Daily Update - August 6th, 2026
Anthropic builds custom silicon team, Sandisk hits $8.97B revenue up 51%, Foxconn July sales reach $29.34B on AI servers, Celestica raises $3B for data center infrastructure.
Hello world, it’s Thursday, August 6.
Anthropic is building an in-house chip design team to develop custom silicon for its Claude models, joining a growing list of AI labs looking to reduce their dependence on Nvidia. Sandisk posted a record quarter at $8.97 billion in revenue, up 51% sequentially, and Foxconn cleared US$29.34 billion in July revenue, a 54.2% year-over-year gain driven almost entirely by AI servers.
Let’s get into it. — Austin & Vik
Be sure to check out the Semi Doped podcast on YouTube or your favorite podcast player!
Foxconn, Wistron Post Record Results as AI Server Demand Outpaces Supply
Foxconn’s July revenue cleared US$29.34 billion, a 54.2% year-over-year gain and the company’s highest monthly total on record, driven almost entirely by AI server orders rather than the iPhone cycle, which has yet to enter mass production for the 18 Pro. The result arrived ahead of schedule in every sense: analysts had expected AI infrastructure to be a second-half story, and the numbers arrived in the first week of August wearing a record badge.
Wistron’s second-quarter net profit jumped 128% on AI server demand that the company says is outpacing available supply, a characterization that helps explain why both ODMs are committing fresh capital. Wistron is expanding AI-related CapEx across facilities in Taiwan and the United States, adding assembly capacity on two continents in parallel. CNBC cited analyst commentary noting that Hon Hai’s strong 2026 performance rests on AI demand and the iPhone upgrade cycle, with both tailwinds running concurrently through the back half of the year.
Behind the demand:
Super Micro reported a $60 billion quarter, a figure that reframes the scale of AI server procurement flowing through the broader supply chain.
Foxconn and Flex have both been building out Mexico-based assembly capacity as a secondary hub for US-destined AI server production.
Austin: ODM assembly is a bottleneck too, just a cheaper one to fix than front-end fab capacity. A leading-edge fab is years and tens of billions. An assembly line is months and a small fraction of that, which is why Wistron can add capacity in Taiwan and the US at once and why Foxconn and Flex are both building Mexico hubs for US-bound servers.
Adding capacity is still scary. You only get paid back if the lines run hot for years, so adding one in 2026 means believing AI server demand is still there in 2029. Guess wrong and you’re eating fixed cost through the downturn. That’s why supply keeps trailing demand even after everyone agrees to build.
imo, this is great for our future. Nothing gets built faster than we can manufacture it.
Arista Earnings and Celestica’s $3B Raise Feed AI Infrastructure Spending Wave; Simmtech posts 11x operating profit
Celestica priced a $3 billion equity offering on August 5, 2026, with proceeds earmarked to accelerate growth across global AI infrastructure, and Bloomberg placed the deal squarely inside a broader data center funding rush reshaping capital markets. The raise is one of the larger single equity transactions in contract electronics this cycle.
Arista Networks reported its second-quarter 2026 financial results in tandem, adding another data point to the case that networking hardware is capturing durable AI spending well past the initial GPU procurement wave.
Simmtech posted a more than elevenfold surge in second-quarter operating profit driven by demand for AI package substrates. Those moves sit a layer below the hyperscaler build: substrate and PCB suppliers are running at a pace that echoes, rather than trails, the system-level growth at companies like Celestica and Arista. The revenue expansion now runs from package substrates through switching silicon and into full rack integration, with multiple tiers of the supply chain reporting accelerating numbers at roughly the same moment.
Austin: I’d never heard of Simmtech before. I’ve heard plenty about substrates, because they’re increasingly showing up as a pain point on earnings calls. During Intel’s Q2 earnings call, Lip-Bu Tan said the industry is facing “one of the most severe supply constraints in its history across leading-edge logic silicon wafers, memory and [substrates],” and that the shortages “will persist for the foreseeable future.” Dave Zinsner said Intel is “actively securing supply of substrates and memory,” and that doing so means “putting money up in advance of getting the substrates.”
Customers prepaying to reserve your output is how a substrate supplier prints 11x operating profit.
So who should investors watch here? Ibiden, Shinko, AT&S, and of course Simmtech.
