMD stock prediction?

Against Monday’s $631.75 close, the bull case is $757, the base is $623 and the bear is $390. The bull scales Stifel’s $700 by $8.20 over $7.58. The base is 40 times 2027 earnings of $15.58. The bear is 25 times that figure.

Why is the bear case a multiple reset, not a bad quarter?

The average target was already $619.51 on 5 October, 1.9% under the price, and the quarter has been guided since 4 August at about $13 billion and a 56% non-GAAP gross margin. Cutting the 2027 multiple from 40.5 times to 25 times produces $390 with earnings left at $15.58. The last close under $390 in this series was 5 May 2026.

Do the OpenAI and Meta warrants dilute the stock today?

Not yet. The Q2 10-Q says none had vested as of 27 June 2026. The two warrants cover up to 320 million shares at $0.01, 19.6% more than the 29 July count. Meta’s final price threshold is $600, which Monday cleared. Purchase milestones still have to be met.

Is $1,250 a usable bull case?

It is the high of the 55-analyst poll on 5 October, and it is 80.2 times 2027 consensus earnings. This prediction does not use it. The $757 bull is scaled from the $700 and $720 cluster at Stifel, Wells Fargo, Rosenblatt and Bank of America.

What is the next catalyst?

Calendars point to 3 November 2026 after the close, still unconfirmed or estimated on the listings reviewed here. The figure that matters is the December-quarter guide, set against about $16.2 billion of Q4 revenue inside a $50.94 billion year. The 10-Q is the next official word on vesting since 27 June.

This is not financial advice.