Either the expensive stocks are right, or the cheap ones are. There is no middle ground.
In the frothy corridors of artificial intelligence investing, a seasoned allocator has paused to notice something the crowd has overlooked: not all enthusiasm is priced equally. Gavin Baker, whose decade of outperformance gives his observations unusual weight, argues that memory and processor stocks sit quietly undervalued while optical chipmakers have been bid to heights that defy the underlying arithmetic. His diagnosis is not a prediction of collapse so much as a reminder that markets, even in their most exuberant moments, must eventually reconcile price with reality.
- A stark and growing valuation chasm has opened inside the AI sector — optical chipmakers like Lumentum trade at triple-digit earnings multiples while memory giants like Micron and even Nvidia remain comparatively cheap.
- Lumentum has surged tenfold in a single year, a gain driven not by superior earnings but by investors simply deciding to pay more — a fragile foundation when sentiment shifts.
- Baker's core tension is binary and unforgiving: if all these companies share the same AI tailwind, both ends of the valuation spectrum cannot simultaneously be correct.
- The semiconductor industry's history of violent boom-and-bust cycles haunts the thesis — if AI demand slows or saturates, the most expensively priced names face the steepest fall.
- The navigation being attempted is disciplined arbitrage — rotate toward the cheap, avoid the crowded, and let valuation serve as a margin of safety rather than an afterthought.
- The current trajectory favors the patient: cheaper stocks like Nvidia have grown into their valuations through earnings, not multiple expansion, giving them structural resilience the optical names lack.
Gavin Baker built his reputation over eight years running Fidelity's OTC Portfolio, compounding returns above 19 percent annually and outpacing 99 percent of peers. Now managing roughly $7 billion at Atreides Management, he carries a Sharpe ratio of 2.46 — a figure that signals not just strong returns, but returns earned without reckless risk-taking.
Speaking recently on the All-In podcast, Baker offered an unsettling read of the AI stock market. He called it "cross-sectionally inefficient" — meaning the prices of companies within the same sector have drifted into incoherence relative to one another. Memory chipmakers like Micron and processor giant Nvidia trade at what he considers bargain levels. Optical component companies like Lumentum and Coherent, meanwhile, command valuations that would only be justified by dramatically superior growth or margins. Neither company offers that superiority.
The logic Baker applies is clean and binary. If memory, processors, and optical components are all riding the same AI wave, then only one set of valuations can be right. Either the cheap stocks are mispriced and will rise, or the expensive ones are mispriced and will fall. There is no version of reality where both ends of the spectrum are correct at once.
What gives the thesis its edge — and its risk — is the semiconductor industry's long memory of brutal cycles. Memory chips in particular have historically swung between glut and shortage, punishing investors who bought at the peak of enthusiasm. The counterargument is that AI represents a secular shift rather than a temporary cycle, permanently elevating demand. If that's true, cyclicality loses its bite. If it isn't, the downside for the most expensive names could be severe.
For practical investors, Baker's framework resolves into a simple discipline: favor the cheap, avoid the crowded. Nvidia's stock has risen on earnings growth, not on investors paying an ever-higher premium — a far more durable foundation than Lumentum's multiple expansion. In a market where momentum can carry prices far beyond reason, valuation remains the one anchor that eventually pulls everything back.
Gavin Baker has spent the better part of a decade proving he knows how to pick technology stocks. During eight years running Fidelity's OTC Portfolio, he delivered compound annual returns above 19 percent and beat 99 percent of his peers according to Morningstar's rankings. Now, as chief investment officer of Atreides Management, a hedge fund overseeing roughly $7 billion in public and private investments, he continues that track record. His Sharpe ratio sits at 2.46—well above what typical hedge funds achieve—which means he's extracting outsized returns without proportionally outsized risk.
