NVIDIA with a P/E ratio of 27.88 times—expensive or cheap

NEXT PICK · Market Insights

NVIDIA with a P/E ratio of 27.88 times—expensive or cheap

Monday, August 31, 2026

A company with a market value of $5.33 trillion posted $96.22 billion in revenue in a single quarter, more than doubling year-on-year, with a net profit of $59.688 billion and a net margin as high as 62%—yet when such a financial report were delivered, its price-to-earnings ratio was only 27.88 times, cheaper than many low-growth consumer stocks.

Even more intriguing, management has just given the first-ever forecast in history: revenue growth for the next fiscal year is about 70%, and it is clearly stated that this figure is not the upper limit on demand, but the upper limit on supply.

So here's the question—a story of both cheap and high-growth prices. Why has the stock price only hovered around $220 over the past month, always just shy of the 52-week high of $236.54?

Latest price $220.78 ▲ +1.48%

Single-quarter revenue $96.22 billion ▲ Year-on-year +106%

Price-to-earnings ratio 27.88 times — Net profit margin: 62%

It is no longer selling graphics cards; it is selling entire computing power factories

Looking at the latest quarter's revenue structure, the answer is quite striking: of the $96.22 billion total revenue, the data center segment alone contributed $89 billion, a year-on-year increase of 117% and a quarter-on-quarter increase of 18%. The remaining gaming graphics cards, professional visualization, and automotive platforms combined are less than one-tenth of the total market.

In other words, those still interpreting NVIDIA as a "graphics card company" in the market are looking at a company that no longer exists. It is now positioned as a data center-scale AI infrastructure provider—not selling a single chip, but a complete system from GPU to interconnect, to cabinet, to software stack.

Where does the irreplaceability of this system come from? The first layer is CUDA. Nearly twenty years of accumulation have made it the default underlying backend for mainstream deep learning frameworks like PyTorch, TensorFlow, and JAX. Millions of developers worldwide have muscle memory built on libraries like cuDNN, cuBLAS, and TensorRT.

The second layer is interconnection. NVLink and NVSwitch form a high-bandwidth, low-latency architecture that enables the 72 GPUs in the NVL72 cabinet to be fully interconnected, treating the entire cabinet as a giant chip for scheduling. AMD's Infinity Fabric and Intel's CXL still have generational gaps in bandwidth and ecosystem maturity.

The third layer is the scale itself. A 75% gross margin and $7.054 billion in R&D investment in a single quarter means it can continuously iterate with intensity others cannot, while also holding priority rights over TSMC's advanced packaging capacity and SK Hynix's HBM capacity.

So perhaps a more fitting analogy is

NVIDIA isn't selling engines, but entire railroads with already laid tracks, signaling systems, and even a driver training manual. Of course, you can build a more fuel-efficient engine yourself, but on which track will you run it?

A price-to-earnings ratio of 27.88 times—is it a discount or a trap?

Let's put the static valuation anchor on the table first

Stock price $220.78, earnings per share over the past twelve months $7.92, price-to-earnings ratio 27.88. Net profit growth for the same period was 126% year-over-year, and dividing the two, PEG is only 0.22—by textbook standards, this is a sign of deep undervaluation.

But another set of figures gives the exact opposite reading. With a market value of $5.33 trillion and $300 billion in revenue over the past twelve months, the price-to-sales ratio remains in the 18-times range; The price-to-book ratio exceeds 20 times, far above the historical midpoint of the semiconductor industry.

For the same company, using profit metrics is cheap, but using assets and revenue metrics is expensive—this kind of rift is itself the source of all the contradictions. It means the market isn't questioning how much NVIDIA is making now, but rather pricing "how many more years this rate of profit can last."

If we follow the Q3 guidance of $108 billion in revenue and the current profit pace, this fiscal year's earnings per share are very likely to enter double digits, and the forward price-to-earnings ratio for the current stock price would be pushed to just over 20 times. So why doesn't the market give it a higher multiple?

The cash flow side is almost flawless. In the first half of this fiscal year, operating cash flow was $74.421 billion, with $22.443 billion in cash on hand plus $76.926 billion in securities, totaling nearly $100 billion, with no interest-bearing debt pressure.

