Chips and Wafers’ Lover
The US will unveil Section 301 semiconductor tariffs around the time China Chairman Xi visits the US for a summit with President Trump, media report, adding Taiwan’s tariff rate will likely be close to the 15% already negotiated with the US due to major investment commitments by Taiwan’s chip and AI stack companies. (Note: We count over US$300 billion in solid commitments). $Taiwan Semiconductor(TSM.US) #GlobalWafers #Wistron #Foxconn #Quanta #semiconductors
Source: Dan Nystedt
Google is a key co-investor alongside Nvidia in the US$3.9 billion convertible bond deal floated by MediaTek, Taiwan media highlight, due to the significance MediaTek’s work on Google TPUs, and both partnerships back a 10-year product time horizon, rather than meeting short term funding needs. #MediaTek #semiconductors
Source: Dan Nystedt
TSMC may raise prices across all process nodes by another 10% to 15% amid strong demand for AI-related chips - and Samsung’s ability to hike prices 10%-15% for its 4nm and 5nm offerings, media report, citing market expectations (NOT from TSMC). The market also reportedly sees follow on price hikes for N2, N3 and N5 advanced processes of 10% by end-2027. $Taiwan Semiconductor(TSM.US) $SSNLF #Samsung
Source: Dan Nystedt
TSMC: Demand for AI compute will 5x annually in coming years, leading overall system compute in 2029 to 50x that of 2024, media report, citing April Li, head of AI and HPC Business Development. TSMC estimates 30 to 40GW per year of AI data centers will be built per year, compared to 5 to 6GW of data centers historically. New advances are vital to keep pace, and TSMC aims to have A12 in mass production in 2029, and 14x-reticle size CoWoS packaging in 2028, and more. By 2030, a single AI chip package will hold over 1 trillion transistors, she said. $Taiwan Semiconductor(TSM.US) $NVIDIA(NVDA.US) $Alphabet(GOOGL.US) $Broadcom(AVGO.US) #semiconductors
Source: Dan Nystedt
Strong demand for Google Tensor Processing Units (TPUs), Intel outsourcing more as it focuses on EMIB, and price increases due to rising costs and demand are all converging into a boom of advanced packaging and testing orders for ASE Technology, media report, adding Google has aggressively expanded TPU volume via partners including Broadcom, MediaTek, Marvell and AMD, leading to substantial add-on orders at ASE. Estimates put Google TPU shipments at 12 to 15 million units by 2028. TSMC remains Google’s chip fabrication and CoWoS partner, with ASE also a major winner from the TPU ramp up. $Advanced Semiconductor Engineering(ASX.US) $Alphabet(GOOGL.US) $Intel(INTC.US) #semiconductors
Source: Dan Nystedt
Google is set to highlight this year’s SEMICON Taiwan (9/2-4), which is already breaking records in scale, media report. The 3-day semiconductor equipment and materials exhibition opens Wednesday, though some forums start today (Monday).
Google’s Chief Technologist for AI Infrastructure Amin Vahdat will deliver a keynote from the eyes of a custom chip developer at the event, making CSPs a new centerpiece. Over 100,000 semiconductor professionals are expected this year, with the exhibition size hitting 4,300 booths, by 1,300 exhibitors. A total of 18 nations have set up in the international pavilion, up 250% over last year, while the Startups area has doubled in size. $Alphabet(GOOGL.US) #SEMI #semiconductorsSource: Dan Nystedt
Nvidia jump starts AI PC market: Asustek’s and MSI’s initial batches of AI PCs designed around Nvidia RTX Spark processors have completely sold out in pre-order, media report, and both have already asked Nvidia for more chips, which are made on TSMC’s 3nm process. There are 2 main RTX Spark chips, the N1 for mainstream AI PCs and the N1x for high-end. $NVIDIA(NVDA.US) $Taiwan Semiconductor(TSM.US) #Asustek #MSI #AIPC
Source: Dan Nystedt
Singapore has increased their Substrate imports as they ramp packaging capacity.