Recently, Baker offered a diagnosis of the artificial intelligence stock market that cuts against much of the current enthusiasm. Speaking on the All-In podcast, he described the sector as "cross-sectionally inefficient," a technical way of saying the prices don't make sense relative to each other. The divergence he's identified is stark. Memory chip manufacturers like Micron and SanDisk are trading at what he considers bargain valuations. Nvidia, the dominant player in AI processors, is also cheap by his measure. Yet optical chipmakers—companies like Lumentum Holdings and Coherent—command triple-digit price-to-earnings ratios despite no obvious superiority in growth or profitability.
Lumentum has jumped tenfold over the past year. Coherent has soared similarly. Both trade at valuations that would make sense only if they were growing dramatically faster or generating far superior margins compared to memory and processor stocks. They are not. The gap between what these optical companies are priced at and what memory and processor companies are priced at suggests a fundamental misalignment.
Baker's reasoning rests on a single assumption: that all these stocks—memory, processors, optical components—are riding the same AI cycle. If that's true, then the math becomes simple. Either the expensive optical stocks are correctly priced and memory and Nvidia should climb substantially higher, or the cheaper stocks are correctly priced and the optical names will eventually collapse. There is no middle ground where both valuations are right.
The semiconductor industry has a long history of violent cycles. Memory chips in particular have swung between gluts and shortages, sending prices and profits careening up and down. Investors have learned to be wary of the sector's boom-and-bust patterns. Optical chips have experienced cycles too, though historically less severe. What makes the current moment different, some argue, is that AI represents not a temporary cycle but a secular shift—a permanent elevation in demand that changes the baseline. If that's true, cyclicality becomes irrelevant. But if supply and demand dynamics shift, if AI adoption slows or saturates, the downside could be brutal.
Valuation discipline matters most when momentum fades. In the short term, a stock can climb on pure enthusiasm regardless of what it costs. Eventually, though, price catches up with reality. The cheaper stocks have an inherent advantage in that race. Lumentum's gains have come largely from multiple expansion—investors simply decided to pay more for each dollar of earnings. Micron and Nvidia have not benefited from that same multiple expansion. Nvidia's valuation has actually contracted even as its stock has risen, meaning the gains came from earnings growth, not from investors repricing the company.
For investors trying to navigate the AI sector, Baker's framework suggests a straightforward approach: buy the cheap names, avoid the expensive ones. If the entire AI sector rises together, the cheaper stocks should outperform over time. If the sector falls, the cheaper stocks will likely fall less. Either way, valuation becomes your margin of safety. The question now is whether the market will eventually agree.
Notable Quotes
The AI sector is cross-sectionally inefficient, with multiples that don't make sense relative to one another— Gavin Baker, Chief Investment Officer of Atreides Management
The Hearth Conversation Another angle on the story
When Baker says the sector is "cross-sectionally inefficient," what does he actually mean? Is he saying some stocks are overpriced or underpriced?
Both. He's saying the relative prices don't match the relative quality or growth. Lumentum and Coherent are trading at multiples that don't align with what Micron and Nvidia are trading at, even though the fundamentals don't justify the gap.
But couldn't the optical stocks be the real winners? Maybe they're expensive because they're more essential to the AI buildout?
That's possible, but Baker's point is that if they are essential and deserve those multiples, then memory and processor stocks should be worth far more than they currently are. The prices aren't internally consistent.
He mentions cyclicality in semiconductors. Does that mean AI demand could just evaporate?
Not evaporate, but shift. Memory chips have historically swung between shortage and glut. If AI adoption slows or the buildout phase ends, inventory could pile up and prices could collapse. That's the risk.
So the cheaper stocks are safer because they have less downside?
Partly. But also because if the sector does well, they have more room to run. If valuations normalize across the sector, the cheap ones catch up faster.
What would prove Baker wrong?
If optical stocks continue climbing while memory and Nvidia stagnate. That would suggest the optical companies really are in a different, better cycle than the others. But that would also contradict the idea that they're all riding the same AI wave.