Kristina Partsinevelos 5.0 16 ideas

NVIDIA faces increasing competitive pressure at the AI inference stage from custom chips developed by Broadcom and hyperscalers, which are cheaper and more efficient, potentially threatening NVIDIA's market dominance and creating investment risk as the AI market becomes more competitive.

Broadcom is establishing itself as a significant second player in the AI chip market through six major custom chip deployments with companies like Meta, Google, OpenAI, and Anthropic, and its multiyear build-out with Meta indicates a sustained revenue ramp, positioning it for growth in the expanding AI infrastructure space.

Nvidia CEO Jen-Hsun Huang stated that margins for enterprise software might come down because AI requires costly hardware and energy. Kristina Partsinevelos highlighted this as a significant, first-time public comment and noted that software companies like Oracle and SAP have extremely high margins (70-90%) from asset-light, subscription models. As AI becomes embedded in every company’s operations, the necessary hardware and energy infrastructure will increase costs, potentially compressing the historically high margins of enterprise software providers. This margin pressure represents a structural headwind for pure-play enterprise software companies, making them less attractive from a profitability perspective. AVOID reflects the risk of multiple contraction as the market digests this shift. Software companies may successfully pass on additional costs to customers, or AI adoption may proceed more slowly than expected, leaving margins intact.

The speaker reported that Super Micro is embroiled in a major smuggling scandal, its stock dropped ~33%, and analysts (Raymond James) say it will suffer a "reputational discount" over its reporting credibility. The report also highlighted past compliance issues (2018 delisting, 2024 auditor resignation). The criminal charges against company insiders, the scale of the alleged scheme, and the history of governance problems create severe reputational and legal risks that undermine investor confidence and the stock's valuation. The combination of immediate legal overhang, severe reputational damage, and a pattern of compliance failures makes the stock unattractive and risky. If Super Micro is fully exonerated, demonstrates robust new compliance controls, or if the legal fallout is contained without further impact on operations, the thesis weakens.

The speaker explicitly suggested that semiconductor equipment makers like Applied Materials and ASML are a "cleaner way" to play the tension in the memory market, as Micron sells off on fears its margins have peaked. The memory upcycle is driving significant capital expenditure from producers like Micron, which directly benefits the equipment suppliers. This investment is expected to continue for years, providing revenue visibility for equipment companies even if memory stock prices stagnate on peak-cycle concerns. These companies represent a way to gain exposure to the memory growth story while potentially sidestepping the direct margin and cycle-timing risks faced by memory producers. If the memory upcycle peaks sooner than expected or capex plans are scaled back, demand for semiconductor manufacturing equipment would decline.

Kristina stated that Nvidia CEO Jen-Hsun Huang's comment on memory optimization is a direct call out to memory suppliers like SK Hynix, Micron, and Sandisk, with Sandisk climbing higher on signals of accelerating demand. Nvidia's technology drives demand for high-bandwidth and low-power memory in data centers, benefiting these suppliers. Watch these memory companies as key beneficiaries of Nvidia's growth and increasing demand, despite current supply tightness. Supply constraints may persist, or competitive pressures could erode margins.

Kristina reported that Morgan Stanley's Joe Moore calls the recent sell-off an attractive entry point and keeps NVIDIA as his top pick in semiconductors. Moore believes the market will eventually get more comfortable with the 2027 growth number, and when it does, the stock will outperform. Long NVIDIA as the sell-off presents an opportunity for future outperformance based on strong analyst conviction. The market may not become comfortable with the growth projections, limiting upside.

"Recently the $2 billion deal with Lumentum and Coherent that's within the optical market. So we're expecting to hear news on how they're going to change their networking copper towards optical." Real-time AI inference requires massive data throughput with minimal latency. Traditional copper networking cannot handle the bandwidth and power efficiency required for next-generation AI clusters. As Nvidia officially transitions the industry standard from copper to optical networking, component suppliers like Lumentum and Coherent will experience a massive, multi-year revenue tailwind as they become critical infrastructure for AI data centers. LONG. These optical suppliers are direct, second-order beneficiaries of Nvidia's architectural shift and massive capital deployment. Supply chain bottlenecks in optical manufacturing, or Nvidia aggressively squeezing margins on its suppliers as volumes scale.

"Google, Amazon, Meta, startups like Cerebras are already all working on alternatives within the inference space." Hyperscalers have been spending billions developing custom in-house silicon (ASICs) specifically because Nvidia's older GPUs were too power-hungry and expensive for inference tasks. However, Nvidia's new purpose-built, memory-heavy inference chip directly attacks this vulnerability. If Nvidia's off-the-shelf solution is vastly superior, it neutralizes the hyperscalers' custom silicon ROI, ensuring they remain heavily dependent on Nvidia's ecosystem rather than breaking free. NEUTRAL. The thesis that hyperscalers will easily pivot away from Nvidia for inference is severely challenged by Nvidia's new product stack. Hyperscalers may still find their internal chips more cost-effective for their specific, proprietary workloads at massive scale, reducing their Nvidia orders over time.

"It's been five months since Nvidia told us they had a $500 billion backorder... If Jensen Huang updates that number and gives visibility into 2027, that's what moves the stock." The market has already priced in the hardware announcements (the new Groq-based inference chip). The critical variable for valuation expansion is proving that AI infrastructure spending is not a short-term bubble. By providing concrete demand visibility into 2027, Nvidia neutralizes the bear thesis of a cyclical peak, forcing analysts to revise long-term earnings models upward. LONG. The combination of capturing the inference TAM (75% of the AI market) and extending demand visibility makes the stock a buy through the GTC event. If management fails to provide clear 2027 visibility or the backlog number stagnates, the stock could suffer a "sell-the-news" contraction.