Inside Hyperscale
In this edition of the Smart Investor newsletter, we spotlight the connective tissue of AI infrastructure. But first, let’s review the latest news and developments.
1
Portfolio News and Updates
❖❖ Alphabet’s (GOOGL) Google seems to make headlines on a daily basis. Last week, we reported that Google is nearing a deal with two leading U.S. private equity firms, Blackstone and KKR, to provide their portfolio companies easy access to Google’s AI models and Google Cloud AI tools. Blackstone and KKR represent well over $2 trillion in AUM and thousands of portfolio companies, creating an enormous, ready-made customer pipeline for Google. This deal strengthens its competitive position against OpenAI and Anthropic, both of which are pursuing similar partnerships to lock in enterprise demand.
However, the tech giant doesn’t stop at the distribution-level tactics. This week, Google and Blackstone announced a joint venture to build a new U.S.-based AI cloud company. Blackstone is committing an initial $5 billion in equity, becoming a majority owner of the JV, while Google is providing proprietary AI chips – TPUs – as well as software and services. The venture offer data-center capacity as compute-as-a-service, with a goal to bring online 500 MW by 2027 and substantially ramp up compute capacity thereafter.
The new joint venture marks Google’s most ambitious effort yet to monetize its proprietary TPUs with external customers, building on its large-scale deals with Anthropic and Meta. At the same time, it helps ease Google’s own acute compute constraints – a challenge shared by all hyperscalers – as surging internal and external demand has forced the company to scramble for additional data center capacity.
The move positions Google in direct competition with NVIDIA’s (NVDA) ecosystem, particularly the GPU-centric “neoclouds” such as CoreWeave and Nebius. However, it is not primarily about displacing NVIDIA’s market share. TPUs are highly specialized accelerators, and the broader AI training and inference market remains heavily GPU-dominated. Instead, by expanding total AI compute supply at a time when demand for power-hungry models is exploding, the JV is expected to relieve some of the market’s most severe bottlenecks while accelerating Google’s TPU commercialization.
Taken together, the portfolio distribution agreements and the infrastructure JV create powerful synergies, pairing massive demand generation with new supply, and open up a significant additional AI revenue growth runway for Google in the ongoing infrastructure gold rush.
❖ On Tuesday, Google kicked off its annual developer conference, Google I/O 2026, with a strong focus on what CEO Sundar Pichai called the “agentic Gemini era” – a shift from AI assistants that generate content to systems that can proactively complete tasks on a user’s behalf.
GOOGL unveiled Gemini Omni, a new multimodal generative AI family designed to create and edit content across text, images, audio, and video from a unified model architecture. The first release, Omni Flash, enables conversational media generation and editing across multiple input types, signaling Google’s push toward fully integrated multimodal AI workflows. Another centerpiece was Gemini Spark, a cloud-based personal AI agent designed to operate across Google services and the web. Google also introduced Daily Brief, a proactive assistant that summarizes inboxes, calendars, and tasks into a personalized digest.
Search received a major AI overhaul, including generative interfaces, adaptive layouts, and new agentic capabilities designed to help users manage ongoing tasks. Underpinning many of these experiences is Gemini 3.5 Flash, Google’s efficient AI model built for fast and scalable agent workloads, alongside expanded Antigravity tooling for building and orchestrating AI agents. Google also previewed deeper AI integration across Android and Chrome, including Android Halo for live agent updates and Chrome integration for Gemini Spark later this year.
1
❖❖ Microsoft (MSFT) rose after one of the most prominent Wall Streeters, Bill Ackman, revealed that his Pershing Square Capital Management fund has built a new multi-billion-dollar position in the stock, while also making MSFT a core holding in his newly launched PSUS fund.
Ackman argued that worries over Azure are overstated, given the cloud’s 40% growth last quarter and $37 billion AI annualized run rate. He also said M365 durability is not a concern, highlighting its resilience against competition thanks to deep enterprise integration via security, identity, compliance, data systems, bundle economics, and Copilot momentum. The billionaire investor also defended Microsoft’s planned $190 billion spending budget for 2026, which he views as an essential growth investment to meet AI and cloud demand amid supply constraints.
