DVB
Deep-Value Investor, Ex-Tech Executive, Agentic Equity Research Builder
Ambarella's turn toward AI chips comes too late, and Nvidia, Qualcomm, AMD and Intel's Mobileye should steamroll it with vertical integration and LLM-style chip design expertise. Once GoPro's main chip supplier, it sat for five years without accomplishing much. It has a history of underperformance and lackluster revenue growth, and co-founder and chief chip designer Leslie Kohn stepped down as part-time CTO. Its 22% jump, which @StockSavvyShay ties to edge AI, came on no news. x.com
Ambarella ran too hard, its multiple expanding to price in a 3–5 year narrative, while next-twelve-month growth decelerates to 10–15%. Its SoCs pair with DRAM and NAND, so spiking memory costs pressure customers' builds and feed back as an indirect demand headwind. New revenue, such as a potential 10-year $800M LTA and $100M of lifetime robotics revenue, is long-dated and not yet secured. Physical-AI stocks will be hammered hardest in any correction, though the setup is interesting longer term. … Show more
Broadcom's custom AI chips for major customers position it as a clear winner in the continuing AI buildout. Kris Patel's case that efficiency and total cost of ownership leave Broadcom room to gain share from incumbent Nvidia is persuasive. Broadcom has not seen peers' upside and is 30% below its highs, making the stock an opportunity. x.com x.com
Ambarella's appeal as a perfect acquisition target is playing out sooner than expected. Edge AI chips make up 80% of its revenue, which hit a record $391M in FY2026, up 37%, and it has 370+ customer projects in production, a Hanwha deal worth $800M+ and a 15-year CVflow software moat with switching costs measured in years. Its CV3-AD chip also handles camera, radar and lidar on a single die. x.com
Broadcom's custom AI chips and networking could drive earnings beyond what its valuation implies over 2-3 years. It is the favorite semiconductor opportunity now, preferred over NVIDIA without replacing it. AI semiconductor revenue is projected at about $58B in FY2026, $115B in FY2027 and potentially $230B in FY2028. At about 17x FY2027 consensus adjusted earnings, risk/reward is compelling if growth materializes; customer commitments add visibility, but delays and financing remain risks. x.com … Show more
Taiwan Semiconductor manufactures chips for competing AI designers, so its upside does not depend on which designer wins. Nearly every AI chip comes from the same fabs; Nvidia, AMD and custom-silicon competition all feed that business. The risk is that the whole AI race slows: 68% of revenue is tied to high-performance computing. x.com
Ambarella could be a robotics winner over the next few years, with a decent chance of accelerating revenue that would be very good for the stock. After a 50% retracement from recent highs, the stock is trying to bounce off the VWAP from its March lows, with earnings now past. As Stanley Druckenmiller says, the present is not what moves stock prices. 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
NVIDIA, Broadcom, Micron and TSMC should be worth less because AI risk has risen while the expected reward has not. The CEOs of Anthropic, OpenAI, SpaceX and Tesla have said AI risks killing all humans and needs to slow down, yet not a single analyst cut ratings or price targets on these chipmakers. In game-theory terms, higher risk with the same reward makes the bet worse. x.com
Inference's move toward edge devices and dense, memory-hungry racks should drive demand for specialized chips, links, power delivery and cooling. McKinsey projects inference at 60% of AI computing demand by 2030. Ambarella makes on-device AI chips; Astera Labs links GPUs, memory and networks; Credo supplies short copper connections; Rambus earns from memory interfaces; Navitas reduces power waste; Modine cools dense racks; and Super Micro builds inference-specific server racks at volume. x.com … Show more
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
