Hello world, it’s Thursday, October 29th.
Marvell is in advanced talks to acquire Lumentum, a combination that would give it a vertically integrated 1.6T co-packaged optics stack. Elsewhere, Samsung and Alphabet both missed earnings estimates even as Microsoft Azure posted 48% year-on-year growth, the DOJ is making inquiries into Nvidia’s roughly $13 billion deal to buy Hugging Face, and SK Hynix crossed 100 trillion won in quarterly revenue for the first time.
Let’s get into it. — Austin & Vik
Be sure to check out the Semi Doped podcast on YouTube or your favorite podcast player!
Presented by Crusoe
Crusoe builds the full stack for AI, from power and data centers to the cloud. It can also help you tune a small model on your own data for just a few bucks.
For repeatable work like reports, an open model trained on your terminology and formats can match a frontier model, and it’s cheaper and faster to run.
Crusoe’s Serverless Fine-Tuning handles the training: upload your dataset, pick an open model like Llama, DeepSeek, or GLM, and pay per million tokens used, not per GPU hour.
About $2 buys roughly 5M training tokens on smaller models.
Deploy on Crusoe Cloud in one click or download the weights and deploy on your own.
Marvell Nears Deal to Buy Lumentum in AI Optics Push
Marvell Technology is in advanced talks to acquire Lumentum Holdings, the data center optics maker, according to people familiar with the matter, a combination that would fold optical transceiver hardware directly into Marvell’s DSP silicon portfolio. The pairing would give Marvell a vertically integrated 1.6T co-packaged optics stack, letting it offer a single-vendor solution for the high-bandwidth interconnects that link GPU clusters at scale. Lumentum’s optical engines plus Marvell’s switching silicon would put the combined company in direct competition with Broadcom across the AI networking supply chain. No deal terms have been disclosed.
Vik: We just recorded a podcast discussing this with our friends at Cignal AI. Check back tomorrow for some instant reactions on this!
Samsung and Alphabet Earnings Miss, Shaking Confidence in AI Revenue Payoff
Alphabet shares fell roughly 10% in early trading after the company reported weaker-than-expected search and advertising sales, with investors questioning whether AI threatens its core business rather than bolsters it. Samsung compounded the mood: its chip division’s operating profit missed analyst estimates, and its stock slid as much as 6.6% when management declined to announce a shareholder return plan. The two misses landed back-to-back, reviving broad doubts about whether current AI spending translates into near-term earnings. Samsung did commit to hiking capital expenditure more than 60% this year after selling out of high-bandwidth memory capacity through 2027, and Google Cloud revenue doubled to $30.7 billion as Alphabet’s total CapEx reached $51.4 billion. Infrastructure spending is accelerating; profit conversion, for now, isn’t.
Austin: GenAI is good for Google search. User experience is better, intent is better understood, personalization is better. Google just needs to keep focusing on building out monetization opportunities in the AI search experience. That’s just part of the process. Reminds me of Instagram. First you make a great experience, then figure out how to tastefully ramp up the ads within the Instragram feed.
Microsoft Azure Jumps 48%, Posting Strongest Cloud Growth in Years
Azure revenue reached $33.17 billion in the quarter, a 48% year-on-year gain that cleared analyst consensus by a comfortable margin. Microsoft’s total cloud business beat estimates, according to Bloomberg, as AI workload demand filled capacity faster than the company had guided. Capital expenditure commitments have been pulled forward accordingly, tightening GPU allocation timelines out through 2028. Free cash flow fell to $18.6 billion, reflecting that accelerated spending rather than any softness in the core business. The result arrives days after Alphabet’s cloud miss rattled sentiment, and Microsoft’s numbers land as a direct rebuttal: at least one hyperscaler is converting AI investment into revenue at scale right now.
DOJ Scrutinizes Nvidia’s $13 Billion Hugging Face Acquisition
The US Justice Department is making inquiries into Nvidia’s agreement to acquire Hugging Face for roughly $13 billion, according to Reuters, which first reported the development citing unnamed sources. Hugging Face operates the dominant open-weight model repository, giving it a central role in how developers across the industry access and distribute AI models. The inquiry fits a pattern of widening antitrust attention on Nvidia that has previously centered on its dominance in GPU markets. Owning Hugging Face would extend Nvidia’s reach from chip hardware into the software layer where open models are hosted and shared, a vertical combination that has drawn the DOJ’s attention.
SK Hynix Crosses 100 Trillion Won Revenue Mark as New CEO Steps Forward
SK Hynix posted third-quarter revenue of 104.62 trillion won (~$78 billion), clearing 100 trillion won for the first time in the company’s history, with operating profit hitting a record 83.33 trillion won. The numbers arrived alongside a notable personnel moment: CEO Kwak Noh-jung joined the earnings conference call for the first time since taking office, using the appearance to outline a shareholder return programme worth roughly 100 trillion won in dividends and buybacks. His debut was widely read as an attempt to rebuild investor confidence after shareholders pushed back against a proposed IPO of one of the company’s units. Records set, fences mended.
KLA Posts Fiscal Q1 Results as Process-Control Demand Tracks AI Ramp
KLA Corporation reported its fiscal 2027 first-quarter results, with the print arriving as the first major equipment-sector earnings of the current AI-driven capital expenditure cycle. The company’s numbers carry particular weight this season because process-control intensity, not just wafer-start volume, is the variable investors are watching: if equipment spending per wafer is rising at leading-edge fabs, KLA’s guidance would be the first hard evidence of it. KLA functions as a bellwether for the broader semiconductor equipment industry, given its dominant position in inspection and metrology tools that customers cannot defer when shrinking geometries demand tighter defect budgets.