But want's most interesting is where all those substrates are coming from - Malaysia.$ATS(ATS.US).VIE $222800.KQSource: Chips & Wafers
Taiwan prosecutors raided a plant owned by ABF substrate giant Unimicron on Friday over the alleged import of substrates and printed circuit boards (PCBs) made at its China plants and then re-labelling them ‘Made in Taiwan’, media report. Some employees are suspected of crimes including document forgery and product origin fraud. The head of Unimicron’s PCB division was released Saturday on bail of NT$15 million (US$475,000). Several other managers were also released for lower amounts. $NVIDIA(NVDA.US) $Alphabet(GOOGL.US) $Amazon(AMZN.US) $Microsoft(MSFT.US) $Meta Platforms(META.US)
Source: Dan Nystedt
TSMC’s 2nd quarter employee profit-sharing payout hit NT$36 billion (US$1.14B), up over-50% from the same time last year, media report, and raising the total 1st half-2026 bonus to NT$70.35B, up 54.3%. TSMC’s profit sharing bonuses are benefitting from the strong AI trend, as the chip giant’s net profit in the 2nd quarter rose 77.4% to NT$706.56B. In 2025, the bonus rose 30% versus 2024. $Taiwan Semiconductor(TSM.US) #semiconductors
Source: Dan Nystedt
Marvell QJul26
- F27 (C26) revenue raised by $500M (+45% y/y) while F28 (C27) revenue raised by $1.5B (+50% y/y); incremental largely driven optical connectivity though custom is also a part- F28 guidance already includes the Google warrant revenue; F29 and beyond will see the step-up- F27 DC growth raised to 60% y/y (vs 50% prior); F28 DC raised to >60% y/y (vs 55% prior); F28 supported by additional supply secured- F28 custom still expected to >2x y/y (not capped); F29 custom >$10B target reiterated with upside bias- Revenue up 13% q/q and 37% y/y to $2.739B at 58.9% non-GAAP GM (flat q/q); both above midpoint- QOct guided up 15% q/q to $3.15B; 4Q accelerates furtherSource: Sravan Kundojjala
Support for a strike at Micron Taiwan over bonus pay hit 80% in a preliminary survey, as the US memory giant faces trouble similar to Samsung in May, when a last-minute deal averted a strike, media report. Taiwan labor unions are pushing to replace the current bonus system with a profit-sharing scheme amid the AI-driven memory chip boom.
Unions say peers offer far higher profit-based bonuses, with Samsung paying out 10.5% of profits and SK Hynix 10%. Formal mediation talks are expected to begin by mid-September at the latest, a mandatory requirement before an official strike vote can be held under Taiwan law. (Note: Taiwan accounts for roughly 60% of Micron’s global production capacity, and anchors the bulk of advanced HBM output) $Micron Tech(MU.US) $SK Hynix(SKHY.US) $SSNLF #Samsung #semiconductorsSource: Dan Nystedt
India rolled out ‘Semicon 2.0’, the 2nd tranche of government funds aimed at building out its semiconductor industry, ₹1.275 trillion rupees (US$13.4 billion) aimed at chip design and manufacturing, media report. The figure is far higher than India’s first tranche of chip investment, ₹760 billion rupees, launched in December 2021. The first phase aimed at silicon chip fabs, compound semiconductor fabs, packaging and testing plants and chip design. That program saw 12 manufacturing deals and investments of over ₹1.64 trillion rupees. #India #Tata #PSMC
Source: Dan Nystedt
SK Hynix’s CEO said the global memory chip shortage is expected to last until the end of 2030 and that the risk of oversupply is low since memory chips are no longer “just commodities” in the AI era, media report. His comments came at a ground breaking ceremony for a US$4 billion advanced chip packaging plant in Indiana, USA, which aims to begin volume production of next-generation HBM (high bandwidth memory) chips in 2029. $SK Hynix(SKHY.US) $SSNLF $Micron Tech(MU.US) #semiconductors
Source: Dan Nystedt
Samsung Electronics aims to reach mass production on FOPLP (Fan-out Panel Level Packaging) by 2028, media report, adding its decades of work in display panels mean it’s taking a very different path than TSMC. Both are moving to rectangular shaped substrates – Samsung on 415 x 510 mm panel format, TSMC on 310 x 310 mm – but Samsung’s past work focused on mature chips. It’s expanding the effort to high-performance chips by 2028. TSMC, which calls its offering CoPoS (Chip-on-Panel-on-Substrate), is designed for high-end graphics chips (GPUs), custom ASICs, and other HPC (high-performance computing) chips. $Taiwan Semiconductor(TSM.US) $SSNLF #Samsung #FOPLP #semiconductors
Source: Dan Nystedt
Synopsys QJul26
- EDA and IP are back on track; IP is still 10% down vs previous peak- EDA at $1.28B is an all-time high, 8% above the prior $1.18B peak of Q3 FY25- On AI disruption, the company sees customer models still needing its physics and sign-off tools; AI-native design is a demand driver rather than a substitution risk, per SNPS- Record hardware revenue again; 12 new and 66 repeat hardware customers in the quarter- IP revenue $474M, up 11% y/y and 4% q/q; first y/y growth after three straight down quarters, second consecutive sequential increase, with further sequential growth expected in Q4Source: Sravan Kundojjala
ABF substrate supply is tight, and key suppliers are looking to expand capacity globally.