Additionally, Ackman believes that the market is underestimating MSFT’s roughly 27% stake in OpenAI, worth an estimated $228 billion. The recently restructured OpenAI revenue-sharing agreement accelerates $6 billion in 2026 cash flows to MSFT, on top of the potential windfall from the ChatGPT maker’s upcoming IPO. That stake has just become even more valuable, as a federal jury dismissed Elon Musk’s lawsuit against OpenAI and Microsoft, removing a potential $134 billion financial overhang and reinforcing investor confidence in Microsoft’s AI strategy.
1
❖❖ Broadcom (AVGO) and Meta Platforms announced an extended multi-year, multi-generation partnership to support Meta’s custom Training and Inference Accelerator (MTIA) chips through 2029. Building on their existing collaboration, the deal includes co-designing chips, advanced packaging, and high-bandwidth Ethernet networking. It features an initial commitment exceeding 1GW – with a roadmap for multi-gigawatt deployment – targeting lower latency and reduced total cost of ownership across tens of thousands of nodes. The expanded agreement further strengthens Broadcom’s custom AI silicon and networking position and increases its long-term revenue visibility.
AVGO’s AI business has been exploding, as it has become a go-to partner for hyperscalers and AI heavyweights developing custom accelerators. The company now has six major AI customers and line-of-sight to over $100 billion in AI chip revenue in 2027.
Broadcom’s core position in the AI buildout was recently highlighted by Wells Fargo. The firm raised its price target to $545 from $430 in one of the most dramatic large-cap semiconductor revisions of the year. Wells Fargo argued that Wall Street still underestimates the scale of AI infrastructure demand tied to hyperscaler data center expansion. According to the firm’s updated model – which captures rising shipments of custom silicon – AI semiconductor revenue run rate is tracking 30-40% above prior Wall Street expectations. Wells Fargo believes that Broadcom could emerge as one of the major beneficiaries of accelerating AI infrastructure spending as hyperscalers continue building larger AI data center networks.
1
❖❖ Netflix (NFLX) reported strong momentum in its advertising business during its Netflix 2026 Upfront event, revealing its ad-supported tier now reaches over 250 million global monthly active viewers – up significantly from 190 million cited last November. The company highlighted robust engagement, with over 80% of ad-tier users active on a weekly basis. NFLX also announced plans to launch its ad tier in 15 additional countries starting in 2027, further expanding its global footprint.
Additionally, in a major content win, the streamer will air five NFL games next season, including the season opener in Australia and first-ever Thanksgiving Eve game, and has extended its NFL partnership through the 2029-2030 season. The package includes high-profile matchups and the NFL Honors show.
Following the news, several analysts reaffirmed their bullish calls on NFLX. BofA cited rapid growth in Netflix’s ad-supported tier, expecting ad revenue to double in 2026 from last year’s $1.5 billion. Meanwhile, JP Morgan said it is positive on Netflix’s reach and content strategy – adding that it is fast becoming “the global TV” – while its Upfront announcements demonstrate progress towards building a scaled advertising strategy.
1
❖❖ Cisco Systems (CSCO) soared after delivering a blowout fiscal Q3 2026 results and raising full-year guidance, with the details of the release prompting multiple price-target hikes and a market-wide reevaluation of the “old guard” member’s position in the AI narrative.
The company reported record revenue of $15.8 billion, up 12% year-over-year, and adjusted EPS of $1.06, up 10.4% – both exceeding the high end of guidance ranges as well as analyst estimates.
Total product revenue was $12.1 billion, up 17% year-over-year, with total orders growth of 35%. Excluding hyperscalers, orders grew 19% year-over-year – with enterprise up 18% and public sector up 27% – indicating broad-based demand. Cisco also recorded accelerating networking product orders growth, with orders up more than 50% year-over-year, marking the seventh consecutive quarter of double-digit growth. Campus networking orders expanded by over 25%, while wireless grew roughly 40% to a fresh record, with Wi-Fi 7 making up half of the mix.
That said, hyperscaler AI momentum was exceptionally strong, with AI infrastructure orders from hyperscalers totaling $1.9 billion in FQ3. For the first three quarters of the fiscal year, these orders arrived at $5.3 billion, exceeding prior full-year expectations of $5 billion. CSCO now expects approximately $9 billion in AI infrastructure orders from hyperscalers for fiscal year 2026, with approximately $4 billion in related revenue expected to be recognized by the end of this fiscal year.
