Nicholas Mugalli
Institutional TMT & Macro Research, AI/Semiconductors/Policy
Qualcomm's reported ~$4 billion purchase of Modular targets Nvidia's real moat, CUDA software, rather than its chips. Modular's hardware-agnostic, write-once-run-anywhere stack is the anti-CUDA, an escape hatch from Nvidia lock-in that lets Qualcomm's chips, and everyone else's, compete in software, where Nvidia has had no rival. Google with TPUs and Amazon with Trainium and Graviton made hardware bets; Qualcomm went after the software layer. x.com
Qualcomm's on-device thesis strengthens as agentic AI drives token demand beyond what the cloud can serve. At Computex, CEO Cristiano Amon projected tokens rising about 40x, from 31.7 billion every 10 seconds in 2026 to 1.27 trillion by 2030; routing all that through data centers at agent latency would not work physically or economically. Inference must spread across phones, PCs, cars and edge devices, where Qualcomm leads, and its new Dragonfly server rack chip pushes into the data center. x.com
Qualcomm could rerate from a handset chipmaker into an AI infrastructure story now that Amazon is co-developing multiple generations of inference silicon with it. As AI shifts toward stacks built around workloads, power and cost, Qualcomm need not beat Nvidia, only help optimize around the GPU. Amazon and Qualcomm are also working on 1.6T links using Qualcomm SerDes and optical DSPs, entering a space led by Broadcom and Marvell. Deployment volumes and further design wins would confirm it. x.com
Nvidia's integrated AI software stack could give it an edge beyond chip performance and pricing. @semivision_tw says the completeness of that stack may become the next competitive battleground for AMD, Broadcom and Marvell; Nvidia's open models can also drive demand for its compute, as @benitoz describes. x.com x.com
Qualcomm is not simply undervalued: as a pure chip designer it lacks both manufacturing and distribution to end developers, leaving it squeezed by the largest companies on earth. It sends designs to TSMC and has no vertical integration. Amazon, which has pricing power through AWS, is now using Qualcomm to design its next chip, pushing it to undercut Broadcom. Nvidia controls the CUDA developer ecosystem, so its designs will likely far outpace Qualcomm's. It could be wrong short term. x.com
Broader adoption of custom AI silicon could benefit Broadcom and Marvell. OpenAI reportedly plans to expand its use of Google's TPUs beyond Nvidia GPUs, and Google TPUs are coming to Oracle Cloud in 2027. Both are potential tailwinds for custom-chip suppliers. x.com
Qualcomm's data center option is not yet in the price, even after a 75% one-month rally that prices its edge AI story in smartphones, cars and glasses. Its AI200 and AI250 data center chips are going to a hyperscaler this year, taking Qualcomm directly into Nvidia's territory with more programmable chips, a point flagged by @KatieTarasov. x.com
Meta starting a cloud business looks bearish for the Qualcomm and Arm chips it had planned to buy. Meta may instead buy more chips suited to neoclouds, such as NVIDIA's. @lokoyacap argues the opposite. x.com x.com
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
Anthropic's $63B custom-chip deal bolsters Broadcom's case for growth from custom AI silicon. OpenAI's Jalapeño chip is seen as next, and the AI chip supercycle is just getting started. x.com
Qualcomm has a terrible position in every market it competes in and is not cheap enough to compensate. It faces Nvidia in cars, Intel and AMD in data-center CPUs, and Samsung and Google on their own phones. China is a material share of revenue, and Apple is aggressively cutting out Qualcomm's modem, a long-term revenue headwind. @citrini, bullish since December, made a good call, but the short stands; it also works as a funding short against long compute names such as AMD, SanDisk and Lumentum. … Show more
NVIDIA's free-model plan could defend chip demand as the two largest AI labs cut reliance on its hardware. OpenAI is building a chip with Broadcom; Anthropic mostly uses Google and Amazon chips. NVIDIA is putting together Reflection AI (in talks), Hugging Face and Groq for free models and paid chips. At 11x expected 2029 earnings versus ~19x for Microsoft, Alphabet and Amazon and 32x for Apple, its price reflects fears of slower AI spending and lost customers. NVIDIA is a 'Maximum Opportunity'. … Show more
