A hardware shortage dubbed "RAMmageddon" is disrupting scientific research worldwide, according to a March 13, 2026 report by Nature journalist Heidi Ledford. The crisis traces directly to AI's explosive growth: as demand for large model training has surged, chip manufacturers have redirected production capacity toward the high-bandwidth memory required by AI systems, starving the market for standard RAM. Prices for conventional chips roughly tripled over the course of 2025, and according to HP, memory now accounts for more than one-third of the cost of building a PC — up from approximately 15% just months prior. Technology consultant Tom Coughlin estimates it will take <a href="/news/2026-03-14-tsmc-n3-wafer-shortage-ai-compute-2026">manufacturers 18 months or more to meaningfully expand supply</a>, placing the end of the shortage well into 2027.
The shortage is widening an already uneven playing field in global research. Well-funded institutions in wealthier countries can largely absorb the higher costs, but resource-constrained labs — particularly in lower-income countries — are being forced to make hard trade-offs. Pravallika Sree Rayanoothala, a plant pathologist at Centurion University of Technology and Management in India, told Nature her team had to narrow the scope of a crop disease forecasting project and resort to chunking data into smaller batches for separate modelling runs. The workaround is functional but costly in time and money. "The project timeline is increasing, and operational expenses are increasing," she said. "Slow model development is delaying tools for early disease prediction."
Abejide Ade-Ibijola, founder of GRIT Lab Africa in Johannesburg, said his institute has been insulated by industry funding, but described a broader pattern among African researchers who lack such support: traveling to universities in wealthier countries to access compute, completing their analysis, and returning home with nothing but a PDF of results. "It's polarized," he said. Matteo Rinaldi, director of the Institute for NanoSystems Innovation at Northeastern University, noted that modern scientific workloads routinely demand substantial memory capacity, making the shortage a direct structural barrier to research. Rayanoothala put it plainly: without a resolution, entire categories of data-intensive science will remain out of reach for labs that can't afford to wait.