Most important take away

Ankur Crawford (Alger EVP/PM) argues AI is not a bubble because underlying numbers are re-rating rapidly and the market is genuinely short compute through at least 2026-2028. The key investing mistake is thinking linearly about an exponentially growing demand curve; the best opportunities are in chip/memory/storage suppliers (Micron, Western Digital), hyperscalers trading at unusually low multiples (Amazon), and AI-native infrastructure plays (Nebius), while much of traditional seat-based SaaS is structurally at risk.

Chapter Summaries

1. Intro & How Ankur Works

Ankur describes meeting Josh’s co-host Adam early in her career, her Stanford PhD background in material science, and how she uses tools like Quartr and Aera to consume conference calls. Her team listens to everything; she personally prioritizes hyperscalers and TSMC.

2. Alger’s Investment Philosophy

Alger is a growth shop founded in the 1960s by Fred Alger. They invest in two buckets: (1) High-Unit-Volume Growth (classic fast-growing disruptors) and (2) Life Cycle Change — mature companies that successfully reinvent themselves (Microsoft under Satya, Apple, Danaher, now Western Digital and Nvidia). Management quality is crucial for the second bucket.

3. AI Is Exponential, Not Linear

Ankur says AI capability has moved faster than she expected three years ago — agents she thought would arrive in 2028 are already here. She cites Claude-related tooling (OpenClock), Perplexity Computer, and a college-counseling app she vibe-coded in 60 hours for her daughter. Investors underperform because they extrapolate linearly from two data points while the actual curve is exponential.

4. Private Market Investing at Alger

Alger participated in the Anthropic $380B round (three weeks before recording) and has historically invested in Palantir pre-IPO. They can put up to 15% of 40-Act funds into private names, which they see as differentiating active growth management now that companies stay private until near-trillion-dollar valuations.

5. The Compute Shortage

“Everyone is short compute.” She expects supply/demand imbalance benefiting sellers through 2026-2028. Bottlenecks: chips, electricity, and increasingly DRAM. Uber reportedly blew its full 2026 compute budget by April. Tokens are “units of intelligence/compute,” and not all tokens are equal — drug-discovery tokens will price very differently than consumer queries.

6. Memory and Storage Re-Rating (Micron, Western Digital, Seagate, SanDisk)

Micron went from $85 (Aug) to ~$450; market cap from ~$175B to over $500B. WDC went from $15B to $125B. She says these are not bubbles: WDC’s 2026-27 earnings estimates jumped from $10 to $25 in three months — it’s a “different company.” Industry consolidation means only three DRAM players and two HDD players, so there’s no alternative to express the trade.

7. Why This Is Not 2000 / Not Solar / Not EVs

Solar failed because tech barriers were low and it relied on subsidies. The 1999 telecom buildout failed because the capex went into long-lived assets (fiber) before demand existed. This time the “car is built” — the only missing input is compute. Hyperscaler capex goes directly into assets that are utilized immediately with 18-month paybacks at current GPU rental rates.

8. Circular Deals & Dr. Burry’s Accounting Critique

She’s relaxed about Nvidia-style seeding (CoreWeave) as normal ecosystem behavior, but more cautious on zero-gross-margin equity-for-revenue deals (AMD/Meta, AMD/OpenAI). Dismisses Michael Burry’s hyperscaler depreciation claim — A100 chips from 2021 are still running; chips don’t die in three years.

9. Stock Picks (Public Holdings)

• Nebius (NBIS) — Spun from Yandex after Ukraine war; 1,500 engineers + $2B cash; potentially “the next AI-native hyperscaler.”
• QXO — Brad Jacobs’ new vehicle consolidating building products (Beacon Roofing, Kodiak). Plans to grow EBITDA from ~$1B to $5B by 2029-30.
• AppLovin (APP) — AI-driven ad engine now expanding from mobile-game ads into e-commerce. Founder-CEO Adam Foroughi keeps headcount flat; got unfairly caught in the software index selloff.
• Amazon (AMZN) — Teens gap PE, cheaper than Costco/Walmart; AWS growing mid-to-high 30s aided by Anthropic; Trainium is a cost-lowering optionality play, not a chip-sale story (Trainium 4 still “okay, not fantastic” vs. Google TPU).

10. Intel (INTC)

Not owned. Impressed with Lip-Bu Tan as a leader (his Cadence turnaround), but the foundry thesis is “a bet on engineers inside the fab” — she’s skeptical Intel Foundry can actually compete with TSMC.

