GOVERNANCE WATCH
By InsidEntity Editorial Desk · Aug 27, 2026 · 12 min read
Nvidia delivered another record quarter. But beneath the $96.2 billion headline, the company changed how it measures adjusted earnings, changed how it presents the business, and disclosed a contingent infrastructure obligation capped at $105 billion.
Nvidia reported its second-quarter fiscal 2027 results on 26 August.
The headline was enormous.
Revenue reached $96.2 billion, up 18% sequentially and 106% year on year.
Data Center revenue reached $89.0 billion, up 117%.
GAAP diluted earnings per share were $2.46, while non-GAAP diluted earnings per share were $2.22.
Both GAAP and non-GAAP gross margins were approximately 75%.
The company also returned approximately $26 billion to shareholders during the quarter.
It was another extraordinary Nvidia quarter.
But the most interesting information was not the size of the beat.
It was what changed around the numbers.
1. Nvidia changed what “non-GAAP” means
Start with the number investors often use to strip away accounting noise.
GAAP EPS: $2.46
Non-GAAP EPS: $2.22
That relationship deserves attention.
Beginning in the first quarter of fiscal 2027, Nvidia changed its non-GAAP methodology so that stock-based compensation was no longer excluded. Nvidia also updated the historical non-GAAP information it presents for comparison.
The change is not merely theoretical.
Nvidia said stock-based compensation was expected to add approximately $1.9 billion to non-GAAP operating expenses in Q1 fiscal 2027. Actual stock-based compensation expense in that quarter was $1.928 billion.
That makes Nvidia’s adjusted earnings more conservative than under its previous methodology.
Melius Research analyst Ben Reitzes has estimated the change has only about a 3% impact on Nvidia’s adjusted EPS, compared with an estimated 14% to 20% impact if companies such as Broadcom, AMD and Marvell applied the same treatment.
That comparison matters.
The methodology changed, but Nvidia is not being affected to the same degree as some of its peers.
So the important observation is not that Nvidia has suddenly started reporting a worse number.
It is that the definition of the number changed, and the size of that adjustment tells us something about Nvidia’s compensation structure relative to its peers.
The ruler changed.
The measurement did not suddenly become useless.
But investors comparing fiscal 2027 adjusted earnings with older periods need to know that they are no longer looking at exactly the same construction.
2. Nvidia also changed the map
The earnings methodology was not the only reporting change.
Nvidia also changed how it presents its business.
Beginning in fiscal 2027, the company changed its presentation of revenue by market platform and recast comparable periods. The new structure gives investors greater visibility into the composition of Data Center revenue, including Hyperscale and AI Clouds, Industrial & Enterprise.
That is useful.
But it also creates another break in the historical series.
Investors now have more information about where Nvidia’s Data Center revenue comes from, while simultaneously having to adjust some of the ways they compare the business with earlier reporting periods.
That distinction matters.
This is a reporting presentation change, not evidence that Nvidia’s underlying business suddenly changed overnight.
The map changed.
The territory did not.
3. Data Center now accounts for 92.5% of revenue
The arithmetic is striking.
$89.0 billion Data Center revenue divided by $96.2 billion total revenue equals 92.5%.
Nvidia’s Data Center business therefore represented approximately 92.5% of total quarterly revenue.
That is business concentration.
It is not the same thing as customer concentration.
But the two questions are increasingly connected.
Nvidia’s new reporting structure gives investors more visibility into the customers and market platforms inside that enormous Data Center business.
And that makes the spending plans of the companies building AI infrastructure increasingly important to Nvidia’s future revenue.
Management has pointed to rapidly rising capital expenditure by the largest hyperscalers.
The implication is straightforward: Nvidia’s growth is increasingly tied not only to demand for its products, but to how aggressively the world’s largest technology companies continue to build AI infrastructure.
That is not a warning about the business.
It is a description of the business.
The customer-concentration question is separate
Nvidia did not disclose the same customer percentages in the Q2 fiscal 2027 results that it had disclosed in earlier periods.
But the historical record provides a useful baseline.
In Q2 fiscal 2026, one direct customer represented 23% of total revenue and a second represented 16%. By fiscal 2026 as a whole, the corresponding figures were 22% and 14%. In Q1 fiscal 2027, three direct customers represented 21%, 17% and 16% of revenue.
The specific customers can change from period to period.
The broader point is more durable: Nvidia has consistently generated a substantial share of revenue from a relatively small number of direct customers.
As Hyperscale becomes an increasingly important part of the Data Center business, that is something worth tracking.
The question for future quarters is not simply whether Data Center remains dominant.
