Data Center Silicon Hardware Engineering Intern Co-op - BS

Broadcom vs. Marvell: the AI supercycle is big enough for both. Here's the better buy. Broadcom and Marvell are both gaining from the AI infrastructure boom, but one stock offers a better risk-reward proposition. Key points. * Broadcom and Marvell have emerged as significant beneficiaries of the global AI infrastructure boom. * Marvell's expanded Google relationship has strengthened its custom silicon opportunity. * Broadcom and Marvell are both seeing faster AI-related growth, but trade at very different valuations. * 10 stocks The Motley Fool, LLC like better than Broadcom" Broadcom (AVGO +0.70%) and Marvell Technology (MRVL +1.15%) are both benefiting from the rapid build-out of artificial intelligence (AI) infrastructure. Broadcom's AI semiconductor revenue surged 221% year-over-year to $16.7 billion in the third quarter of fiscal 2026 (ended Aug. 2). Marvell Technology also earned nearly 79% of its revenue from the data center segment in the second quarter of fiscal 2027 (ended Aug 1, 2026), with the segment benefiting from strong AI-related demand. Hence, while both companies appear well-positioned to benefit from the AI supercycle, investors may want to know which stock offers the better mix of growth, profitability, and valuation. Let's find out. Marvell's Google deal does not necessarily weaken Broadcom. Marvell recently expanded its partnership with Alphabet's (GOOG +0.61%) (GOOGL +0.46%) Google to develop several custom silicon products linked to Google's TPU (custom AI chips designed by Google) ecosystem. These chips will help Google run AI models, move and store data, and manage memory more efficiently. Google can buy up to roughly 59 million Marvell shares at $206.58 each. However, most of these shares become available only if Google buys large amounts of Marvell's custom chips. After a small initial portion, the remaining warrants vest gradually as Marvell generates more revenue from Google. Hence, most of the potential equity dilution will happen only if the partnership also brings Marvell substantial business. But, Marvell's growing relationship with Google does not necessarily imply that Broadcom is losing Google's business. Broadcom has entered into a long-term agreement to develop and supply chips for future generations of Google's TPUs. The agreement also covers networking and other components for Google's next-generation AI racks through as late as 2031. Hence, Google appears to be broadening its supplier base as its AI infrastructure requirements expand. Missed Nvidia in 2009? This rare signal is flashing again. In 2009, a "Double Down" signal flashed for a little-known chipmaker called Nvidia. If you'd invested $5,000 then, you'd be sitting on $2,925,679 today.* Now, for the first time in years, that same "Total Conviction" signal is flashing for a company 1/100th the size of Nvidia. It's a key player in the $1.8 trillion space race, and with the stock recently sitting 20% off its highs, the window to get in early is closing fast. *Stock Advisor returns as of September 27, 2026. Marvell Technology Premium Feature Moneyball Superscore Today's Change (1.15%) $2.99 Current Price Key data points. Market Cap Day's Range $257.43 - $267.48 52wk Range $70.69 - $329.88 Gross Margin Dividend Yield Broadcom's AI growth remains strong despite its scale. Broadcom expects its AI semiconductor business to generate about $21.7 billion in revenue in the fourth quarter of fiscal 2026, up 236% on year-over-year basis. The chipmaker expects about $58 billion of AI semiconductor revenue in fiscal 2026. The longer-term expectations for the company's AI business are even more impressive. Management expects AI semiconductor revenue of roughly $115 billion in fiscal 2027 and around $230 billion in fiscal 2028. Broadcom is also generating significant cash from its growth. The company's free cash flow reached $13.7 billion in the third quarter, which was nearly 46% of revenue. The company also has a large infrastructure software business, which generated around $8.8 billion of revenue in the third quarter. Hence, Broadcom has another source of earnings even as its semiconductor business becomes increasingly dependent on AI spending. Marvell's growth still depends on future execution. Marvell's growth outlook has continued to improve. Management now expects fiscal 2027 revenue to grow 45% year-over-year to $12 billion, up from its previous guidance of $11.5 billion. The company also expects fiscal 2028 revenue to rise about 50% year-over-year to $18 billion, compared with its earlier outlook of $16.5 billion. Marvell also expects its Data Center business to grow about 60% in fiscal 2027 and more than 60% in fiscal 2028. Marvell's custom chips business is expected to more than double in fiscal 2028. Connectivity solutions are expected to be the largest contributor to the company's $1.5 billion increase in its fiscal 2028 revenue outlook. The