AI Is Insatiable
View original at spectrum.ieee.orgIEEE Spectrum - Technical Title: AI Is Insatiable Date: 2026-04-06 14:22 Source: https://spectrum.ieee.org/high-bandwidth-memory-shortage <img src="https://spectrum.ieee.org/media-library/robot-hand-catching-falling-computer-chips-from-an-open-snack-bag-in-pop-art-style.png?id=65425799&width=1200&height=800&coordinates…
What we drew from this source
The claims Via News extracted from this document. We point to the source; we don't replace it.
If any of the big three HBM companies—Micron, Samsung, and SK Hynix—say that they are adjusting the schedule of the arrival of new production, that'd be an important signal
60% confidenceStartups developing all sorts of products might pivot toward creative redesigns that use less memory
60% confidenceData centers might steer toward hardware that sacrifices some performance for less memory as adaptation to shortage
60% confidenceMakers of AI processors, notably Nvidia and AMD, are demanding more and more memory for each of their chips, driven by needs of firms like Google, Microsoft, OpenAI, and Anthropic
60% confidenceAI electricity consumption could account for up to 12 percent of all U.S. power by 2028
60% confidenceAI hyperscalers' ravenous appetite for memory is causing current DRAM shortage, particularly for high bandwidth memory (HBM), which is a major constraint on the speed at which large language models run
60% confidenceWater consumption for cooling AI data centers is predicted to double or even quadruple by 2028 compared to 2023
60% confidenceIf any of the big three HBM companies—Micron, Samsung, and SK Hynix—say that they are adjusting the schedule of the arrival of new production, that'd be an important signal
60% confidenceAI electricity consumption could account for up to 12 percent of all U.S. power by 2028
60% confidenceConstraints like shortages can lead to interesting technology solutions
60% confidenceData centers might steer toward hardware that sacrifices some performance for less memory, and startups might pivot toward creative redesigns that use less memory as constraints lead to interesting technology solutions
60% confidenceGenerative AI queries consumed 15 terawatt-hours in 2025 and are projected to consume 347 TWh by 2030
60% confidenceAI hyperscalers' ravenous appetite for memory is a major constraint on the speed at which large language models run
60% confidenceWater consumption for cooling AI data centers is predicted to double or even quadruple by 2028 compared to 2023
60% confidenceGenerative AI queries consumed 15 terawatt-hours in 2025 and are projected to consume 347 TWh by 2030
60% confidence