Supply-chain intelligence for the AI era. From semiconductors to frontier labs and the infrastructure behind it all.skylabintel.substack.comJoined September 2026
Infineon has opened a Bangkok hub that combines the full semiconductor backend flow with wafer test in one greenfield site.
Module A currently employs about 350 people and is expected to grow to roughly 1,000 as it ramps. The site can expand to five modules, giving Infineon a staged path to add assembly-and-test capacity as demand develops.
Advanced automation and a digital manufacturing backbone support assembly, test and wafer test for automotive, industrial and energy-infrastructure chips. The combination gives Infineon one scalable site for both wafer test and downstream packaging operations, this is in addition to Infineons large presence in Penang, Malaysia.
$IFNNY Infineon
Applied Materials has NT$6.984B ($220M USD) of disclosed machinery purchases from UMC, a foundry expanding into silicon photonics and advanced packaging.
UMC's October 1 disclosure names Applied Materials as the supplier. It covers purchases from December 23, 2025 through October 1, 2026 for production and research.
In July, UMC raised its 2026 capital-spending budget to US$2B. Its expansion plan includes cleanroom installation and tool purchases for silicon photonics in Singapore, plus a new fab building in Tainan to expand its advanced-packaging footprint.
$AMAT Applied Materials | $UMC United Microelectronics
ASE Inc. disclosed NT$1.104B of purchases from Advantest in its October 2 filing. The cumulative total covers June 29 through October 2, excluding tax.
Advantest supplies the systems that test chip performance and quality. In its July results, it said more complex AI and high-performance chips were driving customers to invest in testing, while higher-performance DRAM lifted memory-tester sales. Its test-system revenue rose 38.7% year on year in the quarter ended June.
$6857.T Advantest | $ASX ASE Technology
Winbond's September sales rose 256.7% year on year as the memory shortage strengthened pricing across its DRAM and flash businesses.
Winbond makes specialty DRAM: the working memory used in electronic systems and NOR flash / SLC NAND, which store software and boot code in servers, networking equipment, vehicles and industrial devices.
In its August results, Winbond reported that Q2 DRAM average selling prices roughly doubled from Q1, reflecting higher prices and a better product mix. DRAM revenue rose 78% even as the amount of memory shipped fell by a low-teens percentage. Its memory factories ran at near full capacity.
September consolidated revenue reached NT$28.249B, up 3.4% from August. Q3 totaled NT$82.331B, 37.6% above Q2.
Micron's shift toward higher-priced server DRAM helps explain the opening for Winbond: AI demand draws the major producers toward advanced memory, tightening supply for older products that remain in high demand with increased pricing power.
$2344.TW Winbond | $DRAM Roundhill Memory ETF | $MU Micron
Hon Hai (Foxconn) crossed NT$1 trillion in monthly revenue for the first time as continued server growth explodes.
September sales reached NT$1.159T ($36.5B USD), up 38.4% year on year and 25.7% from August. Q3 totaled NT$3.027T, up 20.4% from Q2.
The company had attributed August's annual growth to increased server-product shipments. Hon Hai expects its AI-related operations to keep growing in Q4. That gives the coming quarter a separate driver after the seasonal September surge.
Hon Hai builds systems around Nvidia's GB200 and GB300 NVL72 platforms.
$2317.TW Hon Hai | $NVDA Nvidia | $DELL Dell
Nan Ya PCB's September revenue growth accelerated to 57.8% year on year, from 40.6% in August. A higher-value product is already in production: substrates for 1.6T switches.
In its August investor presentation, the company said mass production of those substrates lifted Q2 revenue and improved product mix and operating profit. Higher server and switch substrate sales had already driven 37.2% growth in the first half.
September revenue reached NT$5.586B, up 4.4% from August. The latest sales gain follows a documented networking-product ramp with a contribution to profitability.
$8046.TW Nan Ya PCB
The Daily DRAM | October 05, 2026 (as of Asia close)
DDR5 16Gb (4800 / 5600): $58.067 | +0.12%
DDR4 16Gb (3200): $83.588 | +0.00%
DDR4 8Gb (3200): $46.321 | +0.00%
DDR5 reached $58.067 per chip, extending its gain since September 18 to 3.69%.
$MU Micron | $SKHY SK hynix | $DRAM Roundhill Memory ETF
The Daily DRAM | October 02, 2026 (as of Asia close)
DDR5 reached $58 while 16Gb DDR4 slipped.
DDR5 16Gb (4800 / 5600): $58.000 | +0.12%
DDR4 16Gb (3200): $83.588 | -0.09%
DDR4 8Gb (3200): $46.321 | +0.00%
DDR5 finished the week at $58.000 per chip, up 3.57% since September 18. Over the same period, 16Gb DDR4 fell 2.52%.
$MU Micron | $SKHY SK hynix | $DRAM Roundhill Memory ETF
The Daily DRAM | October 01, 2026 (as of Asia close)
DDR5 rose again while DDR4 benchmarks stalled.
