Author: Godot

NVIDIA released its FY 2027 Q1 financial report last night. Taking NVIDIA's financial report as an example, let's talk about how to interpret the financial report from the perspectives of Vivek Arya, an analyst at Bank of America BofA, and Mark Mahaney, an analyst at Evercore ISI.

Step 1: First, figure out what this company sells and who it sells to.

In other words, business model.

The same number can mean completely different things depending on the business model. A 50% revenue increase is normal for a software company, but not so normal for a shipyard, because a shipyard's capacity is physically limited. A 50% increase indicates either a price increase or that the company has squeezed out competitors' orders.

Once you understand the business model, you can further segment it.

First, why are customers willing to pay for the products this company sells? Is it because they have no other choice, because they are too cheap, or because they are easy to use?

NVIDIA has virtually no real alternatives. CUDA is the software-based AI standard, and switching chips is equivalent to wiping out the software investments of the past decade. This is its true competitive advantage.

Second, who are the customers? Are they a few large clients, or individual customers?

NVIDIA's financial report revealed that three direct customers accounted for 54% of its revenue, indicating a high concentration. Any order cut by any one of these customers would have a significant impact, which is a cause for concern among institutions.

If major cloud service providers targeting end users, such as Microsoft, Google, and Amazon, slow down their capital expenditures, it will impact NVIDIA's revenue.

Third, how does the company categorize its business?

This time, NVIDIA changed its classification criteria from Data Center/Gaming/Automotive/Professional Visualization to Data Center (further subdivided into Ultra-large Customers and ACIE)/Edge Computing.

A company's initiative to change its classification is an attempt to get the market to view it from a new perspective.

Combined with NVIDIA's AI investment portfolio and Jensen Huang's AI Factory concept, NVIDIA wants institutions to evaluate its value as a one-stop solution provider for AI factories, rather than just selling GPUs.

Step 2: Examine the quality of revenue growth

First, year-on-year, month-on-month, and acceleration.

NVIDIA's year-on-year growth rate has accelerated for three consecutive quarters. Despite its growing size, the growth rate is still accelerating, indicating that the inflection point in end-user demand has not yet been seen.

Second, where does the growth come from?

NVIDIA's data center growth was 92%, while edge computing only grew by 29%, indicating strong growth momentum but also risks due to excessive concentration of risk.

Hyperscale, a major client, and other clients saw their ACIE growth rates increase by 12% and 31% respectively quarter-over-quarter. This is a data point that is most sensitive to institutional analysts. The faster growth of ACIE indicates that the customer base is expanding, with demand spilling over from a few cloud giants to governments, enterprises, and AI startups around the world. This kind of growth is more sustainable.

Third, is the growth genuine or borrowed?

1) Accounts Receivable Turnover (DSO): NVIDIA's DSO this quarter was 45 days, even shorter than the 51 days last quarter. Customers are paying faster than before, demand is real, and customers are scrambling to get their orders.

2) Deferred Revenue: $3.117 billion, of which $1.7 billion were customer prepayments. Customers paying in advance, before the goods are even shipped, is the strongest evidence of genuine demand.

Why are these two metrics important? Because while revenue surged, accounts receivable also surged, and payment collection slowed down, meaning goods were sold but money wasn't received, potentially indicating unsold inventory. NVIDIA's healthy performance in both metrics suggests that 85% of its revenue growth is genuine.

Step 3: Look at the profit margin

Profits can fluctuate due to one-off events, but what truly reflects earning power is the profit margin, which is the percentage of profit to revenue.

1) Gross profit margin reflects pricing power

If you sell something for 100 yuan, how much is left after deducting direct costs? Hardware companies are considered excellent if they can achieve 30-40%, while software companies typically achieve 70-80%.

NVIDIA's ability to achieve a 75% gross profit margin as a chip manufacturing hardware company demonstrates that it is no ordinary hardware company. It possesses extremely strong pricing power.

Excluding the period last year when the US prohibited sales to China and inventory had to be scrapped, the gross profit margin jumped from 60.5% to 74.9%.

The year-on-year growth indicates that NVIDIA is able to pass on cost pressures to downstream customers amid rising HBM and advanced packaging prices from TSMC.

2) Operating profit margin reflects operating efficiency.

Operating profit margin is higher than gross profit margin by deducting operating expenses (R&D, sales, and administration). NVIDIA's operating profit margin this quarter was 65.6%, meaning that for every 100 units sold, after deducting all daily operating expenses, 65 units were left.

Furthermore, revenue increased by 85%, while operating profit increased by 147%. Profit growth far exceeded revenue growth; this is called operating leverage, indicating that the larger the company, the higher its efficiency, and the greater the profit generated from each additional unit of revenue. This is a characteristic of top-tier business models.

