Mandeep Singh

Alphabet's TPU and vertical integration differentiate it from other hyperscalers. As long as they can defend the search business, the company is expected to deliver a clear earnings beat and raise guidance.

Meta is not only deploying Nvidia GPUs but is also "talking about using NVIDIA CPUs," moving away from the traditional x86 architecture. Historically, an AI server rack consisted of Nvidia GPUs paired with Intel or AMD CPUs. By switching to Nvidia's own CPUs (Grace), Nvidia captures 100% of the silicon value in the rack. This increases revenue per unit and deepens the competitive moat by creating a tightly integrated, proprietary ecosystem. Long NVDA as they successfully expand from being a GPU accelerator company to a full-stack data center provider. Regulatory intervention regarding market dominance or supply chain bottlenecks at TSMC.

Intel is buying back the rest of its Irish data center JV with Apollo for $14B, a deal that makes strategic sense amid a data center construction boom and a push for onshore manufacturing. The move allows Intel to double down on controlling manufacturing capacity and supply chain diversification, key themes in the current geopolitical and tech landscape. The strategic rationale for consolidating this asset is clear and aligns with favorable macro trends, supporting a positive view. The capital outlay could strain Intel's finances if the broader semiconductor cycle turns down.

South Korea is backing Samsung and SK hynix with an $800 billion investment plan to expand memory capacity. This government support ensures capacity expansion even if market dynamics change, similar to how TSMC benefited from Taiwan's backing. Memory demand is outstripping supply for data centers and AI, and new capacity won't come online before late 2027 or 2028, keeping conditions tight and benefiting incumbents.

The shift to agentic AI is driving huge demand for memory, with Micron's gross margins surging from negative to 81% in a few quarters. Supply remains extremely tight, and despite capacity additions, the shortage will persist beyond 2026. The cycle length and room to grow will exceed expectations, supporting continued price and margin expansion.

Hyperscalers have a combined backlog of commitments totaling about $2 trillion, driving steady demand for chips; the AI capex cycle will continue until a major spender like Meta explicitly cuts spending.

Snowflake is well-positioned because agentic AI requires connecting enterprise databases to LLMs, and Snowflake has launched its own coding agent, making it a key beneficiary.

South Korea is backing Samsung and SK hynix with an $800 billion investment plan to expand memory capacity. This government support ensures capacity expansion even if market dynamics change, similar to how TSMC benefited from Taiwan's backing. Memory demand is outstripping supply for data centers and AI, and new capacity won't come online before late 2027 or 2028, keeping conditions tight and benefiting incumbents.

SpaceX стремительно превращается в гиперскейлера вычислительных мощностей для ИИ. Приобретение Cursor может вывести компанию в лидеры среди поставщиков передовых моделей ИИ, конкурирующих с Anthropic и OpenAI, за счет использования орбитальных вычислений и наземных центров обработки данных. Ожидается, что выручка вырастет в три раза с текущих 18 миллиардов долларов благодаря сделкам в сфере вычислительных мощностей. SpaceX/xAI располагает центрами обработки данных с доступом к электроэнергии, что является ключевым конкурентным преимуществом на фоне того, как другие компании сталкиваются с трудностями при обеспечении своих чипов питанием. Бизнес по аренде вычислительных мощностей предполагает премиальное ценообразование на уровне 11 долларов в час за графические процессоры Blackwell, что выше, чем у новых облачных провайдеров, таких как CoreWeave и Nebius, и будет способствовать росту маржинальности. SpaceX может увеличить капитальные затраты до 100 миллиардов долларов, сравнявшись с другими гиперскейлерами в 2026–2027 годах.

Legacy company Dell is guiding to ~48% top line growth due to new AI server line, driving a reset in expectations and stock surge.

The recent volatility in AI infrastructure and semiconductor names is driven by market structure and leverage issues, not fundamentals. The fundamentals remain very solid with accelerating cloud growth and constrained supply chains for components.

SpaceX's enormous CapEx plans and the $1 trillion revenue target lack credibility given its minimal operating cash flow; the company will need massive debt or equity issuance, and the market is clearly skeptical, driving the stock down sharply.

Meta and Microsoft are already in bear markets with questionable AI ROI. There is a definite case where one of them blinks and cuts capex, which would further hurt their stocks.

Memory chip companies operate as an oligopoly, similar to oil, and have not increased capacity, leading to sustained memory shortage and elevated prices. The shift from AI training to inferencing multiplies memory demand per GPU by ten, reinforcing the bullish case. Prices are not expected to decline in the next 6-12 months.

Anthropic's momentum and demand for its models drives usage of Google and Broadcom chips, giving Broadcom a nice tailwind.