Penguin Solutions Raises Guidance Again as Memory Business Surges 111%, But CFO Exit and Thinning Margins Temper the Story
Q3 Fiscal 2026 Earnings Call, July 7, 2026
Penguin Solutions delivered a record third quarter and raised its full-year outlook for the second time in as many months, cementing its pitch to investors that it is becoming an "AI factory platform" company sitting at the intersection of memory and compute infrastructure. Net sales hit $479 million, up 48% year-over-year and 40% sequentially, while non-GAAP diluted EPS came in at $0.84, up 79% from a year ago. But the quarter also revealed some less flattering dynamics: gross margins contracted, the core Advanced Computing segment grew a tepid 4%, working capital consumed $75 million in cash, and the company is losing its CFO effective the day after the call.
Memory Is Now the Growth Engine, Driven by Both Volume and Price
The headline number was Integrated Memory, which posted net sales of $275 million, up 111% year-over-year and 60% sequentially, now representing 57% of total company revenue. Management raised its full-year memory growth guidance to 90-95%, up sharply from prior expectations. CFO Nate Olmstead was direct about the source of that strength: "Both net sales and profits were significantly higher than expected, driven primarily by accelerating AI-driven demand for our memory products." When pressed by Goldman Sachs' Katherine Campagna on how much of the raise was pricing versus new product demand like CXL cards, CEO Kash Shaikh confirmed the outlook reflects "both the volumes as well as the pricing," and clarified that CXL memory expansion cards remain a smaller, faster-growing piece layered on top of the larger core data-center memory module business. Needham's Matthew Calitri pushed further on pricing sustainability, asking whether enterprise customers were starting to balk at higher memory prices. Shaikh's answer was notable for what it implied about the durability of demand rather than any pushback: "While the prices are going up, and it may stabilize at some point, we see increased demand," pointing to backlog that continues to grow faster than shipments.
Preliminary Fiscal 2027 Guide: 30% Growth, But With Caveats
In a move that will likely dominate investor conversations, management offered an early look at fiscal 2027, guiding to approximately 30% growth in both net sales and non-GAAP EPS from the midpoint of the raised FY2026 outlook. This is unusually specific for a preliminary view and signals management's confidence that the current AI-driven demand cycle has legs beyond a single year. However, Stifel's Brian Chin flagged a potential disconnect: if Q4 memory revenue approaches $300 million on a run-rate basis and continues to benefit from pricing, growth could exceed the 30% blended figure, implying a much softer contribution from Advanced Computing. Olmstead confirmed as much, guiding to "mid-teens" growth for Advanced Computing next year, still weighed down by the wind-down of the higher-margin Penguin Edge business and the prior transition away from Meta as a hyperscale customer. Notably, the EPS portion of the FY2027 preliminary guide already bakes in a higher diluted share count from convertible debt dilution, a detail investors should not overlook given shares outstanding are set to jump from roughly 56 million to 62 million in Q4 alone.
Margins Under Pressure Even as Profits Grow
Despite record operating income of $64 million (up 67% year-over-year), non-GAAP gross margin fell to 28.1%, down 3.6 percentage points year-over-year and 3.1 points sequentially. Management attributed this to the ongoing wind-down of the high-margin Penguin Edge business and a mix shift toward lower-margin memory volume, partially offset by favorable memory pricing. The full-year gross margin outlook was nudged up only slightly to 28.5%, and management flagged that Q4 will see "some downward pressure on gross margins" as pricing tailwinds moderate. This is a reminder that even as topline numbers impress, the profitability profile of the business is becoming increasingly dependent on memory pricing cycles that historically have proven volatile.
Advanced Computing Growth Is Weaker Than Headline Numbers Suggest
Advanced Computing net sales were $138 million, up just 4% year-over-year, a stark contrast to the memory business. Within that segment, the non-hyperscale AI infrastructure business grew 81% year-over-year and now makes up 58% of Advanced Computing sales, up from 33% a year ago, but this strength is being masked by the continued exit from hyperscale hardware sales and the Edge business wind-down, which together are reducing total company growth by roughly 14 percentage points and Advanced Computing growth by about 30 percentage points. Investors should treat the segment's official 4% growth figure with caution given how much of the underlying business is being deliberately reshaped away from lower-margin, lower-strategic-value revenue streams.
CFO Departure Adds Uncertainty at an Inopportune Moment
Nate Olmstead is stepping down as CFO effective July 8, moving to "pursue an opportunity in a different industry," with Aaron Johnson, currently VP of Finance and Accounting, stepping in as interim CFO starting July 9 while a formal search for a permanent replacement is conducted. Shaikh sought to reassure investors that "this transition does not change our operating priorities, financial discipline or focus on execution," and both executives emphasized continuity given Johnson's two years working directly with Olmstead. Still, a CFO exit in the middle of an inflection point in the business, right as the company navigates working capital strain, convertible debt dilution, and a complex FY2027 planning cycle, is worth watching closely, particularly if the permanent search extends deep into the new fiscal year.
