What NVIDIA’s August 26 Earnings Could Reveal About the Sustainability of the AI Investment Cycle
NVIDIA will report its fiscal 2027 second-quarter results on August 26, 2026.
For a company that has repeatedly broken revenue records, another earnings beat would hardly be surprising. NVIDIA has exceeded its own revenue guidance quarter after quarter, while demand for AI computing infrastructure has continued to expand at an extraordinary pace.
But I believe this earnings report matters for a different reason.
The question is no longer simply whether AI demand exists.
The more important question is whether the earnings being generated by the AI ecosystem can continue growing fast enough to justify the enormous amount of capital required to build it, particularly when the price of that capital is rising.
There are therefore two questions I will be asking when NVIDIA reports.
Can NVIDIA outrun the rising cost of capital?
And, more importantly:
Can NVIDIA’s customers earn enough on AI capital to keep NVIDIA’s earnings growing faster than the market’s discount rate?
The distinction matters because NVIDIA itself is not financially constrained in the conventional sense.
Its customers are financing the infrastructure boom.
Microsoft, Alphabet, Amazon and Meta are deploying unprecedented amounts of capital into GPUs, servers, networking equipment, data centers, power infrastructure and other AI capacity.
As long as those investments generate returns sufficiently above their cost of capital, the cycle can continue.
But if earnings growth begins slowing while capital intensity continues rising, the bond market — rather than a collapse in AI demand — may eventually become the force that challenges AI valuations.
That is what makes NVIDIA’s August 26 report more than another semiconductor earnings release.
It is becoming a test of the economics of the entire AI capital cycle.
1. NVIDIA Is Still Growing at an Extraordinary Rate
There is little evidence so far that AI demand is collapsing.
NVIDIA’s revenue increased from $46.7 billion in Q2 FY2026 to $81.6 billion in Q1 FY2027.
Data Center revenue increased from $41.1 billion to $75.2 billion over the same period.
Even more strikingly, year-over-year growth accelerated.
Source: NVIDIA Investor Relations, Q2 FY2026 through Q1 FY2027.
The combination is important.
NVIDIA has not been generating extraordinary revenue growth by sacrificing profitability. Gross margin returned to roughly 75% and remained there even while quarterly revenue expanded dramatically.
So far, NVIDIA has been delivering both growth and extraordinary economics.
But the next question is increasingly about the velocity of that growth.
2. The $91 Billion Number Is Important — But the Next Guidance May Matter More
When NVIDIA reported Q1 FY2027 results in May, management guided Q2 revenue to $91.0 billion, plus or minus 2%.
The company also guided GAAP gross margin to 74.9% and non-GAAP gross margin to 75.0%, both plus or minus 50 basis points.
NVIDIA explicitly stated that the forecast assumed no Data Center compute revenue from China.
That would still represent enormous absolute growth.
But it would also be noticeably slower than the roughly 20% sequential growth NVIDIA generated in Q3 FY2026, Q4 FY2026 and Q1 FY2027.
This is why the distinction between growth and growth velocity is becoming increasingly important.
NVIDIA has also consistently exceeded its own revenue guidance.
Sources: NVIDIA Investor Relations.
A beat on August 26 may therefore not be enough by itself. The market has become accustomed to NVIDIA beating its own guidance.
What matters more is whether the company can continue pushing the forward growth curve higher.
A beat tells us where demand has been. Guidance tells us whether the capital cycle is still accelerating.
For that reason, NVIDIA’s Q3 FY2027 revenue guidance may ultimately be more important than the headline Q2 result.
3. NVIDIA’s Earnings Are Rising — But So Is the Price of Money
There is another side to the equation.
AI earnings are growing at extraordinary speed. But the risk-free rate remains unusually high.
As of August 18, the U.S. Treasury’s official daily par yield curve showed:
10-year Treasury: 4.71%
30-year Treasury: 5.28%
Source: U.S. Department of the Treasury.
U.S. Treasury Daily Treasury Par Yield Curve Rates
These yields matter directly to both equity valuation and corporate capital allocation.
A higher Treasury yield raises the return investors can earn without taking equity risk. It also raises corporate borrowing costs and the hurdle rates companies apply when deciding whether a new investment project is worth pursuing.
The transmission mechanism is straightforward:
Treasury yields rise → risk-free rate rises → corporate cost of capital rises → project hurdle rates rise → AI infrastructure must generate higher returns to justify the same investment
This is particularly important because AI is becoming much more physically capital-intensive than the early software-driven narrative suggested.
The market therefore has to compare two moving variables.
Earnings growth versus the discount rate and 2. NVIDIA’s customers.
But so is the return investors can demand from capital elsewhere.The valuation question is ultimately about which force wins.
4. NVIDIA’s Real Cost-of-Capital Risk Sits With Its Customers
This is the most important distinction in the analysis.
NVIDIA does not need to finance the entire AI buildout itself. Its customers do.