connectivity business is benefiting from stronger demand for optical products, network switches, and technologies that help connect larger AI computing systems. Premium Feature Moneyball Superscore Today's Change (0.70%) $2.45 Current Price Market Cap Day's Range $349.43 - $354.43 52wk Range $289.96 - $495.00 Gross Margin Dividend Yield Broadcom offers a more attractive valuation. Analysts expect Broadcom to generate about $173.5 billion in fiscal 2027 revenue, representing nearly 64% year-over-year growth. Marvell's fiscal 2028 revenue is expected to rise about 51% year-over-year to $18.2 billion. These periods are reasonably comparable, since Broadcom's fiscal 2027 ends in November 2027 while Marvell's fiscal 2028 ends in January 2028. However, Broadcom trades at only about 21.6 times estimated fiscal 2027 earnings, compared with roughly 57.3 times fiscal 2028 earnings for Marvell. Broadcom also trades at around 14.2 times expected fiscal 2027 EBITDA, while Marvell trades at roughly 30.2 times expected fiscal 2028 EBITDA. Hence, investors are paying a substantially higher valuation for Marvell even though its expected revenue growth is not higher. Marvell could still outperform if its custom chip and connectivity businesses grow well above current expectations. But Broadcom does not require the same level of future execution to justify its valuation. Broadcom appears to offer a better risk-reward proposition today. Marvell has significant growth potential, but Broadcom currently offers faster expected revenue growth at a much lower forward valuation. Is Broadcom a smart long-term play? Before you buy stock in Broadcom, consider this: The Motley Fool Stock Advisor analyst team just identified what they believe are the 10 best stocks for investors to buy now... and Broadcom wasn't one of them. The 10 stocks that made the cut could produce monster returns in the coming years. Consider when Netflix made this list on December 17, 2004... if you invested $1,000 at the time of its recommendation, you'd have $383,680!* Or when Nvidia made this list on April 15, 2005... if you invested $1,000 at the time of its recommendation, you'd have $1,382,954!* Now, it's worth noting Stock Advisor's total average return is 937% - a market-crushing outperformance compared to 214% for the S&P 500. Don't miss the latest top 10 list, available with Stock Advisor, and join an investing community built by individual investors for individual investors. *Stock Advisor returns as of September 27, 2026.

AT&S and Marvell Technology expand collaboration to Support next-generation AI Infrastructure. Sep. 22, 2026 4:47 PM ET Source: EQS Group AG | Long-term Agreement Expands Advanced IC Substrate Capacity to Support Growing Demand for AI Infrastructure AT&S Today Announced an Expanded Collaboration With Silicon Valley Company LEOBEN, AUSTRIA - September 22, 2026 (NEWMEDIAWIRE) - Marvell Technology, a leading provider of data infrastructure semiconductor solutions, to increase advanced IC substrate capacity for next-generation AI and cloud infrastructure. The agreement builds on the companies' existing relationship and supports Marvell Technology's growing demand for advanced substrates as AI deployments continue to scale. Marvell Technology is the additional customer identified in AT&S's previously announced expansion of its Kulim manufacturing site. The agreement represents the next phase of the strategic collaboration between AT&S and Marvell Technology and underscores the growing importance of advanced IC substrate technologies as semiconductor designs become larger and more complex. As previously announced, the expansion of the Kulim site includes the fit-out of Plant 2 and the construction of a new manufacturing facility for IC substrate cores and advanced packaging. Supported by long-term customer commitments, the investment will provide additional capacity to meet growing demand for advanced IC substrates and packaging. Building a long-term partnership "Today marks an important milestone in our relationship with Marvell Technology. At AT&S, we are focused on enabling our customers' success through leading-edge technology, manufacturing excellence, and long-term partnerships," said Michael Mertin, CEO of AT&S. "Marvell Technology is one of the semiconductor industry's leading innovators in AI and data infrastructure, and together we are well positioned to capture the significant opportunities created by the global build-out of AI infrastructure. This partnership reflects our shared ambition to drive innovation, create value, and achieve long-term success in the market." "As our data center business continues to expand, securing the manufacturing capacity and advanced technology capabilities required to support our customers remains a critical priority," said Vinay Krishna, Senior Vice President and Chief Supply Chain Officer at Marvell Technology. "Our expanded collaboration with AT&S further strengthens our