Step 4: Cash Flow

1) Operating cash flow and net profit

Operating cash flow is the amount of money actually received minus the amount actually paid out. NVIDIA's operating cash flow this quarter was $50.3 billion, and its net profit was $58.3 billion.

Why is cash flow less than profit? 15.9 billion of the net profit comes from the appreciation of equity investments. It's more profit on paper than actual cash received.

Excluding this portion, the cash flow generated by core operations matches the operating profit, indicating that the profit is genuine.

2) Free Cash Flow (FCF) = Operating Cash Flow - Capital Expenditures

FCF (Financial Cash Flow) is truly disposable income. NVIDIA's FCF this quarter was $48.6 billion, while capital expenditures were only $1.76 billion. Hardware investment was minimal, allowing most of the money to be used for stock buybacks, dividends, and ecosystem investments.

This is actually quite important, meaning that NVIDIA is not actually a capital-intensive company. That's right, it's not, which may contradict common perception.

For asset-heavy companies like TSMC, most of their profits are reinvested in factory construction, leaving limited funds for shareholders. NVIDIA, on the other hand, can freely use 90 out of every 100 dollars it earns. This explains why it was able to announce an $80 billion share buyback and a 25-fold increase in dividends.

Step 5: Examine the balance sheet

A balance sheet reflects a company's current financial situation and can also provide signals for the future.

in stock

NVIDIA's inventory has more than doubled year-over-year to 25.8 billion.

Raw materials: 3.8 billion → 6.6 billion

Work in progress: 8.8 billion → 9.9 billion

Finished Goods: 8.8 billion → 9.2 billion

Raw material prices have skyrocketed, while finished product prices have remained relatively stable. The company is stockpiling raw materials for the mass production of Vera Rubin in the second half of the year, not because sales are sluggish.

Supply commitment: How confident the company is about the future

NVIDIA's total supply commitments amount to $145 billion, including $119 billion in manufacturing capacity contracts, $30 billion in cloud service contracts, and $6 billion in other contracts. This is the money the company has already staked.

First, NVIDIA itself is very confident about demand over the next 1-2 years; otherwise, it wouldn't have signed so many contracts. Second, this is also the source of risk: if demand suddenly declines, these $119 billion in capacity contracts will become new impairment losses.

Step 6: Review management guidelines and conference calls to find out what management cares about.

NVIDIA gave a Q2 guidance of 91 billion, excluding revenue from data centers in China. Once China lifts its ban, it will be a pure boon.

The company reiterated its long-term commitment to generating $1 trillion in revenue from Blackwell and Rubin by 2025-2027. The fact that management dares to make this statement indicates an extremely high level of order stability.

The terms that Jensen Huang repeatedly mentioned in the conference call were the directions he wanted the market to focus on: Agentic AI, Tokens per dollar, Vera CPU, and the $200 billion new market.

Agentic AI and Vera CPUs demonstrate NVIDIA's ambition to seize the CPU market, highlighting the importance of CPUs in the agentic era.

Then look at how the Q&A section handles analysts' follow-up questions. "What is the new customer segmentation?" "Customer concentration?" "Is Vera CPU included in Rubin?" These are the points that institutions are most concerned about.

The management's ability to provide clear answers each time demonstrates that they were prepared for these questions and that their narratives can withstand scrutiny.

Step 7: Valuation

The price-to-earnings ratio (P/E) is the most commonly used, but it's not very useful. Semiconductor stocks, in general, appear to be overvalued. Next is Basic EPS * P/E.

DCF (Discounted Cash Flow) is the most rigorous way to estimate a company's value. It involves calculating how much free cash flow a company can earn each year in the future, discounting it back to its current value, and then summing these values ​​together.

However, future growth assumptions need to be made: will NVIDIA still grow by 20% or 50% in five years? And the discount rate (WACC) needs to be considered—whether to use 10% or 9%—a 1% change in the discount rate can lead to a $30-40 change in the target price. This is why there are differing target prices among institutions.

If long-term narratives such as "CPU renaissance," "token economics," and "AI factory capital return revolution" hold true, valuation multiples can be permanently inflated; conversely, they will collapse.

If a DCF model strictly discounts cash flows and does not take narratives into account, the advantage is that it resists bubbles, but the disadvantage is that it may underestimate true platform companies and thus miss out on all semiconductor stocks.

Therefore, ultimately, we need to consider the assumptions themselves and whether the promises can be fulfilled, and how likely they are to be fulfilled. If you believe that the demand for AI inference will continue to explode, and that Anthropic and OpenAI can grow tenfold, then an optimistic valuation is justified.

Conversely, if there are concerns about the Hyperscaler CapEx cycle and ASIC replacement, a conservative valuation is the right approach. This is also one of the most interesting aspects of the US stock market.