Equity Research
Rating: BUY | 12-M Price Target: $620 | Upside: +31% | Current Price: $473
Rating and Price Target:
We have a Buy rating and $620 12-month price target on AMD (about 31% above the current $473 price target). Our price target is mostly in line with Wall Street analysts’ expectations already, which we see as consensus confirming our views rather than disagreeing with us.
Here’s why we think AMD is a buy: For years, AMD has been a business focused on selling computer processors (CPUs) and graphics processing units (GPUs) for gamers. But now it’s rapidly becoming a company that sells CPUs and GPUs for artificial intelligence (“AI”) data centers. This puts it in direct competition with Nvidia. Already, in Q4 2023, AMD’s “Data Center” segment accounted for 58% of revenue and grew 107% annually. Major customers like OpenAI, Meta, and Anthropic have signed contracts to buy AMD AI chips for multiple years, strengthening our confidence in AMD’s growth outlook. However, Nvidia controls the vast majority of the AI chip market today. And they have big advantages in software that are likely to persist. We believe AMC’s stock price reflects management’s guidance for solid, realistic growth. Not stellar growth. This gives us some upside if they can exceed expectations.
our view: One question we think investors are asking themselves is whether AMD will be able to continue increasing its profit margins by selling more AI chips. So far, company results suggest it can – but it hasn’t been proven yet over multiple quarters.
Company Overview and Business Segments
AMD generates revenue from four segments. Data Center comprises EPYC server chips (used for the "brains" in cloud computing) and Instinct AI accelerator chips (used to train and run AI models). Client includes Ryzen chips found in laptops and PCs. Gaming is made up of Radeon graphics cards and specialized chips AMD designs and sells to Sony and Microsoft to go in their PlayStation and Xbox consoles, respectively. Embedded consists of chips that were part of company Xilinx, which AMD acquired in 20|22, that are used in industrial systems, airplanes, military technology, and cars.
Feed powering all four segments is shared technology AMD uses to build its chips. AMD uses what's called a "chiplet" design. Basically instead of creating one large piece of silicon that would be used in a GPU or CPU, it creates smaller chips that it stitches together. It's cheaper this way and allows AMD to be flexible about which chiplets it uses in which chips. AMD has two different graphics chip designs it uses: CDNA (optimized for AI and data centers) and RDNA (optimized for gaming), as well as a software platform called ROCm, which is designed to compete with Nvidia's CUDA software that most AI developers use today. AMD also just launched Helios, its first-ever full server rack offering that bundles GPUs, CPUs, and networking hardware/software to go direct after Nvidia's highest-margin rack-level AI products.
Our view: We think AMD's chiplet design gives it the longest long tail. Not only can it be used across all four of AMD's segments, but it also gives AMD a real cost advantage over Intel, which is still years away from catching up with its own manufacturing technology.
Recent Financial Performance
AMD has visibly accelerated its growth over the past four quarters. Revenue increased from $9.25 billion in Q3 2025 to $11.54 billion in Q2 2026, or about 50% higher than the same quarter a year earlier. Gross margin also increased, rising from roughly 52% to almost 54% (gross margin is the percentage of revenue remaining after deducting production costs). Data Center revenue more than doubled during the same time frame, increasing from $4.34 billion to $6.72 billion. Even the Embedded segment, which includes Xilinx, grew revenue by 19% year over year in the latest quarter (best growth rate in 3 years), indicating that an inventory overhang in that business has largely been resolved.
For Q3 2026, AMD provided revenue guidance of roughly $13 billion, give or take $300 million. This would represent year-over-year growth of about 41% and is above analysts' expectations. Management also guided for gross margin to be near 56%. Looking ahead, Wall Street currently expects AMD's revenue to jump from about $51 billion this year to nearly $89 billion in 2027. Analysts are modeling EPS nearly doubling over this time period. Note: The large increase in earnings is a bit suspicious and should be compared to updated analyst estimates rather than the figure above.
In mid-2025, AMD incurred approximately $800 million in charges due to new U.S. government export restrictions on selling certain AI chips to China. This was recorded as a one-time charge and should not be interpreted as weakening business.
Our view: We like that growth is coming from both server chips and AI chips, rather than just one segment of the business. However, we are cautious that cash generation decelerated somewhat in the quarter due to AMD investing significantly more in building out new AI systems. This is the direct cost of this growth effort so it will be something to monitor closely.
Growth Drivers and Strategic Bets
AMD has successfully moved beyond its previous MI300 AI chip generation to MI325, then MI350, and now has its newest MI400 and MI450 chips available. These will be sold alongside its servers, CPUs, and networking hardware as its new "Helios" rack product, competing directly with Nvidia's newest Blackwell systems and soon-to-launch Rubin systems. AMD has already announced several large, multi-year commitments to the new hardware: OpenAI signed a deal to purchase enough AI chip generation capacity to run 6 gigawatts worth of computing hardware (for comparison, this is approximately how much power it takes to run a large city), Meta followed that up with a commitment for another 6 gigawatts, and last month Anthropic signed a deal for 2 gigawatts. AMD also invested up to $5 billion into Anthropic directly. Both Microsoft and Oracle have also signed on as customers.
On the software front, AMD is doubling down on ROCm, its open-source alternative to Nvidia's CUDA software framework. AMD recently added tools that automatically generate and optimize code for its chips, which it says helped improve software test pass rates from 37% to 93% in just a few months' time. Those tools have reportedly allowed AMD to close most of the gap with Nvidia's hardware for AI inference (the process of running an already-trained AI model), though AMD lags Nvidia far more substantially when it comes to training purposes. Training new AI models from scratch is something AMD's hardware struggles with significantly, and remains a blind spot for the company.