One interesting read through is MEC ($4971.TYO), a Japanese specialty chemicals company with an effective monopoly on the CZ series adhesion promotion chemistry used in ABF substrate manufacturing.The chemistry is critical. It enables copper circuitry to bond to the build up film and is applied at every layer of the substrate.MEC’s latest disclosures show sales to China growing far faster than to any other region.This looks like an important signal that Chinese substrate makers are investing aggressively to build the technical expertise and capacity needed to compete in ABF.Some of the increase may come from non Chinese suppliers manufacturing in China, such as AT&S and Zhen Ding.But it is hard to see that alone explaining the magnitude of the increase.The takeaway: China’s push into ABF substrates looks increasingly real, and MEC’s chemistry sales may be one of the better ways to track it.Source: Chips & Wafers
Intel at DB conference
- Upbeat on foundry and process node developments; 18A yields tracking ahead of targets; calls14A defect density curve the best since 22nm- Strong internal demand for 14A; increased confidence on 14A external engagements; 0.9 PDK in Oct- EMIB-T ramp from back half of ’27; multiple billions of dollars of EMIB per customer; lower capex and higher ROIC vs wafer business- Sees 40% GM and 30% OPM for advanced packaging; wafer business will also have similar margins and depends on 14A execution as the best in this business is now close to 70%Source: Sravan Kundojjala
Qualcomm sees HBC (High Bandwidth Compute) coming to smartphones, XR and cars too in the future. HBC Gen 1 commercialization in 2027 and Gen 2 in 2028. Sees lots of interest from hyperscalers. Adopts system integratrator model with compute die from Qualcomm, TSMC manufacturing, DRAM from stacking from someone else. With HBC, Qualcomm claims 6x the bandwidth per watt of HBM designs.
Source: Sravan Kundojjala
Nvidia Key Comments From 2nd quarter earnings call
Supply ChainNvidia CFO Colette Kress: “Looking ahead, our preliminary expectation is for fiscal year 2028 revenue to grow approximately 70% year-over-year. Although we will work to close the supply-demand gap, we expect supply to remain a bottleneck at least through the end of fiscal year 28.”(Analysts expected 44% for fiscal 2028. Nvidia’s Fiscal 2028 starts Feb. 1, 2027, and ends Jan. 31, 2028)Nvidia CEO Jensen Huang: “Incredibly, we are seeing demand acceleration even at our scale. Customers’ forecasts point to our growth doubling next year. However, as mentioned earlier, we expect to grow approximately 70% as we are supply constrained.”CFO: “Memory scarcity today is being driven in large part by the AI buildout itself and unlike a component that simply raises our cost with no offset benefit. Tighter memory supply is a symptom of the same demand surge that’s driving our own growth.” Nvidia Release: “We’ve partnered with our extensive network of suppliers to secure the critical components needed to meet demand for the next several years. Our commitments increased from $119 billion last quarter to $279 billion, primarily related to the procurement of memory.”CFO: “Many of you have expressed concerns regarding our gross margins as component costs have risen significantly. As you are already aware, we are experiencing extreme pricing conditions in memory. The magnitude of the price increase has exceeded our prior expectations and are headed even higher into next year.”CFO: “We have now absorbed the cost increases. We have also repriced our products in the marketplace.”Nvidia guided 3rd quarter GAAP and non-GAAP gross margin guide at 74% +/- 50 basis points, saying margins will bottom in Q4 in the 71-72% range. In fiscal 2028, margins will be around 72% to 73%Vera RubinCFO: “We commenced production shipments of Vera Rubin earlier this month, having already received purchase orders from every major hyperscaler, AI cloud and system OEM. We expect Vera Rubin to mark the fastest product ramp in Nvidia’s history.”