It is whether the customer base inside that 92.5% becomes more concentrated or more diversified.
4. Then there is the $105 billion disclosure
The most unusual disclosure sits outside the income statement.
On 17 August, Nvidia entered into multiple residual value guarantees with SB Energy relating to leases covering approximately 4.25 gigawatts of IT load at the PORTS-Pike Technology Campus in Ohio.
Nvidia’s aggregate payment obligation under its initial commitment is capped at $105 billion.
The headline number is enormous.
But the legal structure matters more than the headline.
This is not a $105 billion cheque.
It is not $105 billion of debt.
It is not a forecasted $105 billion loss.
It is a capped contingent obligation connected to the residual value of infrastructure supporting OpenAI’s lease.
And the filing gives unusually specific information about when that exposure can become relevant.
When can it actually trigger?
The guarantees generally become effective when the applicable leases commence, subject to specified conditions including the lessor satisfying ready-for-service requirements.
Those conditions are expected to begin in 2028.
A Trigger Event occurs if:
- OpenAI becomes insolvent and that results in a lease default; or
- OpenAI fails to make payments under a lease.
If a Trigger Event occurs, Nvidia does not simply write a $105 billion cheque.
The agreement gives Nvidia several options. It can:
- assume the applicable lease;
- require SB Energy to seek to relet the premises;
- initiate a sale process;
- allow the lease to terminate; or
- defer exercising those remedies for up to one year while paying specified project costs.
The filing also states that OpenAI has agreed to reimburse and indemnify Nvidia for amounts actually paid to SB Energy under the agreements.
That changes the character of the headline.
The right way to describe this is not: “Nvidia faces a $105 billion liability.”
The more accurate description is: Nvidia has accepted a contingent obligation capped at $105 billion, with the relevant guarantees tied to infrastructure expected to come online from 2028 and subject to specific trigger events and mitigation mechanisms.
That is still significant.
But it is a very different statement.
5. Why the $26 billion comparison still matters
Nvidia returned approximately $26 billion to shareholders during the quarter.
The maximum aggregate payment obligation under the Ohio guarantees is $105 billion.
These figures are not economically equivalent.
One is cash actually returned during three months.
The other is a maximum contingent contractual exposure that may never result in a comparable cash payment.
But the comparison establishes scale.
$26 billion returned. $105 billion maximum contingent exposure.
The second figure is roughly four times the first.
That does not mean Nvidia is expected to pay four times what it returned to shareholders.
It means the scale of Nvidia’s involvement in AI infrastructure has moved beyond the conventional supplier relationship.
The company is providing products.
It is investing in infrastructure.
It is helping secure land, power and data-centre capacity.
And it is providing credit support around a major customer infrastructure project.
That is a materially broader role.
6. Nvidia is no longer just selling the infrastructure
The $1.5 billion investment in SB Energy makes that broader role particularly visible.
Nvidia agreed to invest $1.5 billion in SB Energy, according to its disclosure of the arrangement, while also becoming the exclusive AI compute infrastructure provider for the Ohio site.
The arrangement connects three parts of the AI infrastructure chain.
Nvidia supplies the computing platform.
OpenAI provides the demand and leases the facility.
SB Energy develops and owns the physical infrastructure.
Nvidia therefore has an economic interest in making the infrastructure work, while simultaneously supplying the technology that will operate inside it.
That does not prove circular financing.
It does, however, make the contractual disclosures more important.
The further Nvidia moves into financing, investing in and securing the infrastructure required to deploy its products, the less useful it becomes to analyse the company solely as a semiconductor supplier.
7. The 0.00 Auditor Independence score needs context
Nvidia’s InsidEntity Company Risk Rating is 3.22, Good.
The four pillar scores are:
| Component | Score | Reading |
|---|---|---|
| Director Independence | 5.00 | Benchmark, by default |
| Shareholder Influence | 5.00 | Benchmark |
| Director Capacity | 2.88 | Caution |
| Auditor Independence | 0.00 | Nil |
| Overall CRR | 3.22 | Good |
The Auditor Independence score is the obvious outlier.
But a 0.00 InsidEntity score should not be presented as a finding that Nvidia’s auditor is actually non-independent.
Nvidia’s independent registered public accounting firm is PwC.
PwC has served as Nvidia’s auditor since 2004, and Nvidia’s Audit Committee again recommended PwC for fiscal 2027. Shareholders were asked to ratify that appointment at the June 2026 annual meeting.
Nvidia’s Audit Committee also states that it received PwC’s required independence communications and considered whether permitted non-audit services were compatible with PwC’s independence.
That is important context.