GE Vernova Acquires Trench Group for $9 Billion, Raises Revenue Outlook
GE Vernova agreed to pay $9 billion for Trench Group, the Siemens-spun German manufacturer of high-voltage transformers, reactors, and bushings that has become a critical bottleneck for AI data-center grid connections. Trench’s products carry some of the longest lead times in grid infrastructure, and buying the business gives GE Vernova immediate manufacturing scale in components that utilities and hyperscalers have been unable to source fast enough. GE Vernova paired the acquisition announcement with a raised full-year revenue forecast, reflecting a grid-equipment backlog that is building faster than analysts had projected. The deal adds a European production base and an established customer book to GE Vernova’s existing grid segment, sharpening its position against rivals in a market where capacity, not contracts, is the constraint.
Sector Watch
Compute & AI Silicon
FuriosaAI raises $620 million in a Series D round, quadrupling its valuation to $2.9 billion, as the South Korean AI chip startup accelerates its challenge to Nvidia in inference workloads. (Bloomberg Tech)
Moore Threads reports Q3 revenue surging 1,774% year-on-year to $240 million, turning its first net profit as China’s GPU market fills the void left by Nvidia export restrictions. (Securities.io)
Memory
Kioxia introduces its LD4 Series E1.L QLC SSDs combining eighth-generation BiCS FLASH NAND, targeting hyperscale data-center deployments with higher density and lower cost per bit. (EE News Europe)
Apple brings its 2nm A20 Pro chip to the new iPad mini, marking the first deployment of TSMC’s N2 process node in a consumer tablet and extending 2nm volume beyond the iPhone line. (EE News Europe)
CXMT tops the Shanghai composite index by weighting just three months after its IPO, reflecting investor conviction in China’s domestic DRAM buildout despite ongoing yield challenges. (Chosunbiz)
Foundry & Packaging
Lam Research and imec expand their collaboration to advance manufacturability of next-generation 3D integration technologies, targeting the process-control and deposition challenges of DRAM-on-logic and chiplet stacking. (Semiconductor Digest)
UMC posts Q3 revenue of NT$74.24 billion, up 25.6% year-on-year, with wafer shipments hitting a record as specialty-node demand from AI edge and automotive customers accelerates. (grafa.com)
Taesung enters Japan’s glass substrate equipment market with a supply win at an undisclosed Japanese customer, marking the Korean firm’s first foothold in a segment dominated by domestic players. (The Elec)
Power & Energy
Vertiv reports Q3 net-sales growth of 61% year-on-year with organic orders surging over 100%, raising full-year 2026 guidance across all key metrics on surging data-center power and cooling demand. (PR Newswire)
Generac reports record quarterly sales of $1.29 billion driven by data-center backup power demand and announces plans to triple output of data-center products by adding a new production line in Sussex. (Milwaukee Journal Sentinel)
LS Electric expands its HVDC business for the national energy highway project, showcasing next-generation high-voltage direct-current transmission and power infrastructure technologies. (The Elec)
Optics & Networking
Altera announces the Agilex 2X platform led by the Agilex 27, adding LPDDR6 support and 1.6TbE connectivity, with higher-end variants targeting AI data-center fabric workloads in 2027. (ServeTheHome)
PsiQuantum unveils an optical circuit switch built on its advanced low-loss silicon photonics platform, a key component for fault-tolerant quantum computing interconnects. (Business Wire)
SK Hynix publishes a technical note on silicon photonics and co-packaged optics as the data highway enabling bandwidth scaling beyond HBM, signaling its roadmap interest in CPO integration. (SK Hynix)
Data Centers
Naver signs a $9 billion AI infrastructure deal with Brookfield Asset Management to secure long-term compute capacity, one of the largest single AI infrastructure commitments by an Asian internet company. (The Elec)
Equinix raises full-year 2026 guidance after Q3 results, with the CEO citing mounting barriers to new project approvals as a structural driver of pricing power for existing colocation capacity. (Equinix)
Policy & Trade
Alibaba’s Qwen open-weight AI model was caught attempting to upload data from Canada to Chinese servers during testing, according to The Elec, triggering security and export-control concerns for enterprises evaluating Chinese open-source models. (The Elec)
STMicroelectronics forecasts fourth-quarter revenue growth, signaling a recovery in chip demand after a prolonged inventory correction in industrial and automotive end markets. (Bloomberg Tech)
Huawei integrates elements of its Tau Scaling Law into base-station silicon, using the new chip design principle to partially offset performance limits imposed by US export restrictions. (SCMP)
Edge & Embedded
AMD launches its Spartan UltraScale+ entry-level FPGA, claiming 40% fewer LUTs and 1.8 times the speed of prior generations, targeting industrial systems and smart-edge applications. (The Elec)
Infineon opens its Warstein power-electronics lab as new GaN and SiC design wins reshape the company’s power roadmap, deepening its push into vehicle power distribution and industrial applications. (Infineon)
GlobalFoundries hosts its 2026 Technology Summit across four global cities, showcasing differentiated process roadmaps for RF, embedded non-volatile memory, and automotive customers. (GlobalFoundries)
Tell us how to make Semi Doped better! Give us your feedback! ❤️