Nvidia Earnings Preview: Can AI Buyers Absorb Higher Costs?

Nvidia reports fiscal second-quarter results Wednesday, August 26, and the headline numbers are unlikely to be the hardest part of the setup. This Nvidia earnings preview comes as Wall Street expects roughly $92 billion in revenue, nearly double the year-ago level, with adjusted earnings around $2.09 per share as hyperscalers continue spending aggressively on AI infrastructure. Options markets are pricing roughly a 6% post-earnings move, while investors broadly expect another strong quarter and potentially another beat-and-raise.

The obvious debate is whether AI demand remains strong enough to support Blackwell growth, mid-70% gross margins and the coming Vera Rubin ramp. But a new variable has entered the equation at exactly the wrong—or right—time. Bloomberg reported over the weekend that some Nvidia-powered server configurations shipping in early 2027 could see prices rise by more than 15%, largely because of surging memory costs.

That changes the earnings question. The market may care less about what Nvidia sold this quarter than whether customers can absorb meaningfully higher AI infrastructure costs over the next two quarters without slowing deployment.

What Wall Street Is Modeling: Nvidia Earnings Preview

The base case is already demanding.

Consensus estimates point to revenue of roughly $92 billion, compared with Nvidia’s prior-quarter guidance of $91 billion, plus or minus 2%. More bullish analysts are looking for something closer to $95 billion, while Jefferies has reportedly modeled approximately $108 billion for the following quarter. The market is therefore not entering Wednesday hoping for evidence that AI demand exists. It is entering expecting evidence that demand remains exceptional.

Margins are equally important. Nvidia guided last quarter to approximately 75% non-GAAP gross margin and maintained its expectation for margins to remain in the mid-70s for the full year. That level matters because it reinforces the idea that Nvidia is not simply shipping extraordinary volumes—it is retaining extraordinary economics while doing so.

The narrative investors have largely accepted is straightforward: Blackwell is ramping rapidly, hyperscaler capital expenditures remain elevated, networking revenue is expanding alongside compute, and Vera Rubin should create another major product cycle beginning in the second half of 2026. Nvidia has also said it sees approximately $1 trillion in combined Blackwell and Rubin revenue from 2025 through calendar 2027.

That is a powerful setup, but it may also be incomplete. That is why this Nvidia earnings preview has to look beyond the headline revenue beat.

The earnings report comes just as the economics of AI infrastructure are becoming more complicated. Nvidia-powered systems increasingly include expensive memory, networking, CPUs, cooling and power infrastructure, which means customers are no longer evaluating a GPU in isolation. They are evaluating the lifetime return on an entire AI factory.

The market is pricing strong demand; it may not be fully pricing what happens when the cost of satisfying that demand rises sharply.

The True Earnings Pivot: Can Customers Absorb The Price Increase?

The most important dynamic on Wednesday may be customer ROI under higher system pricing.

Bloomberg’s reported 15%+ increases apply to certain Nvidia-powered server systems expected to ship in early 2027, including Vera Rubin and Grace Blackwell configurations. The distinction matters. This is not necessarily a confirmed 15% increase in Nvidia GPU pricing itself. Rather, higher memory and component costs could lift the price of the complete systems customers must deploy.

That creates a direct test of Nvidia’s central economic argument.

On the previous earnings call, Jensen Huang repeatedly emphasized that customers should not judge AI infrastructure based on the purchase price of the GPU. The relevant metrics, he argued, are tokens per dollar, tokens per watt, utilization, uptime, time to production and the lifetime economics of the AI factory. Nvidia also said GB300 had improved throughput by 2.7 times and reduced cost per token by approximately 60% in only six months.

Vera Rubin takes that argument considerably further. Management has said the platform could deliver up to 35 times greater inference throughput and as much as 10 times higher AI-factory revenue compared with Blackwell. Nvidia plans to begin production shipments in Q3, with a larger ramp in Q4 and another substantial step-up early next year.

If those economics hold in real deployments, a 15% increase in system cost could be relatively minor. A customer earning materially more revenue from each rack may rationally accept higher upfront costs because the return on deployed capital still improves.

The problem is that AI infrastructure is not unconstrained by capital. Nvidia itself has described power and capital as limiting resources for AI factory operators. A 15% increase applied across tens of thousands of racks becomes meaningful even for hyperscalers, while smaller AI clouds, enterprises and sovereign customers may be much more sensitive to higher financing and infrastructure costs.

There is also a revealing precedent. During the prior quarter, Nvidia said consumer demand in its Edge Computing business had softened modestly because of higher memory and system prices. Data-center buyers operate under very different economics, but the comment demonstrates that component inflation can eventually translate into demand elasticity.