What illustrates this even more is the strength of capital returns

In just one quarter, the company repurchased $19.732 billion in stock and paid $6.047 billion in dividends, with a quarterly dividend yield of $0.25 per share. A company still expanding at ultra-fast speed with the ease to return over $20 billion in cash to shareholders is almost unprecedented in semiconductor history.

Mature, yet still accelerating—this is a contradictory entity

By any traditional measure, NVIDIA is in a mature stage: last fiscal year, full-year revenue was $215.9 billion, net profit was $120.1 billion, products have long been commercialized, and demand comes from real capital expenditures by hyperscale cloud providers, not concept narratives.

But while mature companies typically see growth decline, NVIDIA is climbing upward—year-over-year growth in the past four quarters has clearly accelerated, with the latest quarter at 106% and data center segments at 117%. It combines the cash flow of mature companies with the slope of growth companies, which is a very rare combination.

Will the market accept this judgment? The technical answer is "cautiously accepted." The current stock price is above the 50-day moving average at $208.42 and the 200-day moving average at $195.83, indicating a complete trend structure; but the upper boundary of the 52-week range between $164.07 and $236.54 has repeatedly blocked it several times.

Moving averages are supporting the bottom, previous highs are pressing against tops—the market is voting with price: I believe in your performance, but I'm not sure how many times I should pay for it.

80% of the market share is being eroded simultaneously by four directions

On the front lines, AMD is the only rival with a hardware spec that can still negotiate. The Instinct MI300X/MI350 series are competitive in memory bandwidth and HBM capacity, with the ROCm ecosystem continuing to catch up. Their pricing is more aggressive for major clients, but their market share in AI servers is still only 10% to 15%, far from NVIDIA's over 80% market share.

Intel's Gaudi 3 is positioned in the cost-performance market, dragged down by process yield, with an almost negligible share. The real threat has never been on this front.

The real variable comes from the customers themselves. Google's TPUs have evolved to the Ironwood generation, and Anthropic has publicly stated it has deployed over one million for inference workloads—this is the first time a custom ASIC has reached a seven-digit order of magnitude at a single customer. Amazon is promoting Trainium while clearly stating that "customers want better value for money."

Meanwhile, Broadcom, as the foundry manufacturer and mass producer of co-packaged optical switch chips for Google's TPU, is pincering NVIDIA's InfiniBand from the network interconnection side. Even OpenAI has claimed that its self-developed Jalapeno processor outperformed NVIDIA's current product line in tests (although this comparison does not include the soon-to-be-released Vera Rubin).

Thus, the true shape of the moat became clear

NVIDIA is almost unshakable in training, because training requires the strongest general-purpose computing power and the most mature software stack; But inference is being gradually eroded by custom chips, because inference competes on performance per watt and unit cost.

So what will be the real winner in the next five years? It's not whose chip is faster, but how much inference will ultimately play in the AI's computing power consumption structure—every percentage point increase in this proportion lowers NVIDIA's market share ceiling by an inch.

The ceiling hasn't been seen yet, but the market share ceiling has already appeared

Looking at the total addressable market, this sector is still expanding: major global cloud providers continue to sharply increase capital expenditures, AI server shipments maintain a growth rate of over 30%, cabinet power consumption has broken several hundred kilowatts, pushing liquid cooling penetration from one-third to over half, and the physical infrastructure of entire data centers is entering a mandatory upgrade cycle.

NVIDIA's own guidance is the best footnote—management expects revenue growth of about 70% next fiscal year, while the CEO says demand is "well over 70%." The real bottleneck is supply: TSMC's advanced packaging capacity is tight, memory chips are also in short supply, and the supply bottleneck will persist at least until the end of that fiscal year.

Markets that can be filled by orders but cannot fill capacity are, in essence, seller's markets. But conversely, this also means revenue growth is capped by physical capacity, making it impossible to hedge cost increases by selling more money.

The story on the market share side is the opposite. Currently, a market share above 80% is almost the industry's limit. As Google, Amazon, and Meta scale up their self-developed chips, most industry research groups expect market share to fall to the 70% to 80% range in the next two to three years—market growth and market share shrinking, these two curves are happening simultaneously.