The AI numbers were what really changed the sentiment, supporting CSCO’s efforts to position itself as the trusted backbone of the AI buildout.
1
❖❖ Keysight Technologies (KEYS) reported the best quarter in company history, capping a record first half with all-time highs in orders, revenue, EPS, and free cash flow.
FQ2 2026 revenue jumped 31% year-over-year to $1.72 billion, above consensus estimates. Gross margin expanded by 770 basis points to an industry-leading 72.3%, while operating margin increased by 820 basis points to 33.3%, supporting analysts’ view that the momentum still has significant runway. Non-GAAP net income rose to $497 million, while non-GAAP EPS climbed to $2.87 – up 69% year-over-year and far above expectations – highlighting substantial operating leverage.
Total orders surged 56% to a record $2.05 billion. Both major segments posted double-digit growth, indicating broad-based demand strength. Communications Solutions Group (CSG) revenue grew 35% year-over-year, driven by 40% growth in commercial communications and 24% growth in aerospace, defense, and government. Electronic Industrial Solutions Group (EISG) revenue increased 24%, including double-digit growth across automotive and energy, general electronics, and semiconductors. Segment operating margins improved to above 33%, reflecting disciplined cost control and pricing power.
Management highlighted broad-based demand strength and order momentum – especially tied to AI infrastructure, next-generation communications, and defense and aerospace spending – while also seeing strength across telecom, automotive, and industrial end markets. As a result, Keysight raised both its FQ3 and FY2026 guidance.
The company expects fiscal Q3 revenue in the range of $1.73-1.75 billion, implying year-over-year growth of 20% at the midpoint, alongside adjusted EPS of $2.43-2.49, representing growth of 43% at the midpoint. Both figures came in well above analyst expectations. FY2026 guidance now includes an additional $375 million in revenue and more than $100 million in run-rate synergies and other operational efficiencies tied to the integration of recent acquisitions. Full-year revenue growth is now expected in the high-20% range, compared with prior guidance for growth above 20%.
1
Portfolio Stocks Under Review
❖ ❖ We are keeping Oracle (ORCL) under review, as the investment case continues to evolve on the back of AI-driven catalysts while investor sentiment remains volatile.
Oracle’s accelerating execution across AI applications and enterprise software remains at the core of the bull thesis. The company continues to expand its agentic AI ecosystem across finance, supply chain, HR, customer experience, and corporate banking, while pushing deeper into autonomous enterprise workflows. ORCL’s broader enterprise franchise also remains solid, with continued leadership positions in areas such as warehouse management and supply chain software, increasingly enhanced with embedded AI functionality.
At the infrastructure level, the narrative continues shifting from speculative AI spending toward execution and monetization. Recent hyperscaler earnings broadly reinforced the view that AI infrastructure demand remains exceptionally strong across the industry, helping support Oracle’s positioning as a secondary hyperscaler and overflow compute provider. The company’s $553 billion RPO backlog remains central to the thesis, while upfront customer commitments, customer-supplied hardware, and partner-backed infrastructure increasingly help offset the capital intensity of ORCL’s AI expansion.
The company also continues expanding its infrastructure footprint and strategic relevance. The Michigan AI data center financing has now formally closed, reinforcing confidence that major Stargate-related projects are moving into execution rather than remaining conceptual. Meanwhile, Project Jupiter in New Mexico continues advancing, strengthening Oracle’s position in large-scale AI infrastructure and power management. ORCL also secured a classified AI contract with the U.S. DoD, adding another layer of validation around the company’s role in secure government and defense infrastructure.
The multicloud strategy is becoming increasingly important as well. Oracle continues deepening partnerships with AWS and Google Cloud, reinforcing a platform-agnostic positioning where OCI increasingly acts as a data and AI layer across multiple ecosystems instead of competing solely as a standalone cloud provider. Strategically, this broadens its addressable market and reduces dependence on any single customer or platform. Importantly, ORCL today is far less of a single-customer story than it was just a year ago, with expanding exposure across hyperscalers, enterprise workloads, and government demand.
