Investors should prioritize Micron (MU) and other memory leaders like SK Hynix, as a structural supply shortage is expected to double or triple DRAM prices through 2027. NVIDIA (NVDA) remains a high-conviction hold as older GPU models maintain high rental value and the upcoming Blackwell launch scales compute power to unprecedented levels. To capitalize on TSMC’s (TSM) projected $100 billion capital expenditure cycle, investors should look toward equipment suppliers ASML, Applied Materials (AMAT), and Lam Research (LRCX). Beyond chips, the skyrocketing demand for AI-driven reinforcement learning makes CPUs a critical, "sold-out" bottleneck worth monitoring in cloud infrastructure. Finally, watch for a major breakthrough in Robotics over the next 6 to 18 months, which will shift market focus toward physical hardware and actuators.

Anthropic (Private)

• Anthropic is experiencing explosive revenue growth, moving from an estimated $9 billion to a projected $40-$45 billion ARR (Annual Recurring Revenue) in a very short window. • The company recently released Opus 4.7 and has an internal model called Mythos that is rumored to be a massive leap in capability (moving from "L4" to "L6" engineer level). • Demand for their tokens is so high that they are operating with high gross margins (estimated floor of 72%), despite the massive compute costs. • They are currently "sold out" of capacity, leading to strict rate limits for non-enterprise users.

Takeaways

• Enterprise Priority: For businesses, securing an Anthropic Enterprise contract is becoming a competitive necessity to bypass rate limits and gain early access to frontier models. • Margin Expansion: The shift from 30% to 70%+ gross margins suggests that frontier AI labs have immense pricing power as long as they maintain a lead in intelligence. • Access Arbitrage: Investment firms (like Citadel) may seek exclusive "first-look" access to tokens, creating a new form of market advantage where the wealthiest firms get the smartest models first.

NVIDIA (NVDA)

• NVIDIA continues to dominate the hardware layer with 75% gross margins. • The "useful life" of GPUs is proving to be much longer than the 5-year estimate many analysts predicted; older A100 and H100 clusters are being re-signed for additional years at higher prices. • The upcoming Blackwell chips represent a massive scale-up in compute power (100,000 Blackwells being equivalent to hundreds of thousands of previous-gen chips).

Takeaways

• Extended Asset Value: The longer-than-expected lifespan of GPUs suggests that the depreciation hit to companies buying them might be lower than feared, while the rental value remains high. • Supply Chain Tightness: Bottlenecks remain in specialized components like copper foil, glass fibers for PCBs, and lasers, which are essential for NVIDIA’s hardware.

Taiwan Semiconductor Manufacturing Co. (TSM)

• TSMC is aggressively increasing capital expenditure (CapEx). Dylan Patel predicts they could spend $100 billion on CapEx by 2028. • They are currently squeezing every possible bit of output from existing fabs to meet demand. • Unlike memory providers, TSMC has been slower to aggressively hike prices, potentially leaving some "value" on the table for their customers (like NVIDIA and Apple).

Takeaways

• Long-term Infrastructure Play: The massive projected CapEx suggests a multi-year "super-cycle" in semiconductor manufacturing equipment. • Downstream Beneficiaries: Companies that supply TSMC, such as ASML, Applied Materials (AMAT), and Lam Research (LRCX), are positioned to benefit from this massive spending trajectory.

Memory & DRAM Sector (MU, SK Hynix, Samsung)

• Memory is identified as a critical bottleneck. Capacity can only grow at low double-digit percentages annually. • DRAM prices are expected to double or triple from current levels because capacity cannot be added fast enough to meet AI demand. • New capacity decided upon today likely won't hit the market until 2027 or 2028.

Takeaways

• Bullish Sentiment: The "memory story" is not overplayed; the supply-demand imbalance is structural and will likely lead to sustained high pricing and record margins for memory makers. • Demand Destruction: High prices will be the only way to "ration" supply, meaning only high-value AI applications will be able to afford the latest memory-intensive hardware.

Investment Themes & Sectors

The "Software-Only Singularity" and Robotics

• We are entering a phase where software implementation is becoming "cheap and easy" due to AI. • The next frontier is Robotics. Patel expects a breakthrough in the next 6 to 18 months in "few-shot learning" for robots (showing a robot a task once or twice for it to learn). • This will shift the focus from software back to physical hardware and actuators.

CPU Demand in AI

• While GPUs get the headlines, CPUs are currently "sold out" and seeing skyrocketing demand. • Reason 1: Reinforcement learning environments (where AI "practices" tasks) run on CPUs, not GPUs. • Reason 2: The "slop code" and apps generated by AI must ultimately be hosted and run on traditional CPU-based cloud servers.

"Phantom GDP" and Deflation

• AI is creating massive economic value that isn't showing up in traditional GDP stats. • Example: One person using AI can do the work of 200 economists or a full engineering team. This is highly deflationary for services but creates "Phantom GDP" where output rises while costs (and thus measured GDP) might actually shrink.

Risk Factors: Social Backlash

• Patel predicts large-scale protests against AI companies within the next three months. • Public perception is turning negative due to fears of job displacement and the "uncharismatic" communication from AI lab CEOs. • Investment Risk: Regulatory crackdowns or social unrest could impact the speed of data center build-outs and power consumption approvals.

Mentioned Private Companies & Tools

• Ramp: AI-automated expense management (Sponsor). • WorkOS: Enterprise-ready APIs for AI startups (Sponsor). • Rogo (Felix): Personal finance AI agent for investment firms (Sponsor). • Vanta: Automated compliance and security (Sponsor). • Ridgeline: Unified platform for asset management (Sponsor). • CoreWeave: Specialized GPU cloud provider.