Sandisk Details AI Storage Strategy, Signs $93.9 Billion Long-term Agreement, HBM Sample to Debut in 2027

BlockBeats News, August 14th. SanDisk stated in its