Penguin Solutions targets AI memory bottlenecks with new hardware

Ask anyone about the AI memory boom this year and the conversation usually follows a script. GPUs come up first, then Micron Technology, and if someone mentions flash storage, Sandisk enters the chat.

Penguin Solutions (Nasdaq: PENG) rarely gets a line in that script. The company is now making a fairly loud case that it deserves one.

On November 2, 2026, Penguin introduced its latest SMART Modular DDR5 Multi-Ranked Dual In-line Memory Modules, known as MRDIMMs. The modules run at DDR5-12800 speeds and are built to ease the memory bandwidth bottlenecks that slow down AI data centers.

What Penguin actually launched

Memory bandwidth measures how quickly data can move between a server’s memory and its processors.

Penguin’s MRDIMMs widen the road with an engineering trick. Each module multiplexes two physical ranks into a single logical rank.

Put simply, two banks of memory take turns serving data so quickly that the system treats them as one faster bank. According to the company, this boosts bandwidth without the speed penalties that multi-rank designs have traditionally carried.

The target workloads are generative AI and large language model inference. Inference is the part where a trained model actually answers questions, and it is notoriously hungry for memory.

The module launch builds on a product Penguin rolled out in March 2026: the MemoryAI KV cache server. That machine uses Compute Express Link, or CXL, an interconnect standard that lets servers tap pools of memory beyond what sits directly beside the processor.

The MemoryAI server delivers up to 11 TB of memory. Its design is aimed squarely at GPU memory constraints during heavily memory-driven workloads.

A KV cache is the working memory a language model keeps while it processes a conversation. The longer the context, the bigger that cache gets, and the more pressure it puts on scarce GPU memory.

The numbers behind the pivot

Penguin’s financials suggest the AI bet is no longer a side project. AI-related revenue now makes up 78% of the company’s total sales.

The memory segment has been the standout. In one recent quarter, it grew 158% year over year.

The company posted record revenue of $567 million for Q4 FY2026, up 68% from the prior year. Full-year sales came in at $1.73 billion, a 26% increase.

Those two growth rates tell a story on their own. A 68% jump in the final quarter against 26% for the full year means growth was heavily back-loaded, with momentum building as the year went on.

Penguin also reports strong order backlogs that stretch across multiple quarters.

Bank of America initiated coverage in October 2026 with a Buy rating and a $100 price target, which the bank’s figures put at approximately 40% upside.

The bank’s analysts highlighted Penguin’s dual role. The company designs AI clusters and also supplies the specialized memory systems that go inside them.

From SMART Global to Penguin

Penguin Solutions was formerly known as SMART Global Holdings. The name change tracked a broader shift toward AI infrastructure, and the SMART Modular brand still lives on in its memory products.

The company also brings operational experience to the table. Penguin has managed more than 100,000 GPUs, logging over 4 billion hours of operational runtime.

What this means for investors and the AI stack

With 78% of sales tied to AI, Penguin’s fortunes are now closely bound to the pace of AI infrastructure spending.

Investors will likely watch three things. First, whether the new MRDIMMs win meaningful adoption in AI servers. Second, whether the memory segment can sustain anything close to its 158% growth pace. Third, whether that multi-quarter backlog converts into revenue on schedule.

For now, Penguin has made its argument clear. The GPU gets the glory, but the company is wagering that the memory feeding it will decide how fast AI actually runs.