Gavin Baker, founder of Atreides Management and one of the most closely followed technology investors on Wall Street, said Wednesday evening that risk/reward has become attractive again, pointing to cheap stocks with durable competitive advantages that he expects to crush earnings estimates for the next six to twelve quarters. Bill Ackman, whose Pershing Square Capital Management manages roughly $15 billion, replied with a single word: "Yes." Neither named specific stocks. The data makes clear what they were looking at.
The Philadelphia Semiconductor Index has fallen 13% in July alone, and is now sitting roughly 25% below its June peak, despite the underlying businesses continuing to report the strongest earnings in their history. MU closed Wednesday at $851, implying a 2027 forward PE of approximately 5.4 times on consensus EPS of $154. MRVL is down more than 32% in the past month. ANET has given back more than 35% from its peak. These are not businesses whose competitive positions have deteriorated. They are businesses whose stock prices have disconnected from their earnings trajectories.
What "Durable Competitive Advantages" Actually Means at This Price Level
The specific language Baker used is analytically precise. A durable competitive advantage is not just a strong business. It is a structural position that prevents competitors from replicating the earnings even if they try. In AI infrastructure, that description fits a narrow set of companies.
Micron is the only US-based HBM manufacturer at scale. HBM manufacturing requires years of process expertise accumulated through incremental yield improvement, and the gap between SK Hynix, which leads, and the rest of the field cannot be closed by capital spending alone. Micron has $100 billion in contracted take-or-pay revenue with floor pricing at margins that management said exceed prior peak cycle levels. That contracted base does not disappear when sentiment turns negative.
Arista Networks holds the dominant position in high-speed Ethernet switching for AI clusters and has surpassed Cisco in data center market share. Every AI cluster that Micron's memory goes into needs Arista's networking layer. That demand does not route around the equipment. Vertiv is the only scaled pure-play in liquid cooling, and Nvidia's reference architectures for Blackwell and Vera Rubin both specify liquid cooling by design. Neither Arista nor Vertiv has a credible near-term replacement.
These are the kinds of structural positions Baker means. The argument is not that the stocks are cheap relative to history. It is that they are cheap relative to the earnings they are structurally positioned to generate over the next eighteen to thirty-six months, which is what six to twelve quarters means in calendar terms.
Why the Sell-Side Has Not Cut Estimates Yet
The most analytically important feature of the current dislocation is that despite MU falling 32% from its all-time high, the consensus 2027 EPS estimate of $149 to $154 has not been materially revised lower. Forty-five analysts cover Micron with a Strong Buy consensus and an average price target of $1,486, representing more than 60% upside from Wednesday's close. Not a single analyst carries a Sell rating.
Analyst models move when the inputs move. The inputs are DRAM and NAND pricing as tracked by TrendForce, Counterpoint, and Omdia, HBM contract pricing, and hyperscaler AI capital expenditure. As of this week, none of those inputs have deteriorated. TrendForce has not cut DRAM price forecasts. The four major hyperscalers have not announced AI capex reductions. TSMC's Q2 report on Wednesday actually raised its full-year capital expenditure guidance from $52 to $56 billion to $60 to $64 billion, an upward revision of $8 billion in a single quarter update. TSMC does not build capacity speculatively. It builds when customers sign agreements. An $8 billion capex revision upward is a statement about demand visibility from Nvidia, Apple, AMD, and every other major fabless designer.
The gap between the stock price and the earnings model is the dislocation Baker is pointing at. Either the earnings model will eventually be cut to justify the price, or the price will recover to reflect the earnings model. The sequence of events over the next six weeks will begin to determine which.
The Korean Leveraged ETF Overhang Is Being Removed
One structural source of selling pressure that has been amplifying the semiconductor correction is now being addressed by regulators. South Korean authorities announced Thursday they are reversing their approval of single-stock leveraged ETFs tied to Samsung Electronics and SK Hynix, citing the outsized market volatility those products contributed to since their launch in May. The assets under management of leveraged memory stock ETFs have already shrunk by 34% from their June peak, according to JPMorgan's Nikolaos Panigirtzoglou, compared to a 13% decline for leveraged equity ETFs broadly. The AUM-to-market-cap ratio for leveraged memory ETFs is three times higher than for the overall equity ETF universe, meaning mechanical rebalancing has been amplifying every down move by a factor that is not present in other sectors. Removing that amplification mechanism reduces the volatility asymmetry that has made the correction feel more severe than the underlying fundamentals warrant.
The Earnings Confirmation Schedule
Baker's six to twelve quarter framing points to a specific earnings realization timeline that begins immediately. SK Hynix reports July 22 and is expected to deliver the strongest quarter in company history. Big tech reports through late July and early August, providing direct disclosure of AI capex commitments from Microsoft, Meta, Google, and Amazon. Micron reports again in late September. Each successive earnings event provides another opportunity for the fundamental demand thesis to be confirmed at the financial statement level.
If the $149 to $154 fiscal 2027 EPS consensus holds or is revised upward through these reporting cycles, a 5.4x forward PE becomes analytically untenable against the earnings evidence. The market is currently pricing in a deterioration of the EPS outlook that has not appeared in the sell-side models and has not been validated by any primary data from the companies or their customers.
Baker's observation is ultimately a statement about asymmetry. At 5x forward earnings, the downside requires the EPS to deteriorate significantly from levels that every current input variable supports. The upside requires the EPS to hold and the multiple to normalize toward even a modest premium to cyclical peers. That asymmetry is what he called attractive. Ackman agreed.
Explore more exclusive insights at nextfin.ai.