supply chain foundation and positions us to scale increasingly complex semiconductor solutions, helping us meet the significant growth opportunities driven by the rapid adoption of AI infrastructure." Responding to structural industry shifts The expansion comes as growing investment in AI infrastructure drives significant demand for more advanced semiconductor technologies. As AI systems become larger and more complex, semiconductor architectures are evolving to deliver greater performance, power efficiency, and scalability. The rapid evolution of semiconductor manufacturing is fundamentally reshaping demand for IC substrates. The transition from monolithic chips to chiplet-based designs is increasing the need for advanced substrate technologies across a broader range of semiconductor solutions, including custom accelerators and increasingly complex heterogeneous computing architectures. At the same time, customers are placing greater emphasis on performance, power efficiency and system-level optimization. These trends are driving demand for larger substrate formats, higher integration density, increased layer counts, and faster innovation cycles, all of which play directly to AT&S's technology strengths. Growth supported by long-term customer commitments The expansion in Kulim is supported by long-term customer agreements that provide visibility for future growth while supporting a disciplined investment strategy. "The agreements supporting this expansion provide a solid foundation for our growth strategy and allow us to invest alongside our customers in the technologies and production capabilities required for the AI era. We remain committed to creating long-term value while maintaining financial flexibility and a strong balance sheet," said Gerrit Steen, CFO of AT&S. AT&S Austria Technologie & Systemtechnik Aktiengesellschaft - Advanced Technologies & Solutions AT&S is a global technology company and leading manufacturer of high-end IC substrates and complex printed circuit boards. AT&S develops and produces leading-edge interconnect technologies for key digital industries: AI infrastructure, high-performance computing, mobile devices, automotive, aerospace, industrial and medical technology. With production sites in Austria (Leoben, Fehring), China (Shanghai, Chongqing), Malaysia (Kulim), India (Nanjangud) and a European competence center for R&D and IC substrate production in Leoben, AT&S is actively shaping the digital transformation - through forward-looking investments in research and development and the responsible use of resources. The company currently employs around 15,000 people. Further information can be found at www.ats.net Media download: You can find more resources in the AT&S media portal: https://ats.canto.de/v/press. Media contact: Gerald Reischl, Vice President Corporate Communications Tel: +43 3842 200 4252; Mobile: +43 664 8859 2452; [email protected] AT&S Austria Technologie & Systemtechnik Aktiengesellschaft Fabriksgasse 13 8700 Leoben / Austria www.ats.net |

At ECOC 2026, Marvell's 2nm optical demos point toward 3.2T AI networks - not yet deployments. Marvell arrived at ECOC 2026 with a broad optical pitch for the AI data center: 38 demonstrations, including what it calls the first 2nm 400G-per-lane optical PAM4 demo, a live 2nm 800G ZR/ZR+ pluggable with MACsec, 2nm 1.6T ZR and coherent-lite O-band demonstrations, and a 102.4-terabit-per-second co-packaged-optics platform using 200G-per-lane silicon photonics. The immediate significance is clear. AI clusters are running into networking limits as fast as they run into compute limits, and Marvell wants to show it has a path from today's 1.6T links toward 3.2T connectivity. The harder question is the one buyers actually care about: do these demonstrations show a deployable path to AI-factory networking at the next speed tier, or do they mostly shift the bottleneck to interoperability, thermal design, qualification and cost? On the evidence Marvell has put forward, the answer is mixed. The portfolio looks like a meaningful component roadmap. It is not yet the same thing as volume-ready infrastructure. What Marvell is actually showing. In its ECOC 2026 announcement, Marvell framed this year's lineup as the next step in a progression it has been building for several years: 5nm 200G-per-lane 1.6T Nova DSPs in 2023, a 3nm 1.6T Ara platform in 2024, and 2nm 800G Libra DSPs announced earlier in 2026. That arc matters because it shows the company is not presenting 2nm optics as a one-off lab stunt. It is trying to establish continuity from SerDes and DSP development through pluggables, coherent links and co-packaged optics. Each part of the lineup targets a different piece of the AI-networking problem. PAM4 increases the amount of data carried per optical lane. Coherent ZR and ZR+ serve longer reaches. Coherent-lite is aimed at shorter or intermediate reaches. MACsec, included in the announced 800G design, brings link-layer