NVIDIA sells accelerators and computing platforms, but someone has to finance the GPUs, data centers, networking systems, electrical infrastructure, cooling equipment, land and energy supply required to deploy them.
That burden increasingly sits with the hyperscalers.
Microsoft, for example, reported $31.9 billion of capital expenditures in fiscal Q3 2026. Roughly two-thirds of that spending went toward shorter-lived assets, primarily GPUs and CPUs.
Microsoft generated $46.7 billion of operating cash flow and $15.8 billion of free cash flow in the same quarter.
The company also said it expected to invest roughly $190 billion in capital expenditures during calendar 2026 and expected to remain capacity constrained through at least 2026.
Refer to Microsoft Investor Relations.
This is currently one of the strongest pieces of evidence supporting the bull case. Microsoft is spending enormous amounts of money, yet demand remains strong enough that the company still expects supply constraints, while free cash flow remains substantially positive.
Alphabet provides a more complicated picture.
Google Cloud revenue increased 82% year over year to approximately $24.8 billion in Q2 2026. At the same time, Alphabet raised its 2026 capital expenditure guidance to $195 billion–$205 billion. Quarterly capital expenditure reached approximately $44.9 billion, exceeding operating cash flow of approximately $39.1 billion and pushing quarterly free cash flow to roughly negative $5.9 billion.
Refer to Rueter’s Alphabet Q2 2026 results and CapEx guidance
The demand signal is exceptional. But the cash-flow signal shows how capital-intensive that demand has become.
Amazon offers another revealing case.
AWS revenue increased 37% year over year to $42.2 billion in Q2, its fastest growth in 18 quarters. AWS operating income reached $16.6 billion.At the same time, Amazon’s trailing-12-month operating cash flow increased to $161.4 billion, while trailing-12-month free cash flow fell to an outflow of $7.6 billion. Amazon attributed the decline primarily to a $66.1 billion year-over-year increase in purchases of property and equipment.
Refer to Amazon Q2 IR
Meta tells a similar story from another direction.
Meta’s Q2 2026 revenue increased 28% year over year to $60.8 billion. Capital expenditures reached $31.08 billion. Operating cash flow was $31.86 billion. Free cash flow was only $784 million. Meta now expects full-year 2026 capital expenditures of $130 billion–$145 billion.
Refer to Meta Investor Relations Q2
Taken together, the scale is extraordinary.
Microsoft expects roughly $190 billion of 2026 capital expenditure.
Alphabet expects $195 billion–$205 billion.
Meta expects $130 billion–$145 billion.
Amazon is also operating at an extremely high investment level, even though its reporting definitions are not directly comparable with those of the other three companies.
The point is not to produce a perfectly comparable combined number. The point is that this is no longer experimental spending. It is an industrial-scale capital cycle.
And that brings us back to the most important question:
How much incremental earnings can the AI ecosystem generate for every incremental dollar of capital deployed?
Strong AI demand and declining capital efficiency can exist at the same time. That distinction will become increasingly important as the risk-free rate rises.
Some financing relationships within the AI ecosystem also deserve monitoring. Strategic investments, long-term capacity commitments and financing arrangements between suppliers, infrastructure operators and customers can strengthen demand.
That does not automatically make the demand artificial. But it does mean investors should distinguish underlying economic demand from demand that may be accelerated by financing relationships within the ecosystem.
The relevant question is not whether the AI cycle is a bubble. It is whether the return on the next dollar of AI capital remains sufficiently above its rising cost.
5. The Cost of AI Does Not Stop at the GPU
The scale of hyperscaler capital spending becomes easier to understand once we look beyond NVIDIA’s GPU.
An AI factory requires a much larger physical system.
GPU → HBM → networking → data center → electrical infrastructure → power generation and grid capacity
This is also why the AI cycle is directly linked to memory semiconductors.
NVIDIA accelerators require increasingly large amounts of high-bandwidth memory, connecting GPU demand directly to the HBM cycle at SK Hynix, Samsung Electronics and Micron Technology.
But memory is only one layer. Power is becoming another major constraint.
NVIDIA is developing an 800 VDC architecture for next-generation AI factories because conventional lower-voltage data-center power distribution becomes increasingly inefficient as rack power moves toward megawatt-scale levels.
According to NVIDIA’s technical documentation, moving to an 800 VDC distribution architecture can allow approximately 85% more power to be transmitted through the same conductor size and reduce copper requirements by around 45%. Source: NVIDIA Developer Technical Blog.
These may appear to be engineering details. Economically, however, they tell us something much larger. The cost of AI does not stop at the GPU.
Each additional layer of compute requires additional memory, networking, electrical equipment, cooling, grid capacity and potentially energy storage.
AI may ultimately raise productivity and lower the cost of producing many goods and services. But getting there requires enormous physical investment first.
That creates one of the defining paradoxes of the current AI cycle:
AI may eventually be economically deflationary in what it produces, while being extraordinarily capital-intensive in what must first be built.
And every one of those investments has a cost of capital.
