DeepSeek Part Two: Why AI Skeptics Are Wrong. 10 Stocks That Will Win.

Tae Kim · Key Context by Tae Kim · March 30, 2026 at 22:35 · ⏱ 5 min read  | Read on Substack ↗
Summary
The article argues that the current selloff in AI stocks is driven by transient macro fears (Iran war, peak AI capex concerns) and ignores the structural shift from training to inference, specifically the explosion in AI agent compute demand. This means the underlying fundamentals for AI infrastructure remain extremely strong, and the market is underestimating the durability and growth of compute requirements from agents, coding assistants, and next-generation models.
  • Inference now accounts for 90% of data center power usage, per Nvidia's chief scientist Bill Dally, and hyperscalers face a severe compute shortage for inference despite hundreds of billions in capex.
  • Agentic coding assistants are driving 15x token demand vs. humans, according to Arm CEO Rene Haas, and companies are in an AI arms race to iterate faster.
  • Intel's CFO disclosed discussions of 3–5 year supply agreements with hyperscalers for CPUs, reflecting durable demand as CPU cores needed per gigawatt of AI data center capacity quadrupled to 120 million.
  • Nvidia's networking segment grew 263% YoY, a leading indicator of future GPU shipments, and Jensen secured the full supply chain from TSMC wafers to HBM to cables.
  • Amazon has rearchitected its entire advertising business around GPUs and LLMs for personalized product listings, showing real-world monetization of AI at scale.
  • Anthropic's upcoming 'Mythos' model and OpenAI's next model are expected to deliver a large leap in reasoning and coding, while Nvidia's Vera Rubin GPU is slated for launch later this year.
Read time 5 min
Length 5,873 chars
Category finance
Trade Ideas
Tae Kim Senior writer, Barron's; author of The Nvidia Way
Author states Nvidia's Jensen 'secured the entire supply chain, from TSMC wafers, HBM memory chips, and CoWoS to the physical connectors and cables,' directly implicating TSMC as the sole advanced fou
Author states Nvidia's Jensen 'secured the entire supply chain, from TSMC wafers, HBM memory chips, and CoWoS to the physical connectors and cables,' directly implicating TSMC as the sole advanced foundry for Nvidia's GPUs and likely for other AI chips mentioned. Risk: Geopolitical risk (Taiwan) and capacity allocation constraints could cap upside; any slowdown in AI spending would reduce wafer demand.
Tae Kim Senior writer, Barron's; author of The Nvidia Way
Author quotes Arm CEO Rene Haas stating token demand increases 15x with AI agents and that CPU cores per gigawatt has quadrupled to 120 million. Arm architecture is widely used in data center CPUs for
Author quotes Arm CEO Rene Haas stating token demand increases 15x with AI agents and that CPU cores per gigawatt has quadrupled to 120 million. Arm architecture is widely used in data center CPUs for orchestration, including Nvidia's Grace CPU and Amazon's Graviton. Risk: Arm's royalty revenue from data center CPUs is still a small fraction of total; competitive pressure from x86 and RISC-V could limit long-term share gains.
Tae Kim Senior writer, Barron's; author of The Nvidia Way
Author explicitly names AMD alongside Intel as seeing surging CPU demand from agentic AI orchestration needs. The article's CPU shortage thesis applies equally to AMD's server CPU lineup.
Author explicitly names AMD alongside Intel as seeing surging CPU demand from agentic AI orchestration needs. The article's CPU shortage thesis applies equally to AMD's server CPU lineup. Risk: AMD's GPU market share in inference remains small relative to Nvidia; CPU gains could be offset by share loss to Arm in some workloads.
Tae Kim Senior writer, Barron's; author of The Nvidia Way
Author extensively details Nvidia's pole position: networking up 263% YoY, acquisition of Groq for ultra-low-latency inference, entire supply chain secured, and chief scientist stating inference is 'T
Author extensively details Nvidia's pole position: networking up 263% YoY, acquisition of Groq for ultra-low-latency inference, entire supply chain secured, and chief scientist stating inference is 'THE JOB now.' Nvidia is the central beneficiary of the agentic AI shift described. Risk: Macro fears could persist longer than expected; any easing of capex plans by hyperscalers would hit NVDA first.
Tae Kim Senior writer, Barron's; author of The Nvidia Way
Author highlights a massive CPU shortage driven by agentic AI orchestration, with Intel's CFO discussing 3–5 year supply agreements with hyperscalers. CPU core demand per gigawatt has quadrupled, dire
Author highlights a massive CPU shortage driven by agentic AI orchestration, with Intel's CFO discussing 3–5 year supply agreements with hyperscalers. CPU core demand per gigawatt has quadrupled, directly benefiting Intel's data center CPU business. Risk: Intel still faces foundry transition risks and competition from AMD and Arm-based CPUs in the data center.
Tae Kim Senior writer, Barron's; author of The Nvidia Way
The article notes HBM memory chips are part of Nvidia's secured supply chain and that inference demand is exploding. HBM tightness is a recurring theme in AI infrastructure; Micron is a major HBM supp
The article notes HBM memory chips are part of Nvidia's secured supply chain and that inference demand is exploding. HBM tightness is a recurring theme in AI infrastructure; Micron is a major HBM supplier alongside SK Hynix and Samsung. Risk: HBM pricing could normalize as supply catches up; Micron's HBM3E qualification and volume ramp is still in early stages versus peers.
More from Key Context by Tae Kim

This newsletter, published March 30, 2026, features Tae Kim discussing TSM, ARM, AMD, NVDA, INTC, MU. 6 trade ideas extracted by AI with direction and confidence scoring.

Speakers: Tae Kim  · Tickers: TSM, ARM, AMD, NVDA, INTC, MU