Artificial intelligence has spent years creating a competition for graphics processing units. Bank of America thinks another constraint is becoming increasingly important: memory.
This is the rationale behind analyst Wamsi Mohan’s recent bullish assessment of Penguin Solutions (PENG), an AI infrastructure and memory firm that Bank of America thinks might profit as AI advances beyond training models to the continual operation of them for users and AI agents.
In an Oct. 9 research note seen by TheStreet, Bank of America initiated coverage of Penguin with a Buy rating and a $100 price objective, representing 39.7% upside from the stock’s $71.57 price at the time of publication.
Mohan says Penguin sits at the “intersection of AI and memory” because it both designs and manages large AI computing clusters and sells memory systems intended to improve their performance.
The difference is important because, according to Bank of America, in certain AI implementations, memory capacity and bandwidth are becoming more important constraints than processing power.
According to the bank, AI-related companies now make up 78% of Penguin’s revenues and have increased by 141% over the previous quarter.
The $100 call is effectively a wager for investors that finding additional chips won’t be the next AI roadblock.
It will ensure that the chips have adequate memory to function effectively.
Bank of America says memory is becoming an AI constraint
The simplest way to understand Penguin’s potential is to consider what happens when a query is posed to an AI model.
The system needs more than processing power. It also requires quick access to data to understand and respond to requests. As discussions lengthen and autonomous AI agents perform more complex tasks, more information must be easily accessible.
This makes memory capacity and bandwidth more important for inference, the process by which a trained AI model reacts to users, according to Bank of America.
To solve that issue, Penguin offers MemoryAI key-value cache servers and Compute Express Link, or CXL, devices. One Tier-1 financial institution has already increased its usage of the technology, according to Mohan.
The time-to-first-token metric, which gauges how soon an AI system starts generating a response, is another parameter the bank emphasizes.
Keeping more information close to a graphics processor can reduce that delay. Bank of America argues that Penguin’s memory expertise gives it an advantage as businesses increasingly judge AI systems not simply by how intelligent they are, but by how quickly they respond.
The bank forecasts fiscal 2028 revenue of $2.87 billion and non-GAAP earnings of $5.43 per share, compared with consensus estimates of $2.66 billion and $5.04, respectively.
Its $100 price objective is based on roughly 18 times its calendar 2028 earnings estimate.
“PENG uniquely sits at the intersection of AI and memory,” according to the Bank of America note.
Penguin Solutions has another way to benefit from AI spending
Penguin’s business involves more than just memory.
Additionally, the firm builds, implements, and oversees massive AI computing systems, exposing it to the infrastructure that businesses need after purchasing thousands of pricey GPUs.
Penguin’s latest quarter provides evidence that demand is accelerating. The company reported record fourth-quarter revenue of $567 million, up 68% from a year earlier, while full-year sales climbed 26% to $1.73 billion.
More Artificial Intelligence:
AI’s biggest business opportunity may not look like AI
AI racks are outgrowing old power: 2 chipmakers have a strategy
BofA splits software into winners and losers as AI agents arrive
Additionally, Penguin revealed a new Neocloud client that chose the business to set up and run a 36,000-GPU AI factory in Norway. According to the firm, Lektra, another Neocloud client, chose Penguin to build distributed AI data centers using current carbon-free electricity. Penguin Solutions
This information helps to understand why Penguin’s high-performance computing background is so important to Bank of America.
For years, Penguin has built and maintained large computer clusters for research and other demanding clients. Mohan contends that it is difficult for more recent infrastructure firms to replicate experience.
Additionally, Bank of America cites Penguin’s “land and expand” approach. According to the bank’s research report, the business recruited 17 AI infrastructure clients in fiscal 2026, and 12 of those clients later increased their installations.
This implies that after initially embracing Penguin’s technology, at least some consumers are going beyond trials and increasing their expenditure.
Penguin Solutions gets a 40% upside call on AI’s next problemJUNG YEON-JE / Getty Images
What Penguin investors should watch next
The simple danger associated with Bank of America’s optimistic argument is that the same memory cycle that initially supports Penguin may later turn against it.
The main danger to his argument, according to Mohan, is a significant correction in memory prices.
Penguin’s earnings and profit margins may benefit from rising memory prices, while declining prices may have the reverse impact, especially if the business has already paid more for inventory.
Investors should also pay attention to cash flow.
In fiscal 2026, Penguin had a negative free cash flow of $163 million, primarily due to the need for greater working capital resulting from its rapid growth. In fiscal 2027 and 2028, Bank of America projects that free cash flow would increase to $69 million and $338 million, respectively.
The concentration of customers increases the danger. According to the BofA study, Penguin’s 10 biggest clients accounted for 66% of fiscal 2025 revenue; thus, the timing of a single significant AI infrastructure project may have a significant impact on quarterly profitability.
Because of these risks, investors can evaluate the next phase really easily.
Penguin must continue to secure significant AI implementations, turn those clients into recurring business, and ultimately transform its expansion into more robust revenue generation.
Penguin may contribute in two ways if agentic AI generates the memory demand that Bank of America anticipates: either by providing more of that memory or by assisting clients in constructing the computing systems that rely on it.
The battle to obtain GPUs marked the beginning of the AI boom.
The $100 price objective set by Bank of America is a wager that the next opportunity will come from figuring out what those GPUs need after they are delivered.