Hello world, it’s Thursday, October 8th.
Samsung Electronics posted a record 107.4 trillion won in Q3 2026 operating profit, a 783% year-on-year surge that Reuters called the highest quarterly profit ever reported by any tech company at roughly $80 billion. Elsewhere, Wolfspeed landed a conditional $1.5 billion U.S. government loan to build out a domestic SiC supply chain, and AMD CEO Lisa Su was in Seoul on October 7 pressing Samsung and SK Hynix for supply deals spanning three to five years.
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 AI infrastructure from the ground up — power, data centers, and the cloud on top.
If you’re running open models — GLM, DeepSeek, Kimi, Qwen — Crusoe’s managed inference is fast: up to 9.9x faster time-to-first-token and 5x the throughput of vLLM, thanks to their MemoryAlloy caching layer and tokenizer fastokens. And it’s the easy kind of fast: pick a model, grab an API key and start inference. No GPUs to provision, no infrastructure or clusters to manage, and it’s pay per million tokens.
$5 in free credits to try it. Send your first request today →
Samsung and TSMC Print All-Time Records as AI Spending Lands in Earnings
Samsung Electronics reported a 783% year-on-year surge in Q3 2026 operating profit to a record 107.4 trillion won, with Reuters noting the result represents the highest quarterly profit ever posted by any technology company at roughly $80 billion. The memory windfall drove almost all of the gain, as hyperscaler AI infrastructure buildouts translated directly into HBM and DRAM demand. TSMC’s quarterly revenue jumped 51% year-on-year, with Bloomberg reporting that AI demand held firm through the period. Both companies reported these results within days of each other, yet their stocks barely moved, reflecting, per Bloomberg, investor expectations already calibrated to torrid AI-driven growth.
Austin: More operating profit than Nvidia!? Nvidia did $63.7B in its July quarter. And that’s with Samsung’s phones, TVs, and foundry reportedly losing money, so memory is carrying all of it.
Vik: $80B a quarter is a stupidly large number. This memory craze has to end at some point, right? This is literally the trillion dollar question.
Zeiss and ASML Map Hyper-NA EUV Path as Sell-Side Upgrades 2027 Outlook
Carl Zeiss and ASML have publicly outlined a Hyper-NA EUV roadmap extending beyond the High-NA tools currently entering production, confirming that chipmakers are already planning for the generation after next. The technical work centers on pushing numerical aperture past the 0.55 ceiling of High-NA systems, a step that requires new optical architectures Zeiss is developing alongside ASML. Wall Street is paying attention. JP Morgan says ASML could signal a stronger-than-expected 2027 revenue outlook at its next investor event, while Morgan Stanley has separately raised its earnings estimates on the basis of ASML’s pricing power and accelerating EUV demand. Customers apparently aren’t waiting to find out.
Vik: Everyone’s planning the node after next, but I called hyper-NA just pointlessly expensive, EUV pushed past its economic limits. The 2027 money is booked anyway - JP Morgan says most EUV orders for the year are already committed. Hyper-NA revenue is a long way off.
Wolfspeed Wins $1.5B Defense Loan to Build Domestic SiC-GaN Supply Chain
The U.S. Department of War’s Office of Strategic Capital has issued a conditional $1.5 billion loan commitment to Wolfspeed (NYSE: WOLF) to finance domestic production of silicon carbide materials and wide-bandgap power devices, with an explicit national-security mandate covering applications in drones, radar, and missile defense. The commitment advances key Department of War initiatives alongside the domestic wide bandgap supply chain. The commitment is conditional, meaning Wolfspeed must satisfy further requirements before funds are disbursed. Demand from AI data centers is running alongside the defense rationale: higher-voltage power networks in large-scale compute facilities have accelerated adoption of SiC and GaN devices, giving Wolfspeed’s expansion a dual commercial and military customer base.
Vik: Wolfspeed overbuilt SiC for EVs and got caught when Chinese capacity took demand down, so a conditional 30 year defense loan matters a lot here. What interests me is the GaN-on-SiC RF epi for electronic warfare, not the data center angle everyone’s tacking on. SiC demand is legit though.
AMD CEO Lisa Su Presses Samsung, SK Hynix for Multi-Year Supply Deals
Lisa Su traveled to Seoul on October 7 to open partnership talks with Samsung Electronics and SK Hynix, telling Korean media at Seoul Dragon City in Yongsan that supply chain complexity requires cooperation plans spanning three to five years. The meetings covered HBM memory and advanced packaging allocations that AMD needs ahead of its next product ramps. Su also met Ryu Je-myung, South Korea’s second vice minister of the Ministry of Science, and announced a dedicated AI research center to deepen AMD’s presence in the country. AMD separately said it plans to integrate Korean NPUs into its own hardware, broadening the scope of the Seoul talks well beyond memory supply. The research center commitment and NPU integration pledge together give Korean counterparts reasons to prioritize AMD alongside existing Nvidia obligations.
Vik: This is AMD trying not to end up last in line for HBM behind Nvidia. Memory is over half a rack’s cost now, so a 3 to 5 year lock makes sense. The AI research center is the sweetener you dangle to get bumped up the allocation queue.
