AMD breaks the trillion mark: What is it buying at 157x P/E?

NEXT PICK · Market Insights

AMD breaks the trillion mark: What is it buying at 157x P/E?

Monday, September 21, 2026

At the close on September 21, AMD was trading at $615.52, soaring 9.95% in a single day, with its market value surpassing $1.00 trillion for the first time—this company, which was still on the brink of bankruptcy in 2016 with a stock price of just two or three dollars, has officially joined the trillion-dollar club.

But if you think this bullish candlestick comes from AMD's own announcement, you're mistaken. The real trigger that day was Meta's AI intelligent app Muse, which topped the US iPhone free chart for three consecutive days.

The Philadelphia Semiconductor Index rose over 3% that day, Arm rose 13%, Intel rose 12%, and AMD's 9.95% was stuck in between—this was a street sentiment resonance, not AMD's personal show. The day's trading volume was about 1.8 times the three-month average.

So here's the question

How can a mobile app ranking that hasn't even contributed a single cent in revenue allow a company's market value to increase by nearly $100 billion in a single day?

Latest price $615.52 ▲ +9.95%

Total market capitalization $1.00 trillion ▲ A historic milestone of breaking one trillion

Price-to-earnings ratio 157.02 times ▲ +73.6% from the 200-day moving average

Three business segments, one lifeline: data centers have already taken over more than half

AMD is headquartered in Santa Clara, California, and its accounts are divided into three parts: data centers, client and gaming, and embedded. But the first part truly determines the stock price.

In the second quarter of fiscal year 2026, the company posted $11.5 billion in revenue for the single quarter, a 50% year-over-year increase and a record high. Of this, data centers accounted for $6.7 billion, a 107% year-on-year increase, more than doubling in just one quarter.

The performance of the other three is sharply differentiated

Client revenue was $3.1 billion, up 23% year-on-year; embedded revenue was $977 million, up 19%, while gaming revenue was $779 million, down 31% year-on-year.

The product matrix is well-developed—EPYC server CPUs and Instinct accelerators for data centers, consumer-facing Ryzen processors, professional graphics Radeon PRO, and embedded FPGAs and adaptive chips.

Customers are highly concentrated in the hands of complete machine manufacturers, public cloud service providers, system integrators, and board manufacturers. This structure is both its leverage and its most vulnerable link.

So what exactly makes AMD irreplaceable? The report gave it a score of 8 out of 10, and when you break it down, it's a four-layer crossover moat.

The first layer is brand and switching costs. EPYC's data center CPU revenue has hit record highs for five consecutive quarters; replacing it means servers must undergo complete machine certification, software optimization, and procurement contract renegotiations.

The second layer is product line synergy locking. From client to server, from AI acceleration to professional graphics, cloud providers can complete procurement all at once within a single supplier system, greatly reducing supply chain management complexity.

The third layer is the scarcity of advanced packaging. TSMC's CoWoS-L packaging capacity is expected to be tight Until 2028, the portion AMD has locked in is something competitors cannot buy in the short term.

The fourth layer is deep tie to TSMC—wafer allocation priority, capacity reserves, technical support. These hidden resources aren't written in contracts, but they determine who can scale up first and get a surge in volume.

If you had to remember AMD in one sentence: It's not a generator of the AI era, but the batch of turbines in the plant that are fully booked—limited capacity, so now everyone has to queue.

A price-to-earnings ratio of 157 times, and a stock price twice as high as any model

Looking at AMD's valuation snapshot: price-to-earnings ratio (P/E) of 157.02, corresponding to earnings per share of $3.92; P/S ratio of about 25.95; Enterprise value multiple (EV/EBITDA) as high as 189.86; Free cash flow yield (FCF yield) of only 0.49%.

Putting these numbers into the coordinate system makes it even more glaring. The historical average price-to-earnings ratio of the Nasdaq semiconductor sector is about 25 to 35 times, while AMD's premium is 5 to 7 times; The industry average price-to-sales ratio is about 8 to 12 times, with a premium of 2 to 3 times.

The most heartbreaking is that 0.49%. During the same period, the 10-year U.S. Treasury yield was about 4.5%—in other words, those who bought AMD today received less than one-ninth of the risk-free rate, and all returns depended solely on capital gains.

Third-party valuation systems offer a more direct conclusion

AMD's current price is 115.5% overvalued relative to its intrinsic value model. In the past 12 months, insiders have net sold about $365 million worth of stock, with no movement in the buying direction.

But the story of cash flow and the story of valuation are two different things.

Full-year free cash flow for fiscal year 2025 is $7.71 billion, up 154% from $3.04 billion in fiscal year 2024, with a free cash flow margin of about 22.3%. Quarterly free cash flow for Q1 2026 is $2.96 billion.

Operating leverage is also being realized

In Q1 2026, operating profit will be $1.48 billion, corresponding to $10.25 billion in revenue, with an operating margin of 14.4%; By Q2, it will rise to $1.99 billion, equivalent to $11.5 billion, with an operating margin of 17.2%, showing two consecutive quarters of growth.