The 0.00 score is a methodological output from InsidEntity’s rating framework.
It is not, by itself, a finding of actual auditor independence impairment.
Those are two different propositions.
And that distinction is exactly why governance ratings need to be read alongside the underlying corporate disclosures.
8. What the CRR actually measures
The CRR is not designed to measure Nvidia’s entire risk profile.
It does not answer how dependent Nvidia is on Data Center revenue, how concentrated its customers are, how sustainable hyperscaler capital expenditure is, how much infrastructure exposure Nvidia has accepted, what happens if OpenAI defaults, or how valuable AI infrastructure will remain over multiple hardware generations.
Those are legitimate questions.
They are simply not what the CRR measures.
The CRR measures governance structure through four pillars: Director Independence, Director Capacity, Auditor Independence and Shareholder Influence.
That is why a company can simultaneously have $96.2 billion of quarterly revenue, 92.5% Data Center concentration, a $105 billion maximum contingent obligation, and a 3.22 Good governance rating.
There is no contradiction.
The numbers are answering different questions.
9. What investors should watch next
The most useful consequence of Nvidia’s disclosures is not another headline number.
It is a better monitoring framework.
The non-GAAP methodology
Fiscal 2027 adjusted earnings now include stock-based compensation.
Investors comparing Nvidia’s adjusted earnings with earlier periods should keep that methodological break visible.
Hyperscale concentration
Data Center represents 92.5% of quarterly revenue.
Historical customer disclosures show that a relatively small number of direct customers have represented substantial portions of revenue.
Future disclosures should show whether Nvidia’s customer concentration is broadening or tightening.
The Ohio guarantees
The $105 billion cap should not be treated as a forecasted loss.
The important variables are the commencement of the leases, the performance of the underlying infrastructure, OpenAI’s ability to meet its obligations, the residual value of the facilities and Nvidia’s ability to use the contractual mitigation mechanisms.
Hyperscaler capital expenditure
Nvidia’s growth is increasingly connected to the infrastructure budgets of the companies building AI systems.
If those budgets continue expanding, Nvidia’s current trajectory has a powerful demand engine behind it.
If the spending cycle changes, the effect on Nvidia could be significant.
Nvidia’s expanding role
Supplier. Investor. Infrastructure participant. Credit supporter.
The more roles Nvidia occupies in the AI infrastructure chain, the more important its contractual and financial disclosures become.
The record quarter is real. So are the questions around it.
Nvidia delivered another extraordinary quarter.
Revenue reached $96.2 billion.
Data Center reached $89.0 billion.
GAAP EPS reached $2.46.
The company returned approximately $26 billion to shareholders.
And Nvidia is guiding to approximately $108 billion of revenue for the next quarter.
The operating performance is not the mystery.
The more interesting story is what sits around it.
Nvidia changed the construction of its non-GAAP earnings.
It changed the way it presents its business.
Data Center now accounts for 92.5% of revenue.
Its historical customer disclosures show substantial concentration among a relatively small number of direct customers.
The company has moved deeper into the infrastructure surrounding AI compute.
And it has accepted a contingent obligation capped at $105 billion tied to a major OpenAI data-centre project, with the relevant guarantees subject to specific conditions and expected to begin from 2028.
None of that means Nvidia has a governance problem.
None of it means the $105 billion will become a $105 billion loss.
And none of it means the 3.22 Good rating is inconsistent with the underlying business.
It means the rating is answering one question while the filings answer others.
The point is not to replace one number with another.
It is to know what the number you are looking at actually means.
About the InsidEntity Company Risk Rating: The InsidEntity Company Risk Rating (CRR) scores companies on a 1 to 5 scale across four governance pillars, weighted 40% Director Independence and 20% each for Director Capacity, Auditor Independence and Shareholder Influence. The same scale applies to pillar scores and to the published overall rating, which is the average of the four pillar scores: 1.00 to 1.99 is Risk, 2.00 to 2.99 is Caution, 3.00 to 3.99 is Good, 4.00 to 4.99 is Excellent, and 5.00 is Benchmark. The CRR measures governance structure. It does not measure business-segment concentration, customer concentration, contingent obligations, investment exposure or the economics of the AI infrastructure cycle. Ratings update as corporate information changes.
Know your entity. This article is independent research and is not financial advice. Figures and statements concerning Nvidia’s financial results and contractual arrangements are based principally on Nvidia’s SEC filings and corporate disclosures. The existence of a contingent obligation does not mean that the obligation will be paid, and nothing in this article alleges wrongdoing by Nvidia, OpenAI, SB Energy or any other company named.