At the same time, the financing debate has not disappeared. Oracle’s AI buildout still requires enormous capital commitments, and the company’s debt-funded expansion continues to make investors uneasy. Reports around pressure on bank balance sheets and the scale of future funding needs reinforce that execution, financing discipline, and deployment timing remain critical variables.
That said, Wall Street sentiment continues to improve. Recent analyst upgrades and price target increases increasingly frame ORCL as a foundational AI infrastructure provider, with growing confidence that consensus cloud and earnings expectations may still be too conservative. Several analysts now argue that the market is underestimating Oracle’s long-term earnings power if even a portion of its backlog converts successfully into recurring AI-related revenue.
All in all, the fundamental story continues to strengthen, supported by improving monetization visibility, expanding multicloud relevance, and continued infrastructure execution. However, ORCL currently appears to trade increasingly in line with broader AI infrastructure sentiment and overall market risk appetite. With the AI infrastructure trade remaining powerful while geopolitical headlines around the Iran war continue driving rapid shifts in market sentiment, we prefer to keep ORCL under review until the broader macro picture becomes at least somewhat clearer.
1
Portfolio Earnings and Dividend Calendar
❖ The Q1 2026 earnings season is ending, but several Smart Investor Portfolio holdings are expected to reveal their results in the coming week. These include one of the most important reports on the market – NVIDIA (NVDA), scheduled for today. Additionally, Palo Alto Networks (PANW) reports on May 26 and Synopsys (SNPS) – on May 27.
❖ The ex-dividend date for Microsoft (MSFT) is May 21, while for RTX (RTX) it is May 22.
w
1
New Buy: Credo Technology (CRDO)
Credo Technology Group Holding Ltd. operates in one of the most critical bottlenecks inside modern AI infrastructure – moving massive amounts of data quickly, reliably, and with manageable power consumption. The company develops high-speed connectivity and optical interconnect solutions that help link GPUs, switches, servers, and memory systems inside hyperscale and AI-focused data centers, where bandwidth demands are rising faster than power budgets can comfortably support. Rather than competing in processors or full networking systems, CRDO focuses on the connective tissue that allows increasingly complex computing clusters to function efficiently at scale. Its technologies are designed to improve signal integrity, reduce power consumption, and support the transition toward faster network architectures, positioning the company near the center of the accelerating buildout surrounding AI training and cloud infrastructure.
1
The Hidden Layer
Credo Technology was founded in 2008 around a highly specific engineering problem that initially attracted far less attention than processors or networking hardware themselves: how to move rapidly growing amounts of data across increasingly dense systems without overwhelming them with power consumption, heat, and signal degradation. In the early cloud era, those constraints were manageable. But as hyperscale infrastructure expanded and AI workloads began scaling across massive GPU clusters, data movement itself emerged as one of the defining bottlenecks inside modern computing.
For much of its early life, CRDO remained a relatively small, engineering-driven company focused heavily on serializer-deserializer, or SerDes,1 intellectual property and physical-layer connectivity technologies. The business spent years operating below the visibility of larger semiconductor names while building expertise in high-speed signal integrity and low-power data transmission. Profitability and large-scale commercial traction came later, particularly as AI infrastructure spending accelerated after 2023.
The company’s broader industry relevance grew materially as hyperscalers pushed toward higher-speed Ethernet architectures and denser AI clusters, where power efficiency became nearly as important as raw bandwidth. Credo emerged as one of the early leaders behind Active Electrical Cables, or AECs,2 helping popularize the category through its ZeroFlap architecture3 designed to extend reach, lower power consumption, and improve reliability compared to traditional passive copper solutions. That positioning moved the company beyond niche IP supply into a more strategic role inside next-generation AI networking infrastructure.
The company’s 2022 public listing provided additional capital to accelerate engineering investment, scale product development, and deepen relationships with hyperscalers and tier-one networking vendors. Management largely maintained a disciplined strategy centered on organic growth and targeted technology expansion instead of broad diversification or large-scale acquisitions.