The long leader holds three cards, each with hard data

Bullish on NVIDIA requires no imagination, because every support point can be found in the latest financial reports and management's statements:

▲ Bull Case

Bears do not deny earnings; they focus on gross margin and denominator

The brilliance of bears lies in never trying to argue that NVIDIA's business is bad—they argue that the best part may already be in the price, and the next worsening part hasn't yet happened:

▼ Bear Case

In the next four quarters, three timelines will move simultaneously

What will truly change the pricing logic of this stock next isn't macro or sentiment, but several clear timing windows for cashing out:

•	Next quarterly earnings report: Verify whether the $108 billion revenue guidance and 74.0% gross margin are met simultaneously; the direction is positive, but gross margin remains a minefield
•	Q4 of this fiscal year: Gross margin bottomed out at 71% to 72% as directed. If it holds or exceeds expectations, it will be an important sentiment turning point
•	Vera Rubin (R100) architecture will be released and see a surge in volume this year: the cost-performance performance of the new generation directly determines whether it can block self-developed ASICs, with a positive impact
•	About 70% of revenue growth guidance for the next fiscal year is being delivered quarter-on-quarter: this is the first time management has made a one-year advance commitment, and any shortcomings will be amplified
•	Advances $100 billion cooperation with OpenAI at a data center in Ohio: The deployment of gigawatt-level computing power is solid demand but will also intensify controversies over circular financing, affecting both directions
•	Price trends in memory and advanced packaging supply chains: If HBM price increases exceed expectations, it will directly erode already declining gross margins, resulting in a negative impact
•	The options market saw $18.5 million in large short orders, 15 times larger than the bullish side—against the backdrop of upward guidance and rising stock prices, this feels more like heavy long positions buying insurance rather than short betting on a crash

From 130.5 billion to 215.9 billion, the fundamentals paid off, but the stock price remained stagnant

Rewinding two fiscal years, NVIDIA's fundamentals trajectory is almost textbook-level: the last full fiscal year had revenue of $215.9 billion, a 65% year-over-year increase, net profit of $120.1 billion, and earnings per share of $4.90; and the previous fiscal year, revenue of $130.5 billion, net profit of $72.88 billion, and earnings per share of $294.

In other words, over two years, the absolute value of earnings per share has more than doubled, while the price-to-earnings ratio has dropped from over thirty times back then to today's 27.88 times—fundamentals outperforming the stock price, valuation multiples shrinking.

The real turning point has been concentrated in recent months. At the end of July, chip stocks collectively pulled back, and in early August, reports about a $500 billion AI infrastructure financing plan sparked concerns about concentration risk, causing the company's market value to shrink by about $130 billion in a single day; Immediately after, the late August earnings report added back over $400 billion in market value.

What does this intense back-and-forth indicate? It shows that the market is no longer divided over NVIDIA's performance; all disagreements revolve around "how long is left in this AI capital spending cycle"—and the 52-week range between $164.07 and $236.54 is a pricing expression of this divergence.

⚠️ Risk Notice • Gross margin has entered a downward channel guided by management, with the fourth quarter of this fiscal year expected to bottom out at 71% to 72% • Memory and advanced packaging supply is tight, and rising costs cannot be offset by increasing shipments • Cloud vendors are accelerating the replacement of self-developed ASICs on the inference side, with deployments at the million-chip level already emerging • China's data center revenue expectations have reached zero, and export controls may tighten further • Massive investment ties with major clients have raised questions about circular financing, and valuation premiums could be revalued at any time • The overall capital outflow of the AI computing power sector continues; even if individual stocks' fundamentals are stable, they will be squeezed and dragged down by sector-specific valuations

🟡 Neutral Growth is certain, but the decline in gross margin is approaching in tandem with the diversion of self-developed development, making valuations neither expensive nor cheap

💬 Discussion Now is the time to hold instead of chasing highs. Its real-time buy/sell levels can be viewed via NVDA on the official account

Data source

•	Source: NextPick real-time market snapshot + NVIDIA official earnings and investor relations disclosures + SEC filing documents + mainstream financial media public reports, data as of August 31, 2026.