As we move through the second quarter of 2026, the Artificial Intelligence (AI) narrative in global markets is evolving from a simple quest for raw processing power to a deeper understanding of the infrastructure required to support increasingly complex models. In a recent interview with Bloomberg, Dan Chung, CEO and CIO of Alger, highlighted a critical shift: the demand for memory, specifically High Bandwidth Memory (HBM), has become "insatiable."

The Memory Wall and Hardware Evolution

For years, investor attention was focused almost exclusively on Nvidia and its GPU architectures. However, as Chung explained, processing power alone is no longer the sole determinant of performance. The so-called "memory wall" has emerged as the single largest obstacle in training Large Language Models (LLMs) and the multimodal systems that dominate the 2026 landscape. Memory is no longer just a storage component; it is the vital conduit that allows data to flow to processors at the speeds required for real-time inference and massive-scale training.

Alger, a firm with a long history of growth investing, sees a massive opportunity in the companies dominating the HBM sector, such as Micron, SK Hynix, and Samsung. The transition to the HBM4 standard, which began in earnest this year, has increased component costs but also significantly boosted margins for manufacturers, creating a new investment stronghold within the technology sector.

Alger’s Strategy and the Macroeconomic Environment

Dan Chung noted that capital expenditure (CapEx) from cloud giants—the Hyperscalers like Microsoft, Google, and Amazon—shows no signs of fatigue. On the contrary, investments in next-generation data centers are accelerating. Chung argues that the market is underestimating the duration of this investment cycle. While many analysts fear an "AI bubble," Alger focuses on the fundamentals: the demand for memory is a direct consequence of the need for efficiency, reduced latency, and higher throughput.

Against the backdrop of the relatively high interest rates that characterized the past two years, semiconductor companies have shown remarkable resilience. Memory, traditionally viewed as a cyclical commodity industry, is transforming into a high-spec, strategically critical sector. A company's ability to reliably produce HBM3E and HBM4 chips is now the "entry ticket" to participation in the global AI economy.

Geopolitics and the Supply Chain

Chung’s analysis extends beyond mere financials into the geopolitical reality of 2026. With the final phases of the U.S. CHIPS Act being implemented and similar initiatives maturing in Europe, the location of memory production has become a matter of national security. Alger is closely monitoring how Micron expands its facilities in Boise and New York, aiming to reduce dependence on East Asian supply chains, even as SK Hynix maintains its technological lead in HBM packaging.

Chung concludes that this "insatiable" demand is not just about the present, but about preparing for Artificial General Intelligence (AGI). Every step toward AGI requires an exponential increase in memory capacity and bandwidth. For investors, this suggests that the "golden age" of semiconductors may still be in its early chapters, with memory acting as the catalyst that will define the winners and losers of the next decade.

  • HBM4 memory has become the primary bottleneck for AI scaling in 2026.
  • Big Tech CapEx remains at record levels, defying bubble fears.
  • The focus is shifting from pure compute to data movement and throughput.
  • Geopolitical supply chain security is now a key driver of semiconductor valuations.