That philosophy remains intact today. CRDO first expanded beyond its internally developed technology base with the 2025 acquisition of Hyperlume, which added microLED-based optical interconnect technology for next-generation AI clusters. In March 2026, the company acquired CoMira Solutions, adding protocol, error-correction (ECC), and security IP that deepened its system-level connectivity capabilities. One month later, Credo announced the significantly larger DustPhotonics acquisition, bringing advanced silicon photonics technology in-house as AI infrastructure shifts toward integrated optical networking. Together, these targeted moves broaden Credo’s platform across both electrical and optical domains while preserving its disciplined, organic-first approach to growth.
—
1 – SerDes (serializer-deserializer): A chip technology that converts data between serial and parallel formats, enabling exceptionally high-speed communication between components inside data centers and networking systems.
2 – AECs (Active Electrical Cables): High-speed copper cables with built-in signal-processing chips that improve reach, reliability, and power efficiency compared to traditional passive cables.
3 – ZeroFlap architecture: Credo’s proprietary AEC design architecture aimed at reducing signal errors, lowering latency, and improving power efficiency in high-speed AI and cloud networking environments.
—
1
Bandwidth Is Power
Credo Technology has become one of the key specialists in high-speed AI connectivity infrastructure, supplying the technologies that allow GPUs, switches, servers, and memory systems to communicate reliably at increasingly extreme data rates. As hyperscalers scale AI clusters toward hundreds of thousands of accelerators, the focus across the industry has expanded well beyond raw compute performance alone. Networking efficiency, power consumption, latency, signal integrity, and memory connectivity have all emerged as critical constraints in large-scale AI deployments – positioning Credo directly inside some of the most important infrastructure bottlenecks of the AI buildout.
At the center of the business are Active Electrical Cables, or AECs, which remain Credo’s flagship product and primary revenue driver. AECs embed active electronics directly into copper cables, allowing them to deliver substantially higher speeds, lower latency, and greater reliability than traditional passive copper solutions while consuming materially less power than optics over shorter distances. In large AI clusters, where even a brief link failure can idle thousands of GPUs simultaneously, that reliability advantage becomes economically meaningful. Management continues describing the industry as still being in the early stages of AEC adoption, with hyperscalers increasingly standardizing around the technology for short-reach AI connectivity. Industry estimates now place Credo’s AEC market share near 90%, reinforcing its position as one of the category’s defining players.
The company’s role, however, increasingly extends far beyond cables alone. Credo’s broader product portfolio now spans optical DSPs, retimers, PCIe connectivity, telemetry software, silicon photonics, and memory interconnect solutions. The common thread across all of them is infrastructure efficiency – reducing power consumption, improving uptime, stabilizing links, and increasing accelerator utilization inside increasingly dense AI environments. As AI spending becomes more scrutinized, hyperscalers are placing greater emphasis on token economics, thermal efficiency, and networking reliability rather than simply adding raw compute capacity. That shift directly strengthens Credo’s positioning.
Importantly, CRDO is no longer simply a “copper story,” increasingly leveraging its ZeroFlap platform across both electrical and optical connectivity architectures. Management has consistently pushed back against the idea that future AI infrastructure will rely on a single connectivity architecture. Instead, the company increasingly frames the future as a heterogeneous environment where copper, pluggable optics, Active LED Cables, near-package optics, and eventually co-packaged optics coexist depending on reach, power, latency, and economics. That dual-roadmap strategy may be one of Credo’s biggest competitive advantages because hyperscalers are unlikely to standardize around a one-technology-only approach across all AI deployments.
That positioning helps explain the company’s aggressive push into optics and broader system-level connectivity. The DustPhotonics acquisition materially strengthens Credo’s optical roadmap by bringing silicon photonics technology in-house and expanding exposure to 800G, 1.6T, and eventually 3.2T optical architectures. Combined with the company’s existing optical DSP business and ZeroFlap optical transceivers, the deal moves CRDO toward a more vertically integrated optical connectivity platform spanning SerDes, DSPs, silicon photonics, and system integration. Management now expects the combined optical portfolio to generate more than $500 million in revenue during fiscal 2027, reinforcing optics as a major future growth engine rather than a side business.