11. SaaS Apocalypse

Disagrees with Jensen Huang’s defense of incumbent SaaS. Thinks Salesforce-style seat-based pricing and automatic annual uplifts cannot continue — customers only use 15-20% of what they pay for. Software terminal operating margins likely reset from ~40% to ~20%. Cybersecurity platforms (Palo Alto, CrowdStrike) more defensible than point solutions (Atlassian called out with “zero moat”). Palantir is owned in non-ETF funds.

12. Mag 7 “Halo” Ranking & Apple’s AI Problem

Ankur agrees Amazon has halo. On Apple: the iPhone rectangle is halo for now, but if Apple doesn’t ship a real agentic Siri, glasses/bracelets/wearables could disintermediate it within three years. She says Apple “doesn’t have the horses” and their security-first positioning limits agent ambition. Best Mag 7 business by gross margin is Nvidia, then Google, then Apple.

Summary

Macro Thesis & Actionable Insights

Compute shortage is real and persistent (2026-2028). She thinks capex plans remain under-sized versus exponential agent/inference demand. Agents unlocked demand faster than expected; Uber maxing out its 2026 compute budget by April is her canonical example. Actionable takeaway: stay long the AI infrastructure stack and don’t sell just because stocks look extended — evaluate on forward earnings revisions, not trailing multiples.

Think exponentially, not linearly. The most repeated thesis of the episode: the biggest PM mistake is connecting two data points and extrapolating a straight line. Investors from healthcare/consumer backgrounds tend to call “bubble” because the numbers don’t fit linear models.

Life-cycle change names are where unrecognized growth lives. Look for stocks trading at value multiples that are actually reinventing. Examples she cites positively: Western Digital, Micron, Nvidia (as a life-cycle change story, not a mere chip cycle), Amazon, Microsoft.

Raise cash tactically, don’t sell the thesis. In February she raised 7% cash across the Concentrated Equity Fund (CNEQ) not by dumping AI names but by trimming across the book, preserving conviction while building dry powder.

Specific Stocks & Investments Mentioned

Bullish / owned:

• Micron (MU) — example of memory re-rating; $85 → ~$450.
• Western Digital (WDC) — EPS estimates revised from $10 to $25 in 3 months; storage demand from AI-generated data.
• Seagate (STX) — other HDD duopolist (mentioned as part of the two-player structure).
• Nvidia (NVDA) — life-cycle change story; highest gross margins in Mag 7; CoreWeave-style seeding not worrisome.
• Amazon (AMZN) — one of her biggest positions; cheapest PE vs. comps; Trainium is cost-lower optionality.
• Nebius (NBIS) — potential next AI-native hyperscaler.
• QXO — Brad Jacobs roll-up of building-products distribution.
• AppLovin (APP) — ad engine expanding into e-commerce; founder-led.
• Palantir (PLTR) — owned outside the ETF; “shepherd” helping enterprises cross the AI chasm.
• Anthropic — participated in $380B private round.
• Alger Concentrated Equity ETF (CNEQ) — her own ETF, launched 2024.

Cautious / not owned:

• Intel (INTC) — good leadership in Lip-Bu Tan but skeptical on foundry execution.
• Salesforce (CRM) — ~10x EV/FCF ex-SBC is not compelling enough; seat-based uplift cycle is breaking.
• Atlassian (TEAM) — “zero moat.”
• Apple (AAPL) — halo at risk if no agentic Siri and no new form factor; still a great sit-on-top of others’ capex.

Circular-deal yellow flags:

• AMD/Meta and AMD/OpenAI deals (equity warrants for revenue at ~0% gross margin) — more worrisome than Nvidia/CoreWeave.

What Would Change Her Mind

• Announced capex plans materially decelerating (she sees the hyperscalers’ ability to dial back quickly as the asymmetric protection vs. 2000-era stranded fiber).
• A valuation-regime handoff problem: the market is currently confused transitioning hyperscalers from a free-cash-flow story to a reinvestment story, which is why they’ve derated despite strong fundamentals. Watch the handoff complete as a signal.

Secondary Takeaways

• All tokens are NOT created equal — this will eventually tier LLM pricing (drug-discovery tokens priced very differently than consumer “what’s the weather” tokens). This tiering could create both pricing-power winners and losers within the model providers.
• Sovereign and enterprise AI adoption is democratizing because non-coders can now build (her own 60-hour college-app vibe-code is the proof point).
• Screenless future (glasses/bracelets) is a multi-year risk for Apple AND for Meta’s ad business (Instagram/TikTok revenue model depends on the rectangle).