Gross margin is the key variable. GAAP gross margin for the full fiscal year 2025 was 50%, rising to 54% by Q2 2026, with a non-GAAP margin of 56%. The company's Q3 guidance is also about 56% non-GAAP gross margin.

So AMD is entirely self-sustaining—R&D and capital expenses can be covered by operating cash flow, with no short-term financing pressure. The issue has never been whether it makes money, but how many years this price has been earning it.

The report's discounted cash flow model (DCF) applies three scenarios: pessimistic, neutral, and optimistic, with a discount rate of 10%. The result is: the reasonable value from all three discounted levels is significantly lower than the current stock price, and even the most optimistic tier is just over one-third of the current price.

So, does today's 9.95% increase count as positive news being realized? The answer leans toward yes. The agent computing power narrative, CoWoS scarcity, and the symbolic significance of a trillion-yuan market cap brought by Meta Muse have all been absorbed by the price all at once.

What illustrates the issue even more is this

The median target prices of over fifty Wall Street institutions were basically caught up or even surpassed by stock prices after Monday's bullish candle, while analysts' target price ranges are absurdly wide. When the median seller has been surpassed by the stock price, what you buy is no longer a discount but the narrative itself.

After graduating from 'Reverse Difficulties,' I am now studying the course 'Valuation Matching.'

AMD has completely moved beyond the concept phase and entered a mature growth phase. In fiscal year 2025, revenue is $34.6 billion, net profit is $4.3 billion, and quarterly revenue in Q2 2026 is set to hit a new high. The scalability of the business model no longer needs to be proven.

The product side also passed the verification stage

EPYC data center revenue has hit new highs for five consecutive quarters, with the Instinct series deployed in leading clients such as Meta, Oracle, and CoreWeave. TrendForce ranks AMD among the top three global fabless design companies.

The industry cycle is also on its side. In 2026, AI server shipments are expected to grow nearly 31% year-on-year, cloud vendor capital expenditures have surged by about 90%, and AMD is currently in the main wave of long-term AI infrastructure price increases, rather than in the exploratory phase.

Market consensus is written on the moving average

The 50-day moving average is $495.85, the 200-day moving average is $354.62, and the stock price is above both moving averages, with the 50-day moving average about 40% above the 200-day line, forming a typical accelerated upward structure.

But what's the other side of this structure? The stock price is 73.6% away from the 200-day moving average, and the 14-day Relative Strength Index (RSI) is at 73.1, entering overbought territory. The steeper the trend, the larger the gap at rebound.

The core conflict has shifted to a different topic

From "Can it be commercialized?" to "Does this valuation match the quality of growth?"

Having beaten Intel, still chasing NVIDIA, Broadcom is biting the edge of the market

AMD is not facing a single opponent, but a net of attacks from three sides.

On the front is NVIDIA. It dominates the AI training and inference market, with over 80% of AI server GPUs under TrendForce's profile. The developer lock-in created by the CUDA ecosystem is the deepest moat in the industry.

AMD's Instinct MI350 series offers cost-performance advantages in some inference scenarios and has secured a multi-year partnership with OpenAI—the first 1 GW MI450 deployment is expected to start in the second half of 2026, and Oracle has signed a supercluster order for 50,000 MI450s.

But the gap in software ecosystems is no joke. ROCm's support for PyTorch and TensorFlow is rapidly improving, but its maturity still lags about one to two years behind CUDA. Hardware can be matched by one generation; developers are used to needing a generation to make changes.

On the back is Intel. Xeon is recovering, with INTC's gains even surpassing AMD's on September 21. Intel's process nodes still face delay risks, but the global capacity layout of IDM 2.0 and the US CHIPS Act subsidies give it supply chain autonomy.

In the data center CPU sector, AMD's advantage is clear—its process technology stands on TSMC's 5nm/3nm side, and the share of x86 servers has risen from about 5% in 2020 to over 30%.

On the other side is Broadcom. Its network chips and custom AI-specific chips are eating into cloud providers' self-developed needs, with price-to-earnings ratios of only 35 to 40 times, indicating much more restrained valuations. Qualcomm is targeting Ryzen with Snapdragon X Elite on client platforms, but Windows on ARM penetration remains limited.

There is also a character with the most subtle identity

TSMC. It is both AMD's manufacturing lifeline and the hand that decides how much AMD can allocate — who allocated capacity and when it was allocated, AMD had no say in the end.

**Remember this pattern in one sentence

AMD is the conqueror in the CPU sector and the catch-up in the AI accelerator sector, while the key to its capacity is held by another company. **

Of the $500 to 600 billion pool, AMD is now only getting a small spoonful

Let's first look at the Total Addressable Market (TAM). AI servers are the most directly relevant segment—shipments in 2026 are expected to grow nearly 31% year-on-year. Assuming about 1.5 million units in 2025 and an average price of about $150,000, the market size is about $225 billion, and in 2026, about $295 billion.

Adding the data center CPU market annual size of about $80 billion to $100 billion, and the AI accelerator and GPU market about $150 billion to $200 billion annually. Combined, AMD's TAM is estimated to be in the $500 to $600 billion range.

