Every few quarters, the debate around Nvidia moves. For a while, it was about whether AI demand was real. Then it was about whether margins could hold. Now, heading into its August 26 earnings report, the question is whether the next product cycle can keep a company already running at this pace from slowing down.
Bank of America thinks it can. And the note behind that view is worth reading before the earnings date arrives.
Table of Contents
Bank of America Nvidia earnings preview and $350 price target
In a note shared with TheStreet on August 7, Bank of America analyst Vivek Arya called Nvidia (NVDA) his top sector pick ahead of the company’s fiscal Q2 FY2027 results. He expects revenue of $94 billion to $95 billion, roughly $3 billion to $4 billion above Nvidia’s own $91 billion guidance. That guidance excludes any China data center compute revenue, meaning actual results could come in higher if modest shipments to that market resume. Third-quarter guidance, he says, should come in at $107 billion to $108 billion, well above the approximately $104 billion Wall Street is currently modeling.
For context, Nvidia reported Q1 FY2027 revenue of $81.6 billion in May, up 85% year over year, with data center revenue of $75.2 billion. The Q2 guidance of $91 billion implied continued sequential growth. BofA’s $94 billion to $95 billion estimate would extend that momentum further.
But Arya isn’t really writing about the next quarter. “The commencement of Vera Rubin next-generation chip deliveries marks the beginning of an extended upgrade cycle spanning multiple quarters,” he wrote. That’s the argument. Not just a beat. A new cycle.
Arya has a $350 price target on Nvidia, representing roughly 56% upside from $223.96 at the time of the note. He points out that the stock is trading at about 16 times forward earnings, its lowest valuation in roughly a decade, even as the earnings trajectory continues to rise, as TheStreet reported.
Why Nvidia Vera Rubin could trigger a multi-quarter upgrade cycle
Nvidia confirmed at GTC Taipei in June that Vera Rubin has entered full production. The platform pairs Rubin GPUs with the new Vera CPU and is expected to be available from cloud partners in the second half of 2026. AWS, Google Cloud, Microsoft and Oracle are already preparing deployments, with OpenAI, Anthropic and SpaceX among the first customers. Nvidia is targeting enough capacity to require 2 gigawatts of power for the buildout, as TheStreet reported.
GPU spot rental prices are near all-time highs, Arya notes in the note. The B200 is running at about $5.66 per hour, the H100 at $2.80 per hour, the A100 at $1.64. That data point matters because it addresses one of the persistent doubts about the AI trade. If customers can still rent compute at those prices and make money from it, they have every reason to keep buying the next generation of hardware. The concern about return on investment fades when the rental market is this strong.
More Nvidia:
The Vera CPU is the part of Rubin that Bank of America finds most interesting beyond the GPU story. An earlier note from the bank called it “the single greatest new addition since the GPU.” Arya’s current note projects Vera CPU sales in the second half of fiscal 2027 at roughly $20 billion, with an annual run rate of $50 billion or more by fiscal 2028. That trajectory, if it holds, would make Nvidia the largest server CPU vendor.
Nvidia gross margin and memory cost inflation outlook for 2027
Memory cost inflation has become one of the louder concerns about Nvidia’s margins. DRAM now makes up 40% to 50% of total production costs, up from 15% to 20% historically. The worry is that as Nvidia moves to more memory-intensive architectures, those rising costs eat into its famously high gross margins faster than the company can price around them.
Arya’s note pushes back on that directly. For Vera Rubin compute racks specifically, the memory cost increase amounts to about 60 basis points of gross margin pressure versus Blackwell Ultra. Gross margins are expected to settle at 73% to 74% over time, down modestly from about 75% now. That is not the kind of structural margin erosion that would break the investment case.
The bigger number is at the pod level. Complete AI pods, which bundle in more memory and storage, could see up to 500 basis points of margin impact. But Arya expects that mix to stay small initially. Nvidia’s long-term supply agreements with SK Hynix and its pricing power, given GPU rental rates near all-time highs, give the company room to pass through costs rather than absorb them.
Nvidia confirmed at GTC Taipei in June that Vera Rubin has entered full productionOta/Getty Images
Nvidia OpenAI circular financing and free cash flow explained
Nvidia has committed roughly $70 billion in direct equity stakes to ecosystem partners. That includes $30 billion to OpenAI, up to $10 billion to Anthropic, and $5 billion to Ilya Sutskever’s Safe Superintelligence. Some investors have questioned whether these arrangements are circular, essentially Nvidia financing the customers who buy Nvidia chips.
Arya’s note addresses this directly. Against the $70 billion in direct investments, Nvidia is expected to generate roughly $470 billion in free cash flow across 2026 and 2027. The $70 billion represents about 15% of that. There is room to keep returning approximately 50% of free cash flow to shareholders while making these investments, according to Benzinga.
The $250 billion backstop tied to an OpenAI and SB Energy campus in Ohio is a different kind of commitment. It is not upfront cash. It is a contingent guarantee that only triggers if OpenAI defaults on the lease, with exposure back-loaded to 2028 and beyond. When Nvidia is expected to be generating $300 billion to $500 billion per year in free cash flow by then, the risk profile looks different than the headline number suggests.
NVDA stock valuation at decade low and the BofA bull case
At 16 times forward earnings, Nvidia is at its cheapest in about a decade. The bank’s EPS projections put Nvidia at more than $13 per share by 2027 and more than $25 by 2030, assuming the AI data center market develops along Arya’s model. That model assumes Nvidia holds more than 70% share of a market the bank sees growing past $1.7 trillion in AI data center systems.
The risks are real. AMD is gaining ground in AI accelerators. The major cloud companies are building more of their own custom chips. China export restrictions remain an overhang. Hyperscaler capital spending could get more uneven if the returns on AI infrastructure disappoint. None of those go away.
But the setup Bank of America is describing is a company with this level of earnings power sitting at a valuation that doesn’t reflect it. Whether August 26 confirms that or not almost misses the point. The bigger question is whether Rubin delivers what Arya’s note says it will. If it does, one earnings report is the start of something, not the thing itself.
