SanDisk’s position in the AI infrastructure supply chain has been turbocharged by the exponential growth of inference workloads. Data center operators are expanding model context windows from 2,048 to 32,000 tokens, a 15× increase that multiplies memory requirements per inference instance. Industry forecasts now project NAND flash demand from AI use cases to grow at a 25% compound annual rate over the next three years, compared with just 8% for traditional storage applications. This structural shift has fueled a three-fold increase in enterprise SSD billings at SanDisk since early 2025, validating the firm’s strategy to prioritize data-center optimized flash products over commodity consumer drives.

The recent ramp of SanDisk’s proprietary BiCS8 232-layer process marks a critical inflection point. Internal benchmarks show BiCS8-based SSDs delivering 45% higher write endurance and 30% faster random read I/O than the prior generation. By collaborating with SK Hynix on co-development of high-bandwidth flash modules, SanDisk expects production capacity for enterprise products using BiCS8 to double by the end of 2026. As data centers iterate on higher compute-to-storage ratios, these enhancements translate directly into lower total cost of ownership, giving SanDisk a competitive edge over peer flash suppliers that remain on 176- or 192-layer nodes.

Since its spin-off in February 2025, SanDisk has outpaced all memory peers with a cumulative revenue increase of 65% and a gross margin expansion of 550 basis points. In Q3 2025, enterprise flash sales accounted for 62% of total revenues, up from 38% a year earlier. Management guided Q4 revenues to grow by 18% sequentially, driven by ramping data-center deployments and an expected 40% quarter-over-quarter increase in module shipments to cloud hyperscalers. With capital expenditures focused on topping up 300,000 wafer starts per month by mid-2026, investors should anticipate sustained double-digit top-line growth through at least fiscal 2027.