DDR5 16Gb (4800 / 5600): $57.933 | +0.06%
DDR4 16Gb (3200): $83.662 | -0.03%
DDR4 8Gb (3200): $46.321 | +0.00%
Since September 18, DDR5 is up 3.45%, while 16Gb DDR4 is down 2.44%. 8Gb DDR4 is up 0.62% over the same period.
$MU Micron | $SKHY SK hynix | $DRAM Roundhill Memory ETF
HBM, CPUs, optics and accelerators dominate the discussion about AI bottlenecks and supply constraints.
But can the silicon wafer industry actually supply the next wave of chip factories?
SUMCO’s president said that, in his view, a 5–10% wafer price increase would not meaningfully change the decision to add capacity.
September coverage of UBS research puts the increase needed to encourage new investment at 40–50% or more.
Expanding chip production requires more qualified wafers. Expanding wafer supply requires prices that justify that new investment, yet leading wafer suppliers remain locked into 2023-era pricing under long-term agreements.
For SUMCO $SUOPY, GlobalWafers and Shin-Etsu $SHECY, repricing could improve earnings at existing plants well before new capacity enters production.
Our latest article maps the customer fab buildout, the wafer supply outlook for 2027–2028, and the economics of financing the next expansion.
The Coming Silicon Wafer Squeeze ↓
The Daily DRAM | September 30, 2026 (as of Asia close)
In anticipation of Micron earnings today, DDR5 and 16Gb DDR4 edged higher after Tuesday's pause.
DDR5 16Gb (4800 / 5600): $57.900 | +0.12%
DDR4 16Gb (3200): $83.686 | +0.03%
DDR4 8Gb (3200): $46.321 | +0.00%
DDR5 is now up 3.39% since September 18. Over the same period, 16Gb DDR4 is down 2.41%, while 8Gb DDR4 is up 0.62%.
$MU Micron | $SKHY SK hynix | $DRAM DRAM ETF
Japan’s flash exports rose another 18% in August
Japan’s packaged flash-memory exports reached ¥217.96B (US$1.38B) in August, up 369% from a year earlier. The export bill kept climbing through the summer despite declining stock prices, with flash price increases doing most of the work.
June: ¥136.13B (~US$0.86B)
July: ¥185.38B (~US$1.18B)
August: ¥217.96B (~US$1.38B)
August export value rose 17.6% from July. The number of exported items rose just 2.6%, while value per item increased 14.6%. Against August last year, item count rose 21.3% and value per item rose 286%.
Japan is the manufacturing base behind Kioxia and Sandisk’s NAND partnership, with fabs in Yokkaichi and Kitakami.
For Micron’s earnings tomorrow, the useful read-through is the continued strength in flash-chip value and pricing. Last quarter, Micron’s NAND revenue rose ~99% from the previous quarter: selling prices increased in the mid-80% range, while memory capacity shipped increased by a mid-single-digit percentage.
Finished SSD and other flash-storage exports were softer: ¥6.47B (~US$41M), up 38% year over year but down 45% from July. August’s strength was concentrated in chips rather than a broad surge in finished storage exports.
$MU Micron | $SNDK Sandisk | $285A.T Kioxia
The US took half of Malaysia’s flash-storage exports in August
Malaysia exported RM16.72B, about US$4.11B, of SSDs and other flash-storage devices in August. Export value was up 468.8% year over year. The US took 50.7% of the total.
The largest destinations were:
• US: RM8.48B.
• China: RM1.97B.
• Mexico: RM1.70B.
Both Sandisk and Micron have large advanced packaging and testing facilities in Batu Kawan, across the bridge from Penang island.
$MU Micron | $SNDK Sandisk
The Daily DRAM | September 29, 2026 (as of Asia close)
DDR5 paused after six gains while 8Gb DDR4 added 0.23%.
DDR5 16Gb (4800 / 5600): $57.833 | +0.00%
DDR4 16Gb (3200): $83.660 | +0.00%
DDR4 8Gb (3200): $46.321 | +0.23%
DDR5 held at $57.833 after six consecutive gains, and 16Gb DDR4 stopped falling. Since September 18, DDR5 is up 3.27%, while 16Gb DDR4 is down 2.44%.
$MU Micron | $SKHY SK hynix | $2408.TW Nanya
Vertiv agreed on September 24 to buy King Environmental Services, an Irish company that prepares, cleans and tests liquid-cooling systems. The deal would expand Vertiv’s fluid-management services across Europe, the Middle East and Africa.
Cooling pipes need more than pumps. Trapped air, corrosion, contamination and the wrong fluid chemistry can reduce how effectively a system removes heat.
King’s work includes flushing pipes, filtering and treating the fluid, balancing flow, and testing cooling systems under load before a facility starts operating. It also provides ongoing maintenance.
