bubble boi
Semiconductor Market Commentary, Thru Project Mascot
ASIC design is a race to the bottom, very bearish for Broadcom, MediaTek and Marvell as more leading-edge wafer capacity comes online from TSMC, Intel and Samsung. Google sources IP and swaps vendors constantly, showing the design houses add little differentiated value, transceivers aside. AMD, Apple and Nvidia do not outsource because it dilutes margins and most IP is one-size-fits-all. At scale, wafer allocation is the real constraint; the shift may take a year or two. x.com x.com
Broadcom and MediaTek risk being bypassed after Google reportedly told TSMC it wants to become a major direct customer, per Tech Taiwan. That points to Google's ultimate goal of an Apple-style COT model, though both partners internally still see the shift as a distant threat. x.com x.com
Semiconductor trade faces a capex growth ceiling even if AI delivers what bulls expect. Spending cannot compound at 70-100% a year indefinitely, and compute per dollar roughly doubles every two years, so the same budget buys more capacity later. Today's chip valuations also demand years of 20% AI earnings growth beyond 2028: roughly 10.9 extra years for AMD, 8.9 for ASML, 7 for Marvell and 6.4 for TSMC, versus 1.6 for Broadcom and 0.8 for Nvidia; Micron needs none under the chart's assumptions. … Show more
Broadcom's integrated optics, electronics and packaging could lower custom-chip system costs because separate suppliers defend their margins. At similar valuations, it is preferred to Nvidia and has more room to beat expectations. Google TPU v8 shipments are estimated at 4.0M in 2026 and 9.1M in 2027, assuming Broadcom supplies the remainder after MediaTek's allocation. Anthropic financing adds value only if chip profit exceeds likely support payments and capacity costs. x.com x.com
US sovereign AI and defense demand for domestically made compute will bring Intel most, if not all, US semiconductor designers as customers. PC weakness also lets Intel shift wafers to higher-margin server CPUs, but that takes a few quarters; capacity to ship more server CPUs is the limiter. The $240 billion 2030 server CPU market forecast is a stretch case, not the base case, versus roughly $200 billion consensus. x.com x.com x.com x.com x.com
Broadcom's roughly 60% margin looks unsustainable because its custom-chip design role is not irreplaceable. Google's TPU v8i has already succeeded, and it is being handled by MediaTek, whose margin is at best in the 30% range. That undercuts the claim that customers need Broadcom's design capability to compete with Nvidia, especially when Broadcom's margin is not far from Nvidia's own. x.com x.com x.com x.com x.com x.com
Rising 1.6T switching demand favors optical suppliers, though their execution risks vary. Bank of America says the market could grow ~101% annually through 2030. Marvell's leading DSPs and Lumentum's supply-short 200G EML lasers are lowest risk; Applied Optoelectronics' capacity expansion and Fabrinet's customer concentration make them medium risk. Aeluma remains pre-commercial, while Lightwave Logic's polymer modulators are a high-risk moonshot despite its TSM partnership. x.com
Broadcom and MediaTek risk losing chip ownership and margin as hyperscalers pull custom silicon in-house for platform control and margin. @jukan05 relays a rumor that Google will move to customer-owned tooling from TPU v9, ordering wafers directly from TSMC like AWS's Annapurna and reducing Broadcom and MediaTek to IP and design-service roles at lower per-chip margins. Even if not now, it will happen eventually, because the end state is all about optimization. x.com
VeriSilicon is China's Broadcom and should take off as custom inference ASIC orders concentrate at mature ASIC design service providers. Jalapeño's 9-month tape-out shows inference ASIC design barriers are lower than assumed, because the slow work of place-and-route, timing signoff and foundry adaptation sits with turnkey platforms. Every hyperscaler wants custom inference chips but won't build back-end teams, so pipelines at Broadcom, Marvell, MediaTek and VeriSilicon should thicken. x.com … Show more
Marvell's end-to-end networking approach is well suited to heterogeneous GPU, CPU and ASIC compute and mixed optical/copper technologies. Its shift from AI loser to winner deserves a lot of credit: its old custom-ASIC focus had lost out to Broadcom and Alchip, but it is now back in ASICs as Google's TPU v10 partner. The stock's excellent run reflects how it is now perceived. x.com
Alphabet and Amazon are set to dominate chip design over the next five years; supply-chain bottlenecks are the main thing holding their chip businesses back, and those won't last forever. Per a Tech Taiwan report shared by @jukan05, Google told TSMC it wants to become a direct major customer, pursuing an Apple-style COT model that bypasses Broadcom and MediaTek. x.com