Cash Flow and Working Capital Strain Are Becoming More Visible
The balance sheet tells a less rosy story than the income statement. Operating activities used $75 million in cash this quarter, compared to $97 million generated in the year-ago period, as receivables ballooned to $704 million from $293 million and inventory rose to $498 million from $184 million, both reflecting higher memory costs and volumes. Cash and short-term investments fell to $440 million, down $295 million year-over-year and $49 million sequentially. While the cash conversion cycle remained within a reasonable 33-day range, the sheer scale of working capital growth underscores that Penguin's memory-led growth story is capital intensive, and investors should monitor whether the company needs external financing to keep funding this expansion as memory volumes and prices continue to climb.
Platform Strategy: Memory Bottleneck as the New Battleground
Much of the call centered on Penguin's positioning as agentic AI shifts infrastructure requirements beyond raw GPU compute. Shaikh's framing was pointed: "Early AI answered questions while agentic AI performs work," and this shift is pushing memory, not compute, to become "one of the primary bottlenecks for large context AI inference performance." The company's MemoryAI KV Cache server, built on Compute Express Link (CXL) technology, is central to this thesis, claimed to deliver up to 2x higher inference performance, up to 8x lower time-to-first-token latency, and memory that is 4x to 5x more cost-effective than GPU HBM. A Tier 1 financial services customer added additional KV Cache servers this quarter for an on-premise AI factory focused on code generation, a use case management flagged as a repeatable pattern for enterprise deployments. The company also highlighted its ClusterWareAI operating system, now expanding into agentic operations tools, and its continued work with Celestial AI (since acquired by Marvell) on photonic memory technology, though these remain earlier-stage initiatives relative to the core memory module business driving current results.
Customer Momentum and Land-and-Expand Traction
Penguin added four new AI infrastructure logos in the quarter and pointed to a broader pattern: across the trailing four quarters, 7 of 13 new AI infrastructure logos have already expanded their spending, while 5 of 16 new memory logos have done the same. Specific wins cited included an expanded Deepgram engagement for voice-AI inference, additional business with a previously disclosed Tier 1 financial institution, a CPU-as-a-Service deployment in South Korea (Haein), a new quantitative trading firm customer, and Spectra, a sovereign supercomputer deployment built with Sandia National Laboratories and NextSilicon. The company was also named an NVIDIA AI Factory Specialized Partner and Dell's 2026 Global Alliances Americas AI Partner of the Year, both signals of ecosystem validation, though these customer wins remain relatively small in absolute dollar terms relative to the scale of memory pricing gains driving the quarter's results.
Penguin Solutions Deep Dive: The Cleanest AI Infrastructure Play Hiding in Plain Sight
The Business Model: From Commodity Memory to AI Factory Architect
Penguin Solutions, formerly known as SMART Global Holdings, operates at the highly lucrative intersection of specialty memory and artificial intelligence infrastructure. The company generates revenue through three distinct segments: Integrated Memory, Advanced Computing, and Optimized LED. The core economic engine of the business relies on purchasing raw memory silicon, such as DRAM and NAND flash, from major semiconductor fabrication plants and engineering these components into high-value, ruggedized memory subsystems. Rather than merely selling these memory modules on the open market, Penguin Solutions integrates them into massive, bespoke high-performance computing clusters and AI data centers. The company designs, builds, deploys, and manages these complex AI factories for its clients, capturing a premium for its engineering expertise, proprietary firmware, and its ClusterWare AI management software. The legacy Optimized LED segment, operating under the Cree LED brand, functions primarily as a cash-generative asset to fund the capital-intensive growth of the AI and computing divisions. By capturing margin at both the component memory layer and the final systems integration layer, Penguin Solutions elevates itself above traditional, low-margin hardware assemblers.
Customers, Competitors, and the Supply Chain Ecosystem
The customer base for Penguin Solutions is currently undergoing a strategic evolution. Historically, the company suffered from heavy customer concentration, relying deeply on hyperscalers, most notably Meta, for massive but lumpy infrastructure orders. Management is actively transitioning the revenue mix to broaden its footprint among enterprise clients, national laboratories, and sovereign wealth entities building localized AI clouds. On the supply side, Penguin Solutions is heavily dependent on upstream semiconductor giants like SK Hynix, Samsung, and Micron Technology for raw memory silicon. Crucially, the company operates as an elite ecosystem partner, recently securing the exclusive designation of NVIDIA AI Factory Specialized Partner, while also serving as a premier AI integration partner for Dell Technologies. The competitive landscape is bifurcated. In the Advanced Computing segment, Penguin Solutions competes against behemoths like Super Micro Computer, Dell Technologies, Hewlett Packard Enterprise, and Lenovo. In the Integrated Memory segment, it faces off against specialty module makers such as Kingston Technology and ADATA, as well as the in-house module divisions of the major memory fabricators themselves.