Samsung and SK Hynix Test Chinese Equipment as US Controls Drive Hedging
Samsung and SK Hynix are evaluating etching tools from Chinese supplier AMEC for use in their China-based fabs, according to a Reuters exclusive, as both chipmakers seek insurance against further tightening of US export restrictions. The assessment covers whether AMEC equipment can serve as a viable substitute for Western tools that could be cut off under future US policy moves. It’s a telling development: American controls intended to limit Chinese semiconductor capability are instead pushing two of the world’s largest memory producers to actively qualify Chinese gear in their own production lines.
The evaluations represent a direct hedging response to US export-control escalation, and if AMEC tools clear qualification, Western equipment suppliers would face a newly credentialed competitor inside blue-chip Korean fabs. Neither Samsung nor SK Hynix has disclosed how far the testing has advanced or which specific process steps are under review. The timing coincides with separate shareholder pressure on both companies: investors have called for larger returns from cash reserves built on AI-driven memory demand, and both chipmakers have signaled they intend to strengthen shareholder return programs in response. Qualifying cheaper domestic Chinese tools would, incidentally, help free up capital for exactly those payouts.
On an adjacent front:
AMEC, formally Advanced Micro-Fabrication Equipment Inc., is China’s leading domestic etch tool maker and has been steadily expanding its product portfolio toward advanced process nodes.
US export controls have already restricted Samsung and SK Hynix from importing certain cutting-edge Western equipment into their China fabs, creating the supply vulnerability that AMEC is now being tested to fill.
Austin: Not good for Lam Research. This is one of the long-term consequences of export controls. They deny China access to Western tools, but they also give Samsung and SK hynix a reason to qualify Chinese alternatives as an insurance policy. Once AMEC is qualified, Lam and Applied Materials are no longer the only game in town inside those China fabs.
Anthropic Assembles Chip Team as AI Labs Seek Nvidia Alternatives
Anthropic is building a dedicated in-house chip design team to develop custom silicon for its Claude models, according to Reuters, with the company actively posting engineering roles to staff the effort. The move follows OpenAI’s Jalapeno chip program, putting two of the three leading frontier AI labs on a path toward proprietary hardware. Neither company has disclosed a timeline for tape-out or production volume.
Sector Watch
Memory
Samsung unveiled zHBM prototype stacking memory directly on AI accelerators alongside a detailed AI memory roadmap targeting the digital memory wall. (The Elec)
CoreWeave and Solidigm sign a multi-year storage agreement to scale AI storage capacity across CoreWeave’s expanding cloud infrastructure. (CoreWeave)
Samsung and Netlist settle all patent disputes under a new five-year memory licensing agreement, closing years of litigation. (Reuters)
Power
Onsemi says AI data center revenue will more than double in 2026 versus 2025 levels, citing sustained hyperscaler power-chip demand. (The Elec)
Power Integrations demonstrates the world’s first 2200V GaN power technology, targeting next-generation high-voltage industrial and data center power systems. (Power Integrations)
Optics & Networking
Veeco receives an order for its LUMINA+ MOCVD system for manufacturing indium phosphide lasers, supporting tightening InP supply for AI optical interconnects. (Veeco)
ams OSRAM reports Q2 revenue at the top of guidance and accelerates its microLED array roadmap for augmented-reality smart glasses. (EE News Europe)
Edge & Compute
Silicon Motion unveils its MonTitan SSD reference design kit at FMS 2026, targeting QoS requirements of agentic AI inference workloads. (Silicon Motion)
NXP Semiconductors will supply UWB chips to BMW for next-generation digital-key and occupant-detection systems across new vehicle programs. (The Elec)
Huawei launches the MateBook Pro S and MateBook Fold 2026, both running its in-house Kirin XE90 and Kirin X90 Plus processors on HarmonyOS. (Huawei Central)
Policy & Trade
GlobalWafers formally warns of emerging 12-inch silicon wafer supply constraints, signaling a potential pricing inflection for the wafer market. (The Elec)
Trump administration moves to impose a 15% tariff in its polysilicon probe, escalating US action against Chinese solar supply-chain dominance. (Reuters)
Indosat seeks a $2 billion loan to finance semiconductor and AI infrastructure buildout in Indonesia, tapping international credit markets. (Bloomberg.com)
Earnings Watch
Cisco scheduled its Q4 fiscal 2026 earnings call for August 12, 2026 at 1:30 PM PT. (Cisco)
Everspin Technologies reported preliminary unaudited Q2 2026 financial results for the MRAM persistent memory maker listed on Nasdaq. (Everspin)
Wolfspeed will report fiscal Q4 2026 earnings on August 19, with details to be released before the call. (Wolfspeed)
Vishay Intertechnology reported fiscal Q2 2026 results for the quarter ended July 4, 2026. (Vishay)