For this Nvidia earnings preview, that makes demand elasticity one of the most important issues hiding beneath the headline numbers.

Wednesday’s real test is whether Nvidia can prove that productivity gains are rising faster than the total cost of deploying its systems.

That mechanism matters more than the price increase itself. If customers continue accelerating purchases despite higher system costs, Nvidia’s pricing power becomes stronger and the perceived durability of its economics improves. If deployments begin stretching out, the market may have to reconsider how much future AI spending is truly insensitive to price.

The Upside Surprise: Higher Prices, No Demand Destruction

The cleanest upside scenario would not simply be a $2 billion or $3 billion revenue beat.

It would be management showing that higher infrastructure costs are not changing customer behavior.

That could appear through several signals: stronger-than-expected third-quarter guidance, continued mid-70% gross-margin confidence, firmer Vera Rubin shipment schedules, expanding purchase commitments or commentary suggesting that major customers are locking in supply despite rising memory costs. Nvidia already said last quarter that Rubin demand was planned, purchase orders were in place and essentially all major customers were preparing for deployment.

If those commitments have strengthened since May, the weekend pricing report could actually reinforce Nvidia’s positioning rather than undermine it.

The psychology would shift from “AI infrastructure is getting too expensive” toward “Nvidia still has enough economic value to pass through inflation without losing demand.” That is a materially different interpretation because it suggests the company is selling infrastructure whose customer value is rising even faster than its cost.

Rubin would become central to that argument.

Jefferies expects Rubin to become increasingly meaningful later in the year and has modeled more than 13,000 Rubin racks shipping by the end of 2026 and more than 120,000 during 2027. The firm also expects the architecture to represent a much larger share of GPU revenue as the cycle develops.

For investors, the important part would not be the exact rack count. It would be evidence that major AI buyers still view Rubin as economically compelling after incorporating higher memory, power and server costs.

That could also strengthen Nvidia’s margin narrative. If system prices rise partly to offset input inflation while customer demand remains robust, investors may become more comfortable that gross margins can remain structurally above what is normally associated with semiconductor hardware.

The stock has also entered the report with an unusual positioning dynamic. Despite extremely strong fundamental expectations, Nvidia shares remain below their May highs, and the stock has declined the day after each of its last four quarterly reports.

That history means expectations are high, but enthusiasm is not entirely unrestrained. It is another reason this Nvidia earnings preview is more about expectations management than simply whether Nvidia beats consensus.

The upside surprise would be Nvidia demonstrating that a 15% increase in system cost still does not meaningfully alter AI deployment decisions.

That would say more about the durability of the AI investment cycle than another routine quarterly beat.

The Downside Surprise: The Economics Start To Tighten

The downside scenario does not require AI spending to collapse.

It only requires the economics to become incrementally less attractive.

The first pressure point would be margins. Memory inflation is particularly relevant because high-bandwidth memory is an essential component of modern AI accelerators, and the broader Rubin platform combines multiple complex chips, networking components and rack-scale systems. If Nvidia cannot fully pass higher component costs through the ecosystem, gross-margin expectations could begin drifting lower.

The second pressure point is customer capital intensity.

Hyperscalers are already spending extraordinary sums to build AI infrastructure. Nvidia said analysts expect hyperscale capital expenditures to exceed $1 trillion in 2027, while management believes total AI infrastructure spending could eventually reach $3 trillion to $4 trillion annually by the end of the decade.

Those numbers support the bull narrative only as long as customers continue earning attractive returns on the capacity they deploy.

Higher server costs, higher financing costs and higher power costs all push in the opposite direction. They do not necessarily eliminate demand, but they can lengthen project timelines, delay marginal deployments or push customers toward alternative architectures when economics are close.

That is where custom silicon becomes more relevant.

Amazon, Alphabet and other hyperscalers remain major Nvidia customers while simultaneously developing or commercializing internally designed AI accelerators. Those alternatives do not need to outperform Nvidia across every workload. They only need to create acceptable economics for specific high-volume applications.

The third pressure point is Nvidia’s increasingly active role in financing the ecosystem itself.

The company has reportedly agreed to provide guarantees of up to $105 billion related to OpenAI’s Ohio data-center development, while it is also working with partners around a much broader $500 billion AI-financing initiative. Analysts have begun paying closer attention to whether such commitments simply accelerate an exceptionally healthy market or introduce elements of vendor-supported demand.

That distinction becomes more important as system prices rise. In this Nvidia earnings preview, financing quality therefore matters almost as much as the absolute level of demand.