At the same time, Hyperlume expands Credo into Active LED Cables, or ALCs, which aim to combine optical reach with AEC-like efficiency and reliability characteristics. CoMira Solutions added protocol, error-correction, and security IP capabilities that deepen the company’s move toward system-level connectivity infrastructure rather than remaining solely a physical-layer component supplier.
The company is also expanding beyond networking interconnects themselves. PCIe Gen6 retimers and AECs position Credo deeper inside modular AI server architectures, while OmniConnect extends the company toward memory and XPU connectivity as inference workloads become increasingly memory constrained. Blue Heron, its 200G retimer platform, further expands exposure into emerging AI scale-up architectures supporting Ethernet, UALink, and ESUN protocols.
Together, these offerings increasingly position CRDO as a broader AI connectivity platform spanning copper, optics, memory interconnects, telemetry software, and system-level networking infrastructure. Its PILOT telemetry software platform further differentiates the company by enabling real-time diagnostics, predictive monitoring, and link-health analytics designed to improve uptime and reliability across large-scale AI clusters. That evolution substantially expands the company’s addressable market while embedding Credo deeper inside hyperscaler architecture decisions and long-cycle AI infrastructure deployments.
Customer concentration remains an important risk, and large competitors including Broadcom (AVGO) – a Smart Investor Portfolio holding – and Marvell continue investing aggressively across adjacent connectivity markets. Hyperscalers also have incentives to develop secondary suppliers or internal alternatives over time. Still, Credo currently benefits from a meaningful technological lead in reliability, power efficiency, and system integration – advantages that become increasingly valuable as AI systems grow denser, more distributed, and more uptime-sensitive.
1
Backplane Gains
Credo’s financial profile increasingly reflects a company that has moved beyond speculative AI exposure into a phase of profitable hypergrowth powered by hyperscale deployment ramps, widening platform adoption, and exceptional operating leverage. Revenue growth remains extraordinary, but what increasingly differentiates CRDO is that profitability, cash generation, and balance-sheet strength are scaling alongside the business rather than deteriorating under expansion pressure.
That shift became especially visible during fiscal Q3 2026. Revenue surged 201.5% year-over-year to $407.0 million, extending a broader pattern of upside execution, with CRDO now beating analyst revenue estimates for seven consecutive quarters and adjusted EPS expectations in every quarter since fiscal Q1 2024. Non-GAAP EPS soared 328% to $1.07, materially ahead of consensus, while non-GAAP operating income climbed to $201.8 million and operating margin expanded to 49.6%. Non-GAAP net margin exceeded 51% – an unusually high level for a semiconductor infrastructure company still growing at breakneck pace.
The acceleration has also been remarkably fast. Fiscal 2024 revenue totaled only about $193 million and still included net losses. Revenue then rose to roughly $437 million during fiscal 2025 before surpassing a $1.3 billion annualized run rate in fiscal 2026, illustrating how quickly large-scale AI deployments are expanding Credo’s opportunity set. The company increasingly combines characteristics rarely seen together in semiconductor infrastructure: hypergrowth, software-like operating leverage, strong gross margins, and meaningful free cash flow generation.
Cash generation has become one of the clearest indicators of that transition. Operating cash flow reached a record $166.2 million during Q3, while free cash flow approached $140 million. CRDO is no longer being valued primarily on future potential – it is already generating substantial earnings and cash from the AI infrastructure cycle. The company exited the quarter with roughly $1.3 billion in cash and short-term investments, giving management flexibility to fund acquisitions, inventory expansion, customer ramps, and elevated R&D spending simultaneously.
Gross margins increasingly sit at the center of the investment debate because they shape whether investors view Credo as a differentiated AI infrastructure platform or simply another hardware supplier exposed to eventual commoditization pressure. Q3 non-GAAP gross margin reached 68.6%, an unusually high level for a semiconductor company scaling this quickly. Management guided fiscal Q4 gross margin toward 64-66% as optics, newer platforms, and broader hyperscaler deployments become a larger part of the revenue mix. While that moderation sparked concerns about long-term margin durability, Credo is still operating at profitability levels that remain exceptional across the broader AI infrastructure ecosystem even while aggressively expanding into new connectivity layers.