Narrowing down to serviceable markets (SAM) by product line coverage: data center CPUs and GPUs are about $200 billion to $250 billion, client processors about $50 billion to $60 billion, gaming GPUs about $30 billion to $40 billion, embedded and custom chips about $20 billion to $30 billion, totaling about $300 billion to $380 billion.

The portion you actually get (SOM) is much more realistic

Data center CPUs are about $50 billion to $70 billion, AI accelerators about $20 billion to $35 billion, and clients about $15 billion to $20 billion, totaling about $85 billion to $125 billion.

Compared to the actual revenue of $34.6 billion in fiscal year 2025, penetration is still very low. The divergence point three years from now is also especially wide: under the optimistic scenario, revenue in fiscal year 2028 could reach $80 billion to $100 billion, in the neutral scenario $60 billion to $75 billion, and in the conservative scenario $45 billion to $55 billion.

Note the width of this range—the gap between the most optimistic and the most conservative is more than double. This is exactly where the 157x P/E ratio is most dangerous: it defaults to the optimistic tier.

Confidence of the bulls: triple validation occurring simultaneously

Those bullish on AMD aren't telling stories; they hold the reports that have already been implemented.

▲ Bull Case

Bears' Calculations: All the good news is already priced in

The bears' logic does not deny fundamentals; they question "how much margin of safety is left at this price."

▼ Bear Case

There are three key points to watch over the next two quarters

The next stock price rhythm will most likely be dominated by the following factors:

•	Q3 earnings report for early November 2026: guidance is about $13 billion, +41% year-on-year, the most direct disprofaction or verification of "whether the acceleration is sustained," with the highest impact
•	Q4 2026: AI accelerators and Radeon price increases of about 10% officially take effect. Whether gross margin can withstand TSMC's price hikes depends on this quarter—a double-edged direction
•	Second half of 2026 to 2027: The actual delivery progress of OpenAI's first 1 GW MI450 deployment and Oracle's 50,000-unit order will determine whether the AI accelerator share can truly rise
•	First half of 2027: If TSMC's CoWoS-L packaging expansion is implemented early, AMD GPU shipment bottlenecks will be lifted, which is positive but also weakens scarcity premiums
•	Full-year 2026: AI server liquid cooling penetration rose from 33% to 53%. The increase in the average price of high-end models drove up EPYC unit prices, affecting the neutral to positive side
•	Continue to observe: can the real computing power consumption of agent applications (such as Meta Muse) be converted into orders, not just download rankings?

After ten years and 300 times, the gains have already outpaced the fundamentals

In 2016, AMD's stock price was only two or three dollars, burdened with debt, and its products were lagging behind across the board. In 2017, Ryzen with Zen architecture was launched; in 2018, EPYC entered servers; and in 2019, Intel was the first to mass-produce the 7nm process—that was when Intel was stuck at 10nm.

2020-2021 was a breakthrough period for data centers, with revenue jumping from $9.76 billion to $16.43 billion, and stock prices pushing from around $50 to break $150 by year-end. During this phase, market expectations were basically realized.

2022 to 2023 was the toughest period. The PC cycle declined, with 2022 revenue of $23.6 billion down 4% year-on-year, and the stock price falling from a peak of $164 to the $60–80 range. But the MI300X was also a foreshadowing of this period.

The main rally will only begin after 2024: fiscal year 2024 revenue of $25.79 billion, +20% year-on-year; $34.6 billion in fiscal year 2025, +34.3% year-on-year; and $11.5 billion in Q2 2026 quarterly growth, +50% year-on-year, with market cap surpassing one trillion yuan.

Over the past decade, AMD's cumulative gain has been about 30,400%, from $10,000 to about $3.05 million—one of the most remarkable rebounds in U.S. stock market history. But revenue during the same period only grew about ninefold.

The huge gap in between is valuation expansion. It has been the main source of returns over the past decade, and also the least reliable part for the next decade.

There are also two deviations worth reviewing

The market overestimated ROCm's speed catching up to CUDA, and underestimated NVIDIA's pace of iteration. These two issues have yet to be corrected to this day.

⚠️ Risk Notice • Valuation is extreme, with a price-to-earnings ratio of 157 times plus a free cash flow yield of 0.49%, with a margin of safety close to zero • Technically, it is overbought, deviating 73.6% from the 200-day moving average, with a beta of 2.48 expanding the pullback • With a concentration of major clients, the capital expenditure decisions of a few cloud providers can directly determine the performance of quarterly revenue • There are already export control precedents, with net MI308-related expenses of approximately $440 million for fiscal year 2025 • With rising costs, TSMC's wafer and memory price increases could squeeze 54% of gross margin if transmission is not smooth • Sentiment is crowded, with semiconductor bulls having been the most crowded trading globally for four consecutive months, with clear negative signals

🟡 Neutral The fundamentals have been confirmed, but the price has already cashed in its victory over the next three years.

💬 Discussion Now is the time to manage positions, not chase. Live buy and sell positions, post AMD to the official account.

Data source

•	Source: NextPick real-time market snapshot + AMD Investor Relations official website earnings announcements + SEC disclosure documents + TrendForce industry data + mainstream financial media public reports, data as of the US East Coast close on September 21, 2026.