In this Talk Your Book episode, hosts Michael Batnick and Ben Carlson sit down with Dr. Ankur Crawford, portfolio manager at Alger and manager of the Alger Concentrated Equity ETF (ticker: CNEQ). Ankur holds a PhD, completed research at Intel, and physically worked inside semiconductor fabrication facilities — giving her a rare ground-level perspective on chip manufacturing that most Wall Street analysts lack.

The conversation revisits Ankur's April appearance on The Compound and Friends, where she had doubted her own bullish memory thesis because the market wasn't confirming it. Since then, Micron has surged 135%, Astera Labs has quadrupled, and Nebius has similarly exploded — vindicating her supply-chain-driven conviction. The discussion covers why the AI build-out is structurally different from past tech booms, how supply constraints across the semiconductor stack actually prevent bubble dynamics, the distinction between open-source and frontier AI business models, and why power infrastructure may be the most overlooked beneficiary of the AI supercycle heading into the next decade.

Market Vindication: Memory and AI Stocks Explode

Micron went from roughly $450 to $1,080 per share — up 135% — in the two months between Ankur's April appearance and this recording, after she had nearly talked herself out of the thesis because the market wasn't confirming it.

Astera Labs moved from ~$95 to close to $400 over the same period; Nebius also started near $95 and surged significantly, reflecting broad market recognition of AI trade duration.

"The market is coming to the realization that some of what is happening in the AI trade has duration" — Ankur, pointing to rising neocloud GPU pricing per hour as direct evidence of structural compute shortage.

Why This AI Cycle Defies Historical Bubble Comparisons

"There is no historical precedent for what is happening today in terms of how fast the innovation is occurring" — Ankur, arguing that using history as a guide is the core mistake bubble callers are making.

Digital innovation has no hardware penetration curve barrier: unlike the iPhone, which required physical device distribution, AI tools can be adopted instantly by anyone, creating near-vertical demand curves.

When software begins writing software, innovation becomes exponential — Ankur says we are at that inflection point now, and linear thinkers systematically underestimate what comes next.

Bubble talk around AI has persisted since at least 2017, according to Ben, suggesting the skepticism has been consistently early and consistently wrong.

Semiconductor Supply Chain Consolidation Creates Structural Pricing Power

Between 2010 and ~2022, the entire semiconductor supply chain — PCB makers, foundries, memory, semicap equipment — consolidated dramatically, shifting from a fragmented GDP-plus grower to a concentrated market.

TSMC's advanced foundry market has gone from roughly 15 competitors 15 years ago to effectively 1.75 players today, with Samsung and Intel still "finding their sea legs."

Even if TSMC wanted to triple capacity, it cannot: ASML, Lam Research, Applied Materials, and KLA-Tencor all face their own supply chain constraints on equipment manufacturing, creating a cascading rate limit on AI buildout.

"We are in fact rate limited by the supply chain, which is a really important aspect of what is happening. So we're almost capping the growth. So we can't get into bubble territory right now" — Ankur.

Ankur spent her PhD at Intel building wafers inside a fabrication facility in a bunny suit — a hands-on background she credits for her appreciation of how difficult and slow semiconductor manufacturing capacity additions truly are.

Valuation Framework: Get the 'E' Right Before Worrying About the PE

"The first thing you need to get right when you think about valuation is the E. Only then can you come up with a PE" — Ankur, arguing that most valuation critiques start from the wrong place.

GEV (GE Vernova) turbine pricing doubled from ~$1,250 to ~$2,500 per megawatt within a year; only 3 companies globally manufacture CCGT turbines, and GEV holds roughly one-third of that market.

Some AI supply chain stocks trade at low double-digit earnings multiples on Alger's differentiated (higher) estimates — appearing cheap rather than expensive once exponential earnings trajectories are modeled correctly.

Micron is the largest holding in the Russell 1000 Value Index, trading at roughly 8x earnings — the market applies a cyclical discount that Ankur believes is wrong given the structural shift in memory demand.