Guidance reinforced the view that demand remains exceptionally strong. Management projected fiscal Q4 revenue of $425-435 million, ahead of analyst expectations and implying another quarter of rapid growth following a record Q3. For FY2026, management reiterated expectations for revenue growth above 170% year-over-year. More importantly, management expects fiscal 2027 revenue to grow more than 50% year-over-year, supported by continued AEC adoption, expanding optical deployments, PCIe Gen6 ramps, and broader hyperscaler programs. The company also expects the DustPhotonics acquisition to become accretive to non-GAAP EPS during fiscal 2027, while combined optical revenue is projected to exceed $500 million.
The balance sheet and supply-chain position further strengthen the story. Inventory rose to roughly $208 million as management prepared for future demand ramps and secured supply across advanced manufacturing nodes including 12nm, 7nm, 5nm, and 3nm. R&D spending also increased sharply as Credo simultaneously expands across optics, memory interconnects, telemetry software, retimers, and broader system-level connectivity infrastructure. The company is no longer scaling a single-product business – it is building a broader AI connectivity platform designed to participate across multiple layers of future hyperscale architectures.
Risks remain substantial, particularly around customer concentration. The top three hyperscaler customers still account for more than 80% of revenue, leaving the business highly exposed to AI spending cycles and customer deployment decisions. At the same time, recent expansion to a fifth hyperscaler customer suggests adoption is gradually broadening beyond the company’s earlier concentrated base.
Credo enters its next phase from a position of unusual financial strength for a company at this stage of growth – combining explosive revenue expansion, elite margins, substantial free cash flow generation, a large net-cash position, and growing exposure across multiple layers of AI connectivity infrastructure rather than a single product cycle.
1
The Bottleneck Trade
Credo operates within a relatively narrow segment of the semiconductor industry focused on high-speed AI connectivity infrastructure, making its closest peers more specialized than traditional chip companies. Astera Labs remains the clearest pure-play comparison, given its similar growth profile and focus on removing data-movement bottlenecks inside hyperscale AI systems. Marvell serves as the scaled incumbent reference across networking silicon, optical DSPs, and interconnect architectures, while MACOM provides a useful benchmark for physical-layer connectivity and optical infrastructure. Rambus adds another comparison point through its licensing-heavy model and growing exposure to memory and interconnect technologies tied to AI architectures. Together, these peers frame CRDO as a specialized AI infrastructure platform positioned around the growing importance of networking efficiency, bandwidth scaling, and data movement inside hyperscale systems.
As the AI buildout continues to expand, stocks sitting at its major pressure points are doing well. Credo, along with several of its closest peers, has clocked triple-digit gains over the past year, with the biggest differences between them showing up in their volatility rather than performance. A high-beta stock, CRDO can look risky, with sharp daily moves often driven more by market-wide macro or geopolitical headlines than by company-specific news or fundamentals. That’s exactly why the Smart Investor sold CRDO in early March after holding it for only a short time in the Portfolio, driven by geopolitical fear and sentiment shock. The Iran war broadly hit high-growth AI infrastructure names and compressed sentiment across the sector, with Credo and similarly volatile peers taking an outsized hit. That turned out to be a mistake, as the market quickly realized that Credo’s growth trajectory was still accelerating underneath the selloff.
Moreover, the company’s strategic position has materially improved over the past several months. Credo is no longer simply an AEC or “copper connectivity” story. The business has evolved into a broader AI connectivity platform spanning copper, optics, DSPs, telemetry software, and memory interconnect infrastructure. At the same time, hyperscaler adoption continued broadening, optical initiatives accelerated faster than expected, and the company deepened its positioning across multiple layers of next-generation AI networking architectures.
The stock subsequently rallied sharply, and we waited for a more favorable entry point before re-establishing the position. That opportunity emerged during last week’s valuation-driven pullback and broader profit-taking across high-growth semiconductor names. As a result, CRDO is returning to the Portfolio as a complementary position alongside Broadcom. While AVGO provides a diversified, lower-volatility AI infrastructure core holding, Credo offers more concentrated exposure to the networking and connectivity bottlenecks increasingly shaping hyperscale AI deployments.