AMD's EPYC server chips achieved a record 34.5% unit share of the market against Intel's Xeon server chips, up from 27.3% just one year ago. AMD's desktop CPU share also hit a new record high. AMD is winning server CPU share from Intel again, and it's not close.
The clear threat to AMD's continued success here is Nvidia, which still overwhelmingly dominates the AI chip market with its GPUs and estimates put its share between 75% and 81% of the total market. The longer-term threat is that large cloud companies are training more of their own AI chips than they buy from either AMD or Nvidia (think Google's TPU, Amazon's Trainium, Microsoft's Maia). That market is said to be growing as fast as 44.6% per year. We believe that could contract the total available market over the long term. To that end, AMD purchased a small, specialized AI chipmaker named Taalas last month.
Our view. With large companies like OpenAI, Meta, and Anthropic all signing these large deals, there's real confidence in the staying power of this growth story. These companies aren't just hoping for success, they are paying for it up front. We have more doubts around how quickly AMD's software ecosystem will catch up to Nvidia's, as there is still a substantial gap when it comes to training new AI models, and we've seen a wide range of time expectations from our different sources.
Capital Allocation and Balance Sheet
AMD has strong balance sheet with little risk. It ended the last quarter with roughly $5.1 billion in cash versus $4.3 billion in total debt. Total shareholder equity was $67.2 billion, so debt makes up a tiny percentage of the company's total value. Free cash flow, or the cash generated by the business, has been between $1.6 billion and $2.6 billion per quarter for the last four quarters. AMD does not offer a dividend. The company buys back a small amount of stock most quarters, but reinvests most of its cash into growth initiatives, particularly R&D and acquisitions.
Capital expenditure, or spending on new equipment and facilities, tripled from $222 million to $808 million per quarter over the last year. This is to build out new AI chip production and packaging capacity. AMD has made several large acquisitions over the last few years. The biggest was Xilinx ($49 billion acquired in 2022) for its Embedded chip business. Another example was ZT Systems ($4.9 billion), so AMD could build entire server systems (not just chips). The company also bought AI chip startup Silo AI for $665 million. The company is buying AI chip startup Taalas, likely later this year.
Our view: Led by CEO Lisa Su and CFO Jean Hu, we believe this spree of acquisitions shows the company is singularly focused on building a comprehensive AI systems business right now. That won't return cash to shareholders via dividends or aggressive buybacks. This should allow AMD to expand its total addressable market over time. But it does create short-term costs and risk as it integrates new businesses. That could weigh on margins in the near term.
Competitive Benchmarking
Keep in mind that AMD's P/Es (price divided by profit per share) is high partly because profits were temporarily depressed by the aforementioned export-related charge. So this comparison is making AMD look "expensive" than it really is.
Our view AMD trades more like Broadcom (another company many view as a winner in AI infrastructure) than it does like a traditional chip company. We believe that premium is only justified if AMD's operating margin (significantly below Nvidia's and Broadcom's today) rises significantly as those AI chip sales ramp.
Moat and Risk Assessment
AMD's two durable competitive advantages are its chiplet manufacturing strategy and increasing server CPU share, as well as existing relationships supplying chips for Sony and Microsoft game consoles that cannot be easily disrupted by competitors. Its software moat with ROCm is tangible but nascent and lagging Nvidia's CUDA software, particularly for training new AI models.
The biggest risks, in order of severity are: Nvidia remaining dominant with strong software lock-in; customer concentration with just a few large customers driving AI chip growth; normal cyclicality in the PC and gaming markets unrelated to the AI play that can impact results during down years; execution risk surrounding manufacturing and packaging of these advanced chips at scale relying on partner TSMC; potential for increased U. S. government restrictions on selling advanced chips to China; and slimmer margins on AI chips vs. Nvidia's due to AMD's software and additional services not yet being as valuable to customers.
Our view: We feel pretty good about AMD's manufacturing and server chip advantages as they are backed up by multiple quarters of actual, reported market share gains. We are not as confident that AMD has a software moat established just yet, and view customer concentration as the risk most likely to impact the stock should one of those three large AI customers reduce spending.
Valuation
To compare AMD's valuation to its peers, let's look at price-to-earnings (P/E) based on trailing twelve-month profit. At ~121x, AMD's P/E ratio is significantly higher than Nvidia's (32.9x), Qualcomm's (18.4x), and Broadcom's (61.4x). However, this is distorted by AMD's one-time export-related charge I mentioned above. If we use this year's projected profit instead, AMD's P/E comes down to ~62x. And if we use next year's projected profit, it's ~30x – which doesn't look quite as excessive if those growth expectations pan out.
When you compare AMD's valuation to its own history, you'll notice that shares actually trade at a lower valuation than they did in 2023 and 2024. This is because AMD's profits have grown faster than its stock price. We think AMD is fairly valued to slightly undervalued today because its average analyst price target already assumes continued good execution, not an ultra-positive outcome. For the stock to move much higher from here, AMD will likely have to beat expectations on things like profit margins and taking market share from competitors, not just meet estimates for revenue and profit growth.
Our view: We see AMD's valuation as priced for its AI chip business to do well over the next several years. Shares already reflect a lot of good news happening, so if AMD is going to continue moving higher from here, it'll likely have to exceed expectations, not just meet them.
Bull, Base, and Bear Scenarios