Market Share and Competitive Advantages
In the broader global server and memory module market, Penguin Solutions is a niche player, commanding approximately a 1 percent overall market share. However, this aggregate figure obscures the company's dominant positioning within the highly specialized sub-segments of bespoke high-performance computing and advanced memory subsystems. The company's primary competitive advantage is its structural uniqueness: it is the only mid-cap technology firm that combines proprietary memory manufacturing with elite systems integration. Pure-play server assemblers are forced to purchase finished memory modules from third parties at standard hardware margins. Penguin Solutions, by contrast, manufactures its own memory subsystems, allowing it to capture the margin at the component layer before integrating it into the final AI cluster. This dual-layer margin capture is reflected in the company's resilient gross margins, which hover around 28 percent, a figure superior to many pure commodity hardware assemblers. Furthermore, the company boasts a 25-year pedigree in high-performance computing. Building an AI factory is not simply a matter of racking servers; it requires solving immense thermal, networking, and memory bottleneck challenges. Penguin's proprietary ClusterWare AI software and deep engineering talent provide a formidable moat against generic IT vendors attempting to pivot into complex AI deployments.
Industry Dynamics: Opportunities and Threats
The artificial intelligence industry is currently undergoing a structural shift from the initial model training phase to the inference and agentic AI phase. This transition represents a massive opportunity for Penguin Solutions. While training massive language models requires brute-force GPU compute, agentic AI workflows and inference tasks are highly memory-bound, requiring vast amounts of fast, efficient memory and general-purpose compute to operate effectively. This dynamic plays directly into Penguin's historical strengths in memory optimization. Additionally, the rising geopolitical trend of sovereign AI, where nation-states mandate that AI infrastructure be built and managed within their own borders, provides a fertile new customer base for bespoke cluster deployments. However, the threats are equally substantial. The business is highly capital intensive, requiring significant working capital to secure memory supply, which has recently driven up accounts receivable and inventory levels. The lumpiness of large-scale AI deployments creates uneven revenue visibility and opens the stock to severe volatility. Furthermore, the deliberate transition away from its historical reliance on Meta introduces near-term execution risk, while the legacy LED business remains highly exposed to cyclical downturns and geopolitical tariff risks.
New Products and Disruptive Entrants
The most consequential technological driver for Penguin Solutions' future growth is the commercialization of Compute Express Link, or CXL, memory expansion technologies. As AI workloads scale, they inevitably hit the memory wall, a bottleneck where processors sit idle waiting for data from memory. Penguin Solutions is aggressively rolling out CXL memory cards and MemoryAI KV solutions, which allow data centers to pool memory across servers, effectively breaking this bottleneck and significantly reducing the total cost of ownership for AI infrastructure. On the competitive frontier, the traditional on-premise AI hardware model faces a credible threat from a new breed of AI-native cloud providers, such as CoreWeave, Lambda Labs, and FluidStack. These well-funded disruptive entrants offer GPU-as-a-Service, allowing enterprises to rent high-performance compute by the hour rather than purchasing bespoke, multi-million-dollar clusters from vendors like Penguin Solutions. While Penguin can and does supply infrastructure to these neo-clouds, a secular shift from enterprise hardware ownership to cloud rental models could compress hardware margins and consolidate purchasing power into the hands of a few aggressive infrastructure-as-a-service providers.
Management Track Record
Chief Executive Officer Mark Adams and President Kash Shaikh have orchestrated a clinical and highly successful portfolio transformation over the last few years. By acquiring Penguin Computing, divesting non-core, low-margin assets like their Brazilian memory manufacturing unit, and officially rebranding the entire entity from SMART Global Holdings to Penguin Solutions in late 2024, they successfully repositioned a sleepy commodity holding company into a pure-play AI infrastructure architect. Management has demonstrated rigorous cost discipline, which was evident in the third quarter of fiscal 2026 when the company delivered a 417 percent year-over-year increase in GAAP operating income despite a highly inflationary memory pricing environment. They have consistently navigated the treacherous cyclicality of the memory market while internally funding the pivot toward artificial intelligence. While the recent abrupt departure of the Chief Financial Officer introduces a minor element of uncertainty, the executive team has built a strong track record of under-promising and over-delivering on their strategic roadmap, proving their ability to execute complex pivots in a rapidly evolving technological landscape.
The Scorecard
Penguin Solutions represents a highly compelling, albeit volatile, vehicle for capturing the next phase of the artificial intelligence supercycle. By owning the intellectual property at the memory layer and possessing the engineering pedigree to integrate those components into massive AI factories, the company has carved out a highly defensible, margin-rich niche. The transition from GPU-heavy model training to memory-bound agentic AI inference perfectly aligns with the company's core competencies in CXL technology and high-performance computing architecture.
However, investors must underwrite the inherent lumpiness of the business model and the execution risks associated with diversifying away from hyperscaler concentration. The working capital intensity required to compete in this space is significant, and the looming threat of AI-native cloud providers shifting enterprise demand from ownership to rental models cannot be ignored. Ultimately, for those willing to stomach the quarter-to-quarter volatility, Penguin Solutions offers a structurally advantaged, back-door play on AI infrastructure that remains underappreciated by the broader market.