Nvidia: The Numbers Are Still Extraordinary. The Financing Around AI Is Getting Stranger.

The Kapital · US Stock Analysis · 20 August 2026

Nvidia’s operating results are still doing something almost no company of this size has ever done. First-quarter fiscal 2027 revenue reached $81.6 billion, up 85% year over year, while Data Center revenue climbed to $75.2 billion. The company guided to roughly $91 billion for the following quarter even without assuming China data-center compute revenue. The uncomfortable part of the story is no longer demand today. It is the increasingly complex financial ecosystem being built to keep tomorrow’s AI infrastructure spending growing.

I have spent much of the AI boom trying not to make the classic mistake of calling an expensive stock overvalued simply because the multiple is high. Nvidia has repeatedly grown into numbers that looked unreasonable a few quarters earlier. Its revenue growth, margins and cash generation have justified a level of optimism that would be reckless for most companies.

But valuation discipline becomes more important, not less, when a company is this successful. The market is no longer asking whether AI infrastructure spending exists. It is asking how long the spending can compound, how much of it produces attractive returns for customers and whether the financing structure around that spending is becoming a source of hidden fragility.

The operating machine is still exceptional

Nvidia reported fiscal first-quarter 2027 revenue of $81.615 billion, an 85% increase from the prior year. Data Center revenue was $75.2 billion, up 92%. GAAP gross margin reached 74.9%. GAAP operating income was above $53 billion.

Those numbers deserve perspective. Nvidia is generating a quarterly revenue base that many global industrial companies do not reach in a full year, while retaining gross margins associated with elite software businesses. That combination of physical product scale and software-like economics is the core reason the company has become so valuable.

The next-quarter outlook was equally aggressive: approximately $91 billion of revenue, plus or minus 2%. Management explicitly said the guidance did not assume data-center compute revenue from China. That means the underlying demand in approved markets was strong enough to support another large sequential step without a business segment that once mattered materially.

The most important number is still gross margin

Revenue growth gets the headlines, but gross margin tells me more about competitive power. A company growing 80% with 75% gross margins is not merely benefiting from a cyclical shortage. It is capturing a large share of the economic value created by customers.

Nvidia’s moat is broader than the GPU. CUDA, networking, libraries, systems design, developer familiarity and a rapidly evolving product roadmap make switching costly. Blackwell and Rubin are not isolated chips; they are parts of an architecture that connects compute, memory and networking.

This is why custom silicon from Google, Amazon and others does not automatically break the thesis. Hyperscalers can build internal accelerators and still buy enormous amounts of Nvidia hardware for frontier models, flexible workloads and customers that want the Nvidia ecosystem.

The real threat is not a single competing chip. It is a gradual change in the bargaining relationship between Nvidia and its largest customers as they become more capable of shifting workloads to internal hardware.

The financing ecosystem is becoming part of the analysis

Reuters reported on August 20 that investors are increasingly scrutinizing the circular nature of AI infrastructure financing. The concern is not fictional. The largest technology companies are investing in AI developers, cloud providers are financing data-center buildouts, semiconductor suppliers are supporting projects and private capital is funding infrastructure whose economics ultimately depend on future AI revenue.

One example cited by Reuters is Nvidia’s guarantee of up to $105 billion connected with an OpenAI data-center project in Ohio. A guarantee is not the same thing as an immediate cash expense, and it does not mean the project will lose money. But it changes the risk map.

If Nvidia supports financing for customers or ecosystem partners that then buy Nvidia equipment, reported demand can remain economically real while becoming financially intertwined. That does not make the revenue illegitimate. It means investors should ask who ultimately carries the risk if the end customer fails to generate sufficient returns.

Why circular financing can be rational

There is a benign interpretation. AI infrastructure is capital intensive, and the ecosystem has a coordination problem. Chip supply, power, data centers and model development all need to expand at roughly the same time. A highly profitable company like Nvidia can help unlock projects that would otherwise be delayed by financing constraints.

If those projects generate strong returns, ecosystem financing can accelerate the growth of the entire market. This is not fundamentally different from industrial companies helping finance equipment purchases, cloud providers giving credits to startups or payment networks subsidizing adoption.

The dangerous version appears when financing begins to create demand that would not exist under normal economic discipline. If projects are built because capital is cheap rather than because expected AI revenue supports the cost, future write-downs become more likely.

That distinction will probably define the next phase of the AI cycle.

The hyperscaler math still looks strong – for now

Amazon, Microsoft, Alphabet and Meta continue to report rapid cloud and AI demand. Capacity constraints have repeatedly appeared in management commentary. This supports the idea that customers are not buying accelerators merely to stockpile them. They are deploying infrastructure into real workloads.

The more difficult question is return on invested capital. A hyperscaler can grow cloud revenue 25% or 35% and still destroy value at the margin if each additional dollar of growth requires disproportionately more capital expenditure.

I therefore watch the ratio between cloud growth and capex growth. If capex continues rising dramatically while cloud growth decelerates, Nvidia’s customers may eventually become more price-sensitive. If cloud growth remains strong and operating margins hold, the investment cycle can last much longer than skeptics expect.

China has become an option rather than the base case

Nvidia’s guidance excluded data-center compute revenue from China, which is both a limitation and a source of optionality. Export restrictions have constrained one of the world’s largest technology markets. If future rules allow Nvidia to sell competitive products there, the revenue opportunity could be significant.