If AI customers can independently finance higher-priced infrastructure because returns remain attractive, the ecosystem looks healthy. If suppliers increasingly need to guarantee financing, invest downstream or support counterparties to maintain deployment velocity, the market may assign a different quality to the resulting revenue growth.

Nvidia already had $145 billion of inventory, purchase commitments and prepaids at the end of Q1, reflecting the enormous scale of the supply chain required to support expected demand.

That is manageable when demand continues accelerating. It becomes more consequential if purchasing behavior begins changing.

The downside surprise would be evidence that AI demand remains large, but is becoming more price-sensitive, financing-dependent or margin-dilutive.

That would not break the long-term AI story. It could, however, change the multiple investors are willing to pay for each incremental dollar of Nvidia earnings.

Beyond Wednesday: Rubin Is The Next Two Calls

This week’s report will describe a quarter dominated by Blackwell.

The next two earnings calls will increasingly describe a company transitioning toward Rubin.

That makes Vera Rubin execution the most important six-to-twelve-month monitoring variable. Nvidia has said production shipments should begin in Q3, accelerate in Q4 and become very large in the following quarter. Management also expects Rubin to outperform Grace Blackwell commercially and has said virtually every major frontier model company plans to adopt it from the outset.

The first signal to watch is whether that schedule holds.

Blackwell demonstrated how technically complex rack-scale transitions can become. Rubin incorporates seven purpose-built chips across five accelerated rack types, so shipment timing, supply availability and system integration will matter as much as silicon performance.

The second signal is customer mix.

Last quarter, Nvidia disclosed approximately $38 billion of hyperscale revenue and $37 billion from its broader AI cloud, industrial and enterprise category. The latter grew faster sequentially.

That diversification matters because price sensitivity may differ dramatically across buyers. Microsoft, Amazon, Meta and Google can absorb higher infrastructure costs more easily than smaller AI clouds or enterprises. If Rubin adoption remains broad across both groups, the economic argument becomes significantly stronger.

The third signal is financing.

Nvidia increasingly describes GPUs and AI infrastructure as financeable assets whose productive lives may extend beyond conventional depreciation assumptions. Last quarter, management noted that rental prices for H100 and even A100 capacity had increased despite newer products entering the market, supporting the argument that older Nvidia infrastructure can remain economically useful.

Investors will need to watch whether that remains true as newer systems become more expensive.

Finally, the CPU opportunity deserves more attention. Nvidia expects Vera to address a roughly $200 billion CPU market and said it had visibility to nearly $20 billion of standalone CPU revenue this year, separate from CPUs bundled into Vera Rubin.

That provides another potential growth vector, particularly as agentic AI creates more orchestration, tool-use and inference workloads.

The next phase of Nvidia’s story is becoming less about selling more GPUs and more about proving that an increasingly expensive full-stack AI factory still generates superior economics.

That is a harder test—but also a more valuable one if Nvidia passes it.

Conclusion: The Headline Beat May Be The Least Interesting Number

Nvidia enters Wednesday with almost everything investors normally want from an earnings setup: extraordinary growth, strong product demand, high margins and a major new architecture approaching production.

Yet that is precisely why the report is difficult.

The market already expects approximately $92 billion of revenue, and some analysts are positioned for substantially more. Another beat may reinforce the existing narrative without materially changing it. What could matter much more is management’s explanation of how Rubin economics, memory inflation, gross margins and customer financing are evolving as the cost of AI infrastructure rises.

Valuation adds another layer.

Nvidia’s LTM P/E stands near 32.88x, compared with approximately 55.95x in July 2025. LTM EV/EBITDA has similarly compressed to around 31.18x, while LTM EV/Revenue sits near 20.36x. The stock still carries premium absolute multiples, but those multiples have declined materially as earnings have caught up with the share price.

That leaves investors with a more nuanced earnings question than whether Nvidia beats consensus.

If Rubin delivers substantially better customer economics while higher prices are absorbed without slowing deployment, the current valuation can be interpreted against another major earnings-growth cycle. If higher memory costs, financing requirements or customer ROI concerns begin reducing deployment intensity, the same multiples may face pressure even if near-term revenue remains strong.

Wednesday is ultimately a test of economic durability, not just quarterly execution.

The most important signals to monitor after the print will be Rubin order visibility, gross-margin resilience, system pricing, customer financing behavior and whether hyperscalers and smaller AI-cloud customers continue expanding deployments at the same pace through the next two quarters.

Disclaimer: We do not hold any positions in the above stock(s). Read our full disclaimer here.