Vertiv expects the acquisition to close in the final quarter of 2026.
$VRT Vertiv
Marvell is securing more US chip capacity for AI optical connections
GlobalFoundries and Marvell expanded a multiyear manufacturing agreement on September 17. It will add capacity at GlobalFoundries’ Burlington, Vermont factory for chips used in high-speed optical connections.
These chips use silicon germanium, a material suited to handling fast signals. GlobalFoundries says its current technology supports optical connections carrying 200 gigabits per second on each lane.
The agreement builds on existing production. It covers chips for plug-in optical modules and designs that place optical connections closer to computing and networking chips.
$MRVL Marvell | $GFS GlobalFoundries
Credo’s new 1.6T optical links can flag network trouble remotely
Credo launched a new family of optical connections for AI networks on September 15. They carry 1.6 terabits of data per second and continuously report the health of the connection.
That speed is twice the capacity of an 800-gigabit link. The monitoring is meant to help operators spot unstable connections before they interrupt an AI workload.
The products combine Credo’s signal-processing chip, optical chip and diagnostic software. Operators can retrieve information from both ends of a connection, update software remotely and inspect a saved history of problems.
$CRDO Credo
Oracle says it has delivered more than 300,000 GPUs since the end of May
Its September 10th update reported more than 300,000 GPUs delivered to AI cloud customers since the end of May, alongside 850 megawatts of additional data-center capacity in its latest quarter.
Cloud infrastructure revenue reached US$7.4B in the quarter ended August 31, up 121% from a year earlier.
Oracle generated US$23.1B of operating cash flow and spent US$28.5B on capital projects during the quarter, leaving negative US$5.4B of free cash flow.
$ORCL Oracle
5 Followers 42 FollowingInstitutional intelligence for private AI investing.
From compute and energy to software, robotics, scientific discovery and longevity.
46 Followers 462 FollowingAI semis read from the supply chain up: DRAM, HBM, NAND, hyperscaler capex. Read-throughs, catalysts, and trades with thesis and invalidation attached
36 Followers 541 FollowingThere is surely nothing other than the single purpose of the present moment. A man's whole life is a succession of moment after moment.
783 Followers 521 FollowingYou can’t search for what you haven’t thought of yet. https://t.co/z7FnZdSnYc tracks 40,000+ stock niches to find what’s moving.
🔍 See what it's finding now
10K Followers 4 FollowingThe industry powering AI, explained by @vikramskr and @austinsemis. Weekly podcast and daily takes. Sign up at https://t.co/wSYNiNJYqL and subscribe on YT
15K Followers 5K FollowingSEMI connects the global electronics supply chain to promote its development and foster the collaboration and innovation vital to the growth of SEMI members.
107K Followers 10K Following"Key Context" Substack covering AI and chips. Reached #1 new bestseller first 24 hrs
Subscribe https://t.co/3N2fHOXQfL
"Be so good they can't ignore you"
35K Followers 2K FollowingEnergy Abundance, AI Power & Data Centers, NBA, & Rap posts. Background in equity & credit markets, startups, & IB. Personal views NOT employers. NOT advice.
13K Followers 254 FollowingDeep-Tech Semiconductor Analyst, MSEE @JohnsHopkins. Expert-driven system architecture deep-dives of AI datacenter hardware at https://t.co/atebRcUn6C
1K Followers 1K FollowingTravis Keam | Building Antfeed ⚡️ | Real-time Korea semis news for global investors tracking $SKHY + Samsung — straight from local sources | English/Korean app
51K Followers 2K FollowingCEO & Principal Analyst @creativestrat. Full-stack tech analysis from semis to software. Advisor to industry leaders. Technologist at heart. Keeper of Bees.
11K Followers 524 FollowingSold my finance co in 2021, now trade/invest full time. Specialty is finding semi/AI infra inflection points before the breakout. Lots of receipts. Zero AI slop
11K Followers 486 FollowingSemicon expert. Ex AMD, Ex Samsung. Now at Renesas. PhD Paris Tech. MBA from INSEAD. Opinions my own.
https://t.co/vrQ6HkBsVC
1.1M Followers 220 FollowingOnly on X, don’t trust fake accs
AI/Semi Supply Chains Research
Nothing is investment advice. No paid promos; may trade/hold names disc, views my own.
55K Followers 1K FollowingFormer journalist, now financial analyst. Based in Taipei. Tweet mainly about semiconductors and Taiwan. Not investment advice. Views are my own.
1.9M Followers 179 FollowingNobel Laureate. Co-Founder & Chair @GoogleDeepMind; Chief Scientist of Alphabet. Founder & CEO @IsomorphicLabs. Building the future...
1.9M Followers 1K FollowingCo-Founder of Coursera; Stanford CS adjunct faculty. Former head of Baidu AI Group/Google Brain. #ai #machinelearning, #deeplearning #MOOCs