“We see Vera Rubin accounting for about 20% of data center revenue in Q3.”(Back-of-envelope: Nvidia guided total Q3 revenue at about $108 billion +/- 2%. So maybe Q3 data center revenue will be $100.4 Billion – and Vera Rubin revenue will hit $20.09 billion for Nvidia.)ChinaCFO: “In Q2, we shipped less than 1% of our total data center revenue in Hopper 200 products to customers based in China in accordance with the US government licenses. And given ongoing geopolitical uncertainty, there is no China data center compute revenue in our forward outlook.” Visibility into fiscal year 2028Jensen: “We just have a lot greater visibility now both upstream and downstream, we’ve just never forecasted a year out…we will continue to work with our supply chain on that…”Source: Dan Nystedt
Intel EMIB advanced packaging has become a confirmed rival to TSMC with strong yields and satisfied customers, media report, citing unnamed supply chain sources, who say Intel’s work with clients Google and MediaTek has been so smooth it indicates that costs and production yields are meeting targets. And with TSMC CoWoS capacity still tight, clients are likely to deepen work with Intel. $Intel(INTC.US) $Taiwan Semiconductor(TSM.US) $Alphabet(GOOGL.US) #mediatek #semiconductors
Source: Dan Nystedt
Is DRAM worth its weight in Gold? Strong AI demand has driven memory chip prices up so fast that 1-kilogram of South Korea DRAM now costs as much as 620 grams of gold – up from 120 grams at the 2023 cyclical trough, according to South Korean trade data cited by media. That’s a 12.5x price surge in under 4-years. $Micron Tech(MU.US) $SSNLF $SK Hynix(SKHY.US) #semiconductors
Source: Dan Nystedt
Apple’s new M6 laptop/compute chip has continued the use of SoIC-MH advanced packaging, but high-end M-series chips (M6 Pro, Max, Ultra) may use 2 advanced packaging techniques, SoIC-MH for interconnect density and WMCM to integrate logic, LPDDR memory and high-speed I/O, media report, noting TSMC is readying capacity in 3 plants: Zhunan AP6, Longtan AP3 and Chiayi AP7. Analysts estimate WMCM capacity could reach 60,000 wafers-per-month (wpm) by end 2026 and over 120,000wpm in 2027. Important to note that the M6 is a monolithic die – everything on one slab of silicon – while the Pro, Max and Ultra versions are multi-die (chiplets) which need to be ‘stitched’ together via advanced packaging techniques.
SoIC-MH: System-on-Integrated-Chips Multi-HeightWMCM: Wafer-Level Multi-Chip Module $Apple(AAPL.US) $Taiwan Semiconductor(TSM.US) #semiconductorsSource: Dan Nystedt
Nvidia’s Groq 3 LPX-based server racks have entered mass production at Foxconn, the primary supplier of the racks, media report, adding Nvidia announced the Groq 3 LPU chip in March. Samsung Foundry is producing the 4nm chips, which feature 500MB of on-chip SRAM per chip, better for inference (decode) and sidestepping the memory chip bottleneck. The Groq 3 LPX rack combines 256 LPUs for 128GB of total on-chip SRAM. 1/2 $NVIDIA(NVDA.US) $SSNLF #Foxconn #Samsung
Source: Dan Nystedt
Nvidia Vera Rubin server rack data throughput is 30-times better than prior generation GB300 NVL 72, with cost-per-token dramatically lowered, giving Taiwan AI server manufacturers visibility to 2028, media report, citing unnamed industry sources. Nvidia announced performance data at the Hot Chips 2026 conference that just wrapped up. Taiwan’s Foxconn, Quanta and Wistron all expect to be in mass production of Vera Rubin servers in the 4th quarter. $NVIDIA(NVDA.US)
Source: Dan Nystedt
