NVIDIA: Selling the AI System—and Proving Its Economics

Aria Vantage | Research cut-off: September 9, 2026

NVIDIA’s strongest investment case is that customers need a working computing system, not simply a faster chip. Accelerators, networking, memory, software and deployment tools must function together. The more difficult that integration becomes, the more valuable a supplier capable of coordinating it can be.

The stock question is whether that advantage produces enough durable cash earnings to justify a company worth approximately $5.4 trillion at the September 9 reference price. Leadership in a growing industry is a foundation for analysis. It does not, by itself, establish that the shares are mispriced.

The system advantage

NVIDIA reported fiscal Q2 revenue of $96.22 billion, including $89.02 billion from Data Center. Gross profit as a percentage of revenue was 75%, while GAAP and adjusted diluted EPS were $2.46 and $2.22. Its Q3 outlook calls for $108 billion of revenue, plus or minus 2%, and 74% gross profit as a percentage of revenue, plus or minus half a percentage point. NVIDIA’s August 26 results.

The next-generation Vera Rubin platform extends the systems approach. NVIDIA describes an architecture combining GPUs, CPUs, networking, storage processing and inference acceleration across coordinated racks. Its claimed throughput gains depend on workload and configuration; they should be treated as vendor comparisons rather than universal customer outcomes. NVIDIA’s Vera Rubin production announcement.

This design matters economically because bottlenecks migrate. Making one processor faster is less useful if communication, memory access or power delivery prevents the rest of the installation from keeping up. Coordinated design can help customers bring usable capacity online sooner and extract more output from a constrained facility. The competitive product is therefore the cost and reliability of completed work, not an isolated benchmark.

Software strengthens that proposition. CUDA supplies compilers, libraries and development tools for GPU computing. Its practical significance is that applications need to be developed, tuned and maintained, not merely placed on hardware. NVIDIA’s CUDA documentation.

Our interpretation is that a mature software environment can reduce adoption friction and raise the effort involved in switching. This is an economic advantage, not a guarantee of permanent dependence. Large customers have resources to support alternatives, while developers can adopt abstraction layers and portable tools. NVIDIA must keep making its entire platform attractive enough that staying remains the better operating decision.

What customers are committing to

There is fresh evidence that customers still want that platform. AWS and NVIDIA announced plans for two million additional GPUs, with the joint release describing deployments in 2027–2028. The collaboration also spans CPUs, networking, models and robotics. This is a deployment plan, not revenue already recognized, and the release does not disclose a price that permits a reliable contract-value calculation. AWS–NVIDIA announcement.

The breadth of the relationship helps the bull case: a customer can build some of its own technology and still buy substantial NVIDIA capacity. Investors should nevertheless avoid counting each announced partnership as an independent source of final demand. Cloud providers, model developers and infrastructure companies can participate in the same underlying project.

Microsoft’s latest call provides a useful demand check from another buyer: management said Azure demand continued to exceed available supply and projected further growth in fiscal 2027 capital spending. It also explained that lease classification would reduce its reported calendar-year capital-expenditure figure without changing its underlying investment expectations. Comparing headline spending numbers without their definitions can produce a false signal. Microsoft’s fiscal Q4 call.

Growth and the cost of delivery

The most consequential new management forecast is approximately 70% revenue growth for NVIDIA’s FY2028, described as constrained by supply. Management also expects gross profit as a percentage of revenue to bottom around 71%–72% in Q4 FY2027 before settling around 72%–73% in FY2028, citing sharply higher memory costs. These are company projections, not independently established outcomes. NVIDIA’s earnings-call transcript, pages 3 and 7.

That combination is the central investment debate. Demand may remain strong while a larger portion of system economics accrues to scarce component suppliers. Growing volumes and a broader product portfolio could more than offset a lower share of sales retained as gross profit. Alternatively, investors could find that extraordinary sales growth produces less incremental profit than expected. A bullish analysis must model both revenue and the cost of fulfilling it.

What the share price requires

Valuation needs equally careful accounting. NVIDIA’s adjusted results now include stock compensation, and the reconciliation removes investment gains, among other items. GAAP EPS is therefore higher than adjusted EPS this quarter. Simply using the larger number as a permanent earnings base would overstate operating comparability. NVIDIA’s accounting reconciliation.

At the $224.215 intraday reference price, approximately 3:20 p.m. Eastern on September 9, four times quarterly adjusted EPS implies about 25.2 times annualized earnings. That is a mechanical snapshot, not a next-year forecast. A 20-times valuation at the same stock price would require roughly $11.21 of annual EPS on a comparable basis. This reverse calculation makes the required earnings progress visible without pretending to establish fair value.

Stock Analysis’s public forecast snapshot showed FY2027 revenue of $411.31 billion and FY2028 revenue of $677.83 billion on September 9. Those third-party estimates imply approximately 65% growth. They already incorporate an ambitious outlook; strong future performance is not an undiscovered idea. The accessible EPS presentation did not provide a sufficiently clear accounting-basis reconciliation for direct comparison here. Public forecast snapshot.

Commitments and competing platforms

There is also a financing dimension. NVIDIA’s latest filing reports $279 billion of supply and capacity commitments and additional obligations involving cloud services, investments and future leases. These are not all conventional debt or equally firm obligations, but they matter when testing cash returns and flexibility. NVIDIA’s fiscal Q2 Form 10-Q, Note 10.

Securing supply can protect growth during a shortage. It can also leave cash tied up or create losses if customers delay purchases. The relevant underwriting question is how much cash remains after delivery, strategic investments and financing commitments, and what happens if several customers simultaneously slow expansion. A simple earnings multiple cannot answer that question.

Competition provides another clear falsifier. Google’s 2026 infrastructure announcement introduced separate TPU 8t and TPU 8i designs for training and inference, while also continuing to offer NVIDIA instances. This is evidence of a customer building a mixed infrastructure portfolio. Google’s Cloud Next announcement.

NVIDIA need not capture every workload to grow. It does need to defend enough of the most valuable opportunities to preserve attractive returns. If customers consistently achieve better economics on specialized alternatives, NVIDIA could face a weaker mix, pricing pressure, or a smaller share of incremental spending. The question is measurable customer value, rather than whether an alternative chip exists.

Scenarios and the next tests

The bullish scenario combines a successful Rubin ramp, continued customer utilization, and rising cash earnings despite higher component costs. A middle scenario delivers rapid sales growth but limited multiple expansion because costs and commitments rise with it. The bearish scenario combines weaker customer monetization, deployment delays and weaker profitability, allowing earnings expectations and the valuation multiple to fall together. Export restrictions add another variable: the current quarterly outlook assumes no China Data Center compute revenue. NVIDIA’s quarterly outlook.

The next results should be judged against the revenue and profitability outlook, Rubin deployment progress, receivables and cash conversion—not excitement around AI alone. Persistent delivery problems, weaker utilization across customers, or commitments growing faster than economically supported orders would require revisiting the case. NVIDIA’s operating evidence supports a strong business thesis; the stock’s return will depend on cash earnings growing faster than the obligations required to support them.

Research note: This article was prepared with AI assistance. It provides general information and analysis, not personalized investment advice. Prices, forecasts and company plans can change.