- Val Zlatev’s $1 billion Analog Century Long/Short hedge fund returned 61% in the first half of 2026 — a record for the strategy since its 2018 launch — driven by positions in memory chipmakers, storage-device makers, and networking equipment companies rather than the more widely owned AI platform names; the fund gained 9.4% in June alone, and Zlatev’s bigger $2.1 billion Market Neutral strategy (which includes a separately managed account from Millennium Management) returned 11.7% in the first half.
- Top winners include Sandisk Corp. (+858% in H1 2026) and Micron Technology (+304% in H1 2026), reflecting Zlatev’s thesis that memory is the constrained resource in the AI buildout — not compute; Zlatev told investors that memory prices have risen four to five times, driven by data centers and the expansion of AI “context windows” (the working memory that lets models process larger inputs), and expects further gains: “The memory guys are not prepared. Prices went up for the roof. They will go more.”
- Other Analog Century positions include Applied Materials, Lumentum Holdings, and Astera Labs — a portfolio skewed toward the hardware and connectivity layer of AI rather than the software or hyperscaler layer; the strategy validates the “picks and shovels” thesis for AI investing, showing that the highest returns in the first half came not from Nvidia (which the firm does hold) but from the components that feed Nvidia’s systems and are in shortest supply.
- The broader hedge fund landscape is producing dramatic AI-driven dispersion: Whale Rock Capital’s flagship fund rose 72.5% in H1 2026, Coatue Management returned 24.5%, and Analog Century’s 61% puts it among the top-performing tech-focused funds in the industry — while the average hedge fund returned a fraction of these figures, illustrating how concentrated the alpha has been in investors who correctly identified memory and infrastructure as the AI trade’s critical bottleneck.
What Happened?
Val Zlatev’s Analog Century Management — a New York-based hedge fund focused on technology, media, and telecom with a particular emphasis on semiconductors and hardware — delivered a record 61% return in the first half of 2026 for its Long/Short strategy. Investor documents reviewed by Bloomberg show the fund gained 9.4% in June alone, with year-to-date performance driven by memory chipmakers (Sandisk, Micron), storage companies, and networking equipment makers rather than the more crowded AI platform trades. Zlatev, who holds a physics doctorate and previously covered semiconductors at Kingdon Capital before founding Analog Century in 2018, has articulated a thesis centered on memory as the binding constraint in the AI buildout: context windows require enormous working memory, data-center scaling requires storage, and neither the memory nor the networking layer was adequately prepared for the speed of AI demand growth. Millennium Management, which provided about $1 billion to the firm in 2024, has one of the eight separately managed accounts in the firm’s Market Neutral strategy.
Why It Matters?
Analog Century’s returns validate a specific and counterintuitive version of the AI investment thesis: the highest returns in H1 2026 came not from the most visible and widely-owned AI names, but from the memory and infrastructure layer that most investors underweighted because it looked like a commodity business. Sandisk’s 858% return and Micron’s 304% return demonstrate the extent to which memory pricing power was underestimated — a function of both structural under-investment in memory capacity relative to AI demand growth and the physics of context window scaling (larger AI models require exponentially more high-bandwidth memory per inference). For institutional investors benchmarked to passive indices, the lesson is that the AI trade’s highest-alpha pocket has been in components and infrastructure, not in the hyperscalers or model developers whose AI revenues were already partially anticipated by the market.
What’s Next?
The key question for memory-focused AI investors is whether Zlatev’s thesis has further runway. Memory prices have already risen four to five times; the next leg of the trade depends on whether context window scaling continues to expand faster than memory supply can respond — a question that hinges on AI model architecture trends and the capex timelines of SK Hynix (which today began its $28 billion US listing to fund HBM expansion), Micron, and Samsung. The Nvidia Kyber NVL144 delay reported Monday could paradoxically benefit the memory thesis by extending the period of supply-demand imbalance in high-bandwidth memory. For investors considering entering now, the key risk is that memory is a historically cyclical business — and the same pricing power that drove Sandisk and Micron to record gains could reverse sharply if capacity additions outpace AI demand growth in 2027 or 2028.
Source: Bloomberg