Valuation remains elevated relative to the broader Technology sector, though the premium is supported by extraordinary growth, elite margins, and rapidly expanding cash generation. Importantly, the recent pullback reduced some of the excess embedded in the stock’s earlier momentum-driven valuation. On a forward non-GAAP P/E basis, CRDO now trades broadly in line with Marvell and Rambus while remaining well below Astera Labs and MACOM. A similar setup appears across forward EV/EBITDA and EV/Sales multiples. Perhaps most notably, Credo’s forward PEG ratio of roughly 0.4 remains exceptionally low both in absolute terms and relative to peers, signaling that a meaningful portion of CRDO’s growth trajectory may still remain underpriced despite the stock’s volatile upward climb.
1
Investing Takeaway
Credo is increasingly evolving beyond its original identity as an AEC specialist into a broader AI connectivity platform spanning copper, optics, DSPs, telemetry software, and memory interconnect infrastructure. Its advantage no longer rests solely on bandwidth speeds, but on enabling hyperscalers to scale AI clusters more efficiently through lower power consumption, higher reliability, and improved networking performance. As AI infrastructure grows denser and more distributed, connectivity itself is becoming one of the industry’s defining bottlenecks – positioning CRDO directly inside a critical layer of next-generation architectures. The company still carries meaningful risks, particularly around customer concentration and valuation volatility, but it now operates from a position of accelerating growth, elite profitability, strong cash generation, and expanding platform breadth. Following the recent pullback, the stock once again appears increasingly aligned with Credo’s underlying growth trajectory and long-term AI connectivity opportunity.
1
New Sell: SPX Technologies (SPXC)
We are selling SPX Technologies after a strong operational run because the macro and market backdrop for industrial infrastructure companies has shifted, reducing the near-term risk-reward despite continued execution by management. SPXC remains a fundamentally strong company with meaningful exposure to long-term data center cooling demand, but the environment that supported the original thesis has become less favorable.
Operationally, the company continues to perform well. First-quarter revenue rose 17.4% year-over-year to $566.8 million, while adjusted EPS increased 22.5% to $1.69. Adjusted EBITDA grew 22.9%, with margins expanding 90 basis points to 22.2%. HVAC revenue increased 22%, driven by strong data center demand and recent acquisitions, while backlog in the segment rose 38% organically. Management also raised full-year guidance, now expecting roughly 15% revenue growth and 18% adjusted EPS growth at the midpoint.
In many respects, the original investment thesis played out as expected. SPX successfully repositioned itself from a diversified industrial business into a more focused thermal-management and infrastructure platform increasingly tied to data center demand. The market recognized that transition, leading to a significant re-rating in the stock. Analysts broadly acknowledged the progress following earnings, with multiple firms lifting price targets on expectations for continued capacity expansion and growing exposure to data center cooling infrastructure.
The issue now is not company-specific – it is macro-driven. Rising oil prices tied to the Iran conflict, surging Treasury yields, and persistent inflation concerns have pressured the broader industrial sector, particularly companies trading at premium valuations. While AI infrastructure leaders continue attracting aggressive dip-buying, SPXC sits in a more difficult middle ground – too AI-adjacent to trade like a traditional industrial, but not deeply embedded enough inside the core AI supply chain to receive the same defensive market positioning as names such as Vertiv.
That backdrop matters because the valuation argument has changed. SPXC now trades above both sector averages and many of its own historical valuation ranges, despite continued execution risks tied to capacity expansion, tariff exposure, input-cost inflation, and margin pressure from ongoing HVAC buildouts. The long-term story remains intact, but in the current environment the upside asymmetry appears less compelling. We will continue monitoring the company and reassess if the macro backdrop improves and valuation becomes more attractive relative to fundamentals.
1
Smart Investor’s Winners Club
The Winners Club represents stocks from the Smart Investor Portfolio that have risen at least 30% since their purchase dates.
Markets have been up and down, but the Club member count remains unchanged at 25 stocks: AVGO, GE, TSM, EME, HWM, ANET, VRT, IBKR, APH, MTZ, GOOGL, STRL, ORCL, PH, ASX, CSCO, KEYS, CRWD, BK, JBL, PANW, ATI, MS, NVT, and RTX.
The first runner-up is still SNPS with a 24.12% gain since we purchased it in early April. Will it break into the winners’ circle, or will another stock outrun it to the finish line?
1
|