But I do not include a large China recovery in my base valuation. Policy can change quickly, product requirements can be tightened, and domestic Chinese accelerator ecosystems are improving. I prefer to value China as upside optionality rather than money the company is entitled to earn.

The valuation question cannot be answered with a historical P/E

A business growing 85% cannot be valued using the same multiple as a mature semiconductor company growing 5%. The problem is that no company can compound at Nvidia’s current rate indefinitely. At some point the law of large numbers becomes the central variable.

If annualized revenue exceeds $350 billion and then grows 25%, Nvidia adds almost $90 billion of revenue in one year. Maintaining 50% growth would require additions that are almost unimaginable by historical semiconductor standards.

My model therefore assumes rapid deceleration even in the bull case. The question is not whether growth slows. It is whether growth slows from 80% to 35% while margins remain exceptional, or from 80% to 10% because customers digest excess capacity.

My three valuation worlds

In my bear case, AI infrastructure enters a digestion period. Hyperscaler capex growth slows sharply, custom silicon gains share, and Nvidia’s gross margin falls toward the mid-60s. Revenue still remains enormous, but the market re-rates the company as a cyclical hardware leader rather than a near-monopoly platform. I would value the equity around $2.3–2.8 trillion.

In my base case, AI inference expands across software, enterprise and consumer applications. Nvidia continues leading frontier training and networking while custom chips take specific workloads. Growth normalizes toward 25%–30% over several years and gross margin stays around 70%. I can justify roughly $3.5–4.2 trillion.

In my bull case, inference demand becomes a new utility-like compute layer, agentic systems multiply token usage and Rubin extends Nvidia’s performance lead. Revenue continues compounding above 30% for longer than the market currently models. Then $5 trillion or more can be defended.

These are intentionally broad ranges. When a company’s future depends on an infrastructure market that is itself being invented in real time, precision becomes theater.

What the bond market is telling Nvidia investors

Technology shares have recently reacted sharply to higher long-term bond yields. Reuters reported that the US 30-year Treasury yield moved near levels not seen since 2007, contributing to pressure across semiconductors.

This matters because Nvidia is a long-duration equity despite its current profits. A large portion of today’s valuation comes from cash flows expected many years into the future. When the discount rate rises, those distant cash flows become less valuable today.

The irony is that Nvidia can report extraordinary earnings while its stock falls because the market is simultaneously raising the hurdle rate applied to those earnings. That is not irrational. It is basic valuation mathematics.

What I will watch next

First, sequential Data Center growth. At this scale, even single-digit sequential growth adds billions of dollars.

Second, gross margin. If it remains near the mid-70s, Nvidia’s pricing power is intact. A sustained decline would tell me customers are gaining negotiating leverage or product mix is becoming less favorable.

Third, accounts receivable and cash conversion. Rapid growth should continue producing cash, not simply accounting earnings.

Fourth, customer financing exposure. I want to distinguish strategic guarantees that unlock profitable projects from structures that effectively subsidize weak demand.

Fifth, hyperscaler cloud growth. Nvidia ultimately sells picks and shovels to companies that need to monetize the gold rush.

The bear case is becoming more sophisticated

A year ago, the weak bearish argument was “AI is a bubble.” That was too crude. Demand was real, revenue was real and cash flow was real.

The stronger bear case in 2026 is different: AI can be revolutionary and still produce poor returns for some infrastructure investors. Overbuilding railroads did not mean railroads were useless. Overbuilding fiber did not mean the internet was a fad. Transformative technologies can attract too much capital at the wrong price.

Nvidia is insulated because it earns money selling the infrastructure. But it is not infinitely insulated. If enough customers realize that incremental projects earn inadequate returns, orders eventually slow.

Why I still would not bet aggressively against Nvidia

The company has consistently executed faster than skeptics expected. Its product cadence is exceptional. It has the financial resources to invest through cycles, a software ecosystem competitors struggle to replicate and customer demand that remains constrained by physical infrastructure such as power and data-center capacity.

Shorting that combination because the multiple looks high has been a poor strategy. The burden of proof should remain on the thesis that demand is about to break.

But I also would not use past execution as permission to ignore valuation. Every great company has a price at which expected returns become ordinary.

My conclusion

Nvidia’s current numbers still justify describing it as one of the strongest businesses in global technology. Revenue growth is extraordinary. Margins remain exceptional. The data-center platform is becoming broader, not narrower.

What has changed is the environment around the company. AI infrastructure is now so large that financing structures, power constraints, customer returns and bond yields matter almost as much as chip benchmarks.

I remain constructive on the company, but I would demand a margin of safety in the stock. The next stage of the thesis will not be decided by whether Blackwell sells. It will be decided by whether the customers buying Blackwell earn enough money to keep buying Rubin at even greater scale.

For a related look at how AI infrastructure can become financially leveraged, see my analysis of CoreWeave.

Sources and data date

Data date: 20 August 2026. Valuation ranges are my analytical scenarios and are not price targets.

• Nvidia: Q1 fiscal 2027 results
• Reuters: AI earnings and financing scrutiny
• Reuters: bond yields and semiconductor selloff

This article reflects my personal analysis and is not investment advice.