6. What NVIDIA Must Prove on August 26
When NVIDIA reports, I will therefore be watching four variables rather than simply asking whether earnings beat expectations.
The fourth test can be extremely crucial.
Suppose NVIDIA reports $95 billion rather than $91 billion. That would be a strong result.
But if management then provides an outlook suggesting substantially slower sequential growth, investors may conclude that the absolute level of AI demand remains enormous while the growth rate is beginning to normalize.
Conversely, if NVIDIA significantly exceeds Q2 guidance and again provides a strong Q3 outlook while maintaining gross margins around 75%, the bull case becomes much stronger.
It would suggest that hyperscaler demand remains powerful enough to absorb rising infrastructure costs and a higher market discount rate.
The distinction can be summarized simply, a beat tells us where demand has been. Guidance tells us whether the capital cycle is still accelerating.
7. The Bull Case Does Not Require Lower Interest Rates
The bullish case for NVIDIA does not necessarily require the 10-year Treasury yield to collapse.
If NVIDIA’s earnings and its customers’ AI monetization continue growing fast enough, a structurally higher Treasury yield can coexist with elevated AI valuations.
The equation would effectively be higher discount rate but even faster earnings growth and therefore valuation remains supported.
Microsoft currently offers evidence that this outcome is possible.
The company is deploying extraordinary amounts of capital while still producing substantial free cash flow and reporting demand strong enough to keep capacity constrained.
AWS growth provides another supporting signal.
If the infrastructure being installed today generates substantially higher cloud revenue, enterprise AI revenue, advertising efficiency and productivity gains over the next several years, today’s investment cycle could remain economically rational even at higher interest rates.
In that scenario, AI would be doing something unusual.
It would be outgrowing the price of money.
8. The Bear Case Does Not Require an AI Collapse
The bearish scenario is more subtle.
NVIDIA does not need to miss earnings. AI demand does not need to collapse. Microsoft, Amazon, Meta and Alphabet do not need to stop building data centers.
Instead, imagine the following combination:
NVIDIA continues producing strong revenue growth. But sequential growth gradually slows. Gross margins remain strong but stop expanding. Hyperscaler capital expenditures continue rising. Depreciation and infrastructure costs increase. Free cash flow remains under pressure.
And the U.S. 10-year Treasury yield remains close to 5%.
Nothing in that scenario would resemble an AI recession. Yet AI equity valuations could still compress.
The reason is simple. Excellent growth can become insufficient when the amount and price of the capital required to generate that growth rise even faster. This is the core risk I believe investors increasingly need to confront.
The greatest risk to AI equities may not be collapsing demand.
It may be excellent growth that is no longer excellent enough relative to the amount and price of capital required to sustain it. That is a very different argument from saying AI is a bubble.
It is an argument about capital efficiency and valuation.
Conclusion: Earnings vs. the Discount Rate
NVIDIA enters its August 26 earnings report from an extraordinary position.
Revenue reached $81.6 billion last quarter.
Data Center revenue reached $75.2 billion.
Year-over-year Data Center growth reached 92%.
Gross margin remained close to 75%.
And NVIDIA has guided the current quarter to approximately $91 billion of revenue if I am not wrong.
There is little evidence in those numbers that AI demand has disappeared. At the same time, the infrastructure required to satisfy that demand is becoming enormous.
Microsoft expects roughly $190 billion of capital expenditure in calendar 2026.
Alphabet expects $195 billion–$205 billion.
Meta expects $130 billion–$145 billion.
Amazon’s rapidly rising infrastructure investment has already pushed trailing-12-month free cash flow into negative territory despite accelerating AWS revenue growth.
Behind those investments lies another layer of spending on HBM, networking, data centers, power systems and grid infrastructure.
Meanwhile, the U.S. 10-year Treasury yield is around 4.7%, while the 30-year yield is above 5%.
That changes the test.
The question facing investors is no longer simply how much more AI infrastructure can the world build, it is what return will that infrastructure earn.
NVIDIA itself has little difficulty financing its growth. The harder cost-of-capital test sits with the companies buying its products.
As long as Microsoft, Amazon, Alphabet, Meta and the broader AI ecosystem can convert enormous capital expenditures into earnings and cash flows that comfortably exceed their rising hurdle rates, NVIDIA’s earnings engine can continue running.
But if growth in economic returns begins lagging behind growth in the capital deployed, NVIDIA can continue reporting good numbers and still face a much more difficult valuation environment.
That is why I will be watching the August 26 results for more than another earnings beat.
I will be watching for evidence that the return on AI capital remains high enough to support the next round of investment.
Because in the end, NVIDIA’s ability to outrun the cost of capital depends on something NVIDIA cannot determine alone:
Can NVIDIA’s customers earn enough on AI capital to keep NVIDIA’s earnings growing faster than the market’s discount rate?
If the answer begins moving toward no, the bond market may become a more powerful constraint on AI valuations than semiconductor demand itself.
That is the real test on August 26.