encryption into the performance conversation. Co-packaged optics moves optical components closer to accelerator and switch silicon, which is why it is being watched closely as port speeds keep rising. That range is one reason the announcement deserves attention. Marvell is not only chasing a bigger single number on a booth placard; it is trying to show relevance across the stack that links accelerators, switches and fiber plant inside and between AI clusters. But it is still a demonstration slate. Marvell did not say these are all shipping products, did not identify which systems are prototypes versus engineering samples versus productized designs, and did not provide the operating details that turn a technology milestone into a procurement decision. Why this matters to AI factories. For AI infrastructure operators, networking is no longer a background line item. Large clusters depend on moving data among accelerators, racks and switching layers without turning expensive compute into idle silicon. Optical links matter here because they can carry more bandwidth over distance than copper, but higher optical speeds do not automatically make a data center cheaper or easier to run. That is why the 2nm claim is important but incomplete. A smaller process node can help enable new bandwidth targets, yet it does not by itself settle the questions that dominate deployment: transceiver cost, thermal behavior, fiber requirements, switch compatibility, manufacturing yield, reliability and serviceability. A live booth demo proves a design can be shown under selected conditions. It does not prove the design can be qualified across a fleet, produced at scale, or supported economically over time. The same caution applies to the 102.4T co-packaged-optics platform. It is a notable signal that Marvell sees co-packaged optics as part of the route from 1.6T toward 3.2T-era AI connectivity. It is not, on the information released, a complete description of a production AI system. Buyers still need to know how that figure maps to switch designs, thermal envelopes, field service models and failure domains. MACsec is another example of why the details matter. Link-layer encryption is increasingly relevant in modern infrastructure, but operators will want to test what it does to power and latency in real deployments, not just whether it can be demonstrated on stage. Where the proof burden shifts next. The cleanest way to understand Marvell's announcement is to compare a component milestone with a network milestone. At the same event, the Ethernet Alliance's ECOC 2026 demonstration spans 400G, 800G and 1.6T Ethernet, 224G SerDes, link training, and an 800G lossless link using Link Layer Retry and Credit-Based Flow Control, with participation from multiple vendors across switches, modules, cables and test platforms. That contrast is the real industry story. Faster DSPs and optical engines are necessary, but they are only one part of an AI-factory network. The burden now shifts to interoperability across vendors, firmware maturity, standards alignment and operational fit at rack and cluster scale. As speeds climb, the limiting layer can move from raw component capability to the less glamorous work of qualification and integration. Marvell's announcement leaves several buyer-grade questions open. There are no published measurements for power per bit, latency, bit-error rate, thermal envelope, optical reach or test duration. The company did not disclose manufacturing yield, production schedule, pricing, named customers or qualification status. It also did not explain how far along each demonstration is on the path from roadmap to revenue hardware. Those omissions do not invalidate the technology. They define the next stage of scrutiny. For operators and system designers evaluating 1.6T-to-3.2T roadmaps, the practical checklist is straightforward: compare solutions at rack and cluster level rather than module level; verify switch, firmware and fiber compatibility; test MACsec's impact on power and latency; and separate an impressive public demo from hardware that is actually qualified for fleet deployment. So Marvell's ECOC showing should be read as a strong signal that optical interconnect is becoming a central competitive layer in AI infrastructure, and that 2nm is now part of that race. It should not be read as proof that the economics and operations are solved. The company appears to be moving the state of the art forward at the component level. The market still has to prove that those gains survive contact with multi-vendor networks, thermal limits and production reality. Isabella Garcia explores the societal impacts of technology for BizTech Weekly. With a background in sociology and a master's in communication, her articles delve into how technology reshapes human interactions, privacy, and its collective futures. Isabella's thought-provoking pieces challenge readers to consider the ethical dimensions of the digital age.