Portfolio holding period target is approximately 3 years with a goal of doubling or tripling; occasional trades are kept on a "short leash." The CNEQ portfolio holds roughly 30 concentrated names.

Hopper Pricing Surge and the Compute Shortage Signal

Hopper GPU (NVIDIA's first-generation AI chip) rental pricing is up 30-40% even after Blackwell and Grace Blackwell chips arrived — the opposite of what was expected, and a direct real-time signal of compute scarcity.

Nebius, which Ankur describes as the first AI-native hyperscaler formed in this era (rather than a traditional neocloud), is seeing this pricing surge on its older chip inventory.

Anthropic's ARR has climbed from $30 billion to approximately $44-49 billion, with Q2 profitability rumors — "when growth investors are telling you they've never seen anything like this, everyone needs to take heed" — Ankur.

Open Source vs. Frontier AI: Where Value Actually Accrues

Deepseek's emergence created a market misconception: open-source models will converge and commoditize, meaning value in open-source accretes to the infrastructure layer, not the model itself.

Frontier closed models retain value with the model holder because they provide differentiated intelligence — the business model is fundamentally different from open-source, not threatened by it.

"I am watching for if any of these open model suppliers can cross the chasm with something super innovative" — Ankur, acknowledging the innovation pace is fast enough that no player can be permanently counted out.

What Would Actually Cause Concern — and What Wouldn't

The one scenario that would genuinely worry Ankur: if $3 trillion in CapEx could be magically placed in the ground instantly, eliminating the supply constraint that currently acts as a natural governor on the cycle.

An algorithmic breakthrough enabling dramatically more output with far less compute would also force a thesis re-examination — though Ankur notes Deepseek did not actually represent this, given the open vs. closed model distinction.

Ankur welcomes skepticism: "I love the skepticism because if everyone was not a skeptic, then everyone would be on the same side as us" — broad consensus would be a more worrying signal than persistent disbelief.

Power Infrastructure: The Most Overlooked AI Beneficiary

Looking back from 2031, Ankur believes power and the industrial ecosystem will be seen as the most underappreciated area — capacity is even harder to bring online than semiconductor fabs.

Independent Power Producers (IPPs) currently trade at low-teens multiples on 2027-2028 earnings estimates — not stretched future-year numbers — making them attractively valued relative to the structural demand from AI data centers and industrial reshoring.

The broader portfolio thesis extends beyond AI: Ankur sees the US economy transitioning from consumer-led to industrial-led growth, with reshoring and CapEx expansion driving a long economic tail — reflected in holdings like QXO (building products) and Heico (airplane parts).

"We don't have enough power for the eventual reshoring of a lot of industry and a lot of capacity back to the US" — Ankur, framing power scarcity as a decade-long structural constraint, not a near-term bottleneck.

Resources Mentioned

Working at Intel

Alger and manager of the Alger Concentrated Equity ETF (ticker: CNEQ). Ankur holds a PhD, completed research at Intel, and physically worked inside semiconductor fabrication facilities — giving her a

We the People Journal (Diary, Notebook)

ut? Look, I think that there's some big picture things that I haven't changed my mind on. There's a paper that we wrote in early '23, probably March, April of '23. It was called AI and the Declining C

Supply Chain Notebook Funny Supply Chain Management Logistics Job Composition Notebook, Purchasing Procurement Officer Department Journal Gift (120 Lined Pages, 6'' x 9'')

can they literally also not manufacture as fast as supply, as fast as demand wants them to? Let's study that supply chain as well. TSMC is not the holder of all equipment, right? So they need to ask

POPRUN Dotted Journal Spiral Notebook, B5-7-1/4 x 9-1/2", 220 Pages, 120 GSM Dot Grid Notebooks, Hardcover Spiral Journals for Writing Journaling Drawing Women Men Work School, Black

lder of all equipment, right? So they need to ask ASML to make them equipment. They need to ask Lam Research and AMAT and KLA-Tencor for that equipment as well. So the equipment that goes into these b