Austin: First time AMD has disclosed SK Hynix memory in its AI systems. HBM4 for MI455 was known to be Samsung only, so a second source is the real news here with supply this tight.
Turba Labs Raises $52 Million to Double Compute by Optimizing Existing Data Centers
Palo Alto-based Turba Labs has raised nearly $52 million in combined seed and Series A financing to optimize AI infrastructure without building new data centers. The two-year-old startup creates a digital twin of data center infrastructure, models customers’ AI use cases, and routes workloads to hardware best suited for their complexity. Co-founder and co-CEO Hans-Juergen Schmidtke said the platform could double available compute by customizing how AI workloads interact with underlying hardware, though the exact capacity unlocked will vary by infrastructure.
Sources: wsj.com
Vik: Huh. Interesting idea. Run and optimize a datacenter buildout using computer simulation before deploying billions. There is a lot of economic value to that.
Austin: Double is a big claim. What’s the baseline, i.e. how much compute is sitting idle in these data centers today?
Applied Digital Posts Q1 2027 Results as HPC Lease Demand Accelerates
Applied Digital reported fiscal first-quarter 2027 earnings as demand for high-performance computing infrastructure continued building across its data-center portfolio, with utilization rates and lease pricing drawing close attention from analysts tracking the broader HPC operator cycle. The company, regarded as one of the first AI-native data-center operators to report this cycle, provides an early read on how AI cloud infrastructure is absorbing capacity. Pentagon interest in AI cloud services has added a new demand vector for operators like Applied Digital, expanding the customer base beyond traditional hyperscalers. As subsequent HPC operators prepare to report, Applied Digital’s figures on contract pricing and facility utilization now serve as the opening data point for the quarter.
Vik: Applied Digital’s whole pitch is durable long term contracts with tier-one investment grade hyperscalers. Do not underestimate military interest in AI compute, especially after the OAI-HuggingFace incident has shown what AI agents can do in unison. Rest assured governments will militarize that.
Sector Watch
Foundry & EDA
Siemens EDA and TSMC are jointly applying agentic AI to semiconductor design workflows, targeting faster RTL-to-GDSII iteration on TSMC process nodes. (The Elec)
Rapidus enlists four Malaysian IC design houses as it actively builds a 2nm customer pipeline ahead of commercial production. (The Business Times)
Rohm expands back-end chip manufacturing outsourcing, signaling a structural shift in how the Japanese IDM manages packaging capacity.
Memory
SK Hynix’s NAND flash memory unit is said to have selected banks for a 2027 U.S. IPO, putting a valuation process in motion for one of the world’s largest flash suppliers. (Bloomberg.com)
Biren shares slump 6 percent following a $4 billion share sale. (The Standard (HK))
BE Semiconductor shares slide as analysts flag memory-shortage pressure on its hybrid-bonding ramp, per Bloomberg coverage. (Bloomberg Tech)
Packaging & Semiconductor Equipment
Aehr Test Systems receives $6M follow-on Sonoma burn-in production orders from its lead hyperscale customer for a next-generation AI processor, extending its burn-in footprint. (Aehr Test Systems)
Mitutoyo opens a dedicated semiconductor metrology center in Dresden, expanding precision measurement support for European chip fabs. (The Elec)
Boss Semiconductor and Japan’s Mirise Technologies partner to develop chiplet interconnect technology for automotive applications. (The Elec)
Power
Renesas adds 100V GaN FETs targeting AI data-center and industrial power conversion, expanding its wide-bandgap device lineup.
LS Electric establishes a Taiwan joint venture to expand its power-infrastructure business on the island. (The Elec)
HD Hyundai Infracore wins a 390 billion won US data-center gas generator engine order, its largest disclosed stateside energy contract. (The Elec)
Data Centers & Cloud Infrastructure
CoreWeave enters India with a 240 MW AI cloud region at AdaniConneX campus, its first international deployment outside North America and Europe. (CoreWeave)
San Francisco passes a data-center moratorium, restricting new builds in the city that is home to many of the leading AI companies. (Bloomberg.com)
Firmus and CDC abandon their planned $51 billion Australian AI data-center project as Firmus shifts focus to Asian expansion. (Light Reading)
Compute & Edge
Microsoft launches the $2,599 Surface Laptop Ultra with an Nvidia RTX Spark chip, its first co-engineered AI PC with Nvidia. (CNBC)
AMD will launch a next-generation APU with FSR 4 support by year-end, expanding its consumer AI compute lineup. (The Elec)
Dell Technologies unveils new Windows PCs and workstations targeting the agentic AI era at a product launch event. (Business Wire)
Policy & Trade
Amazon vows to stop signing data-center NDAs with local governments, following Microsoft’s similar pledge amid scrutiny of secret deals. (Law360)
FCC approves SpaceX to deploy a 15,000-satellite D2D LEO constellation, with a nine-year completion deadline through October 2035. (Light Reading)
SEMI Europe joins the European Commission’s STEAM Executive Panel to address the semiconductor industry’s growing skills shortage. (EE News Europe)
Tell us how to make Semi Doped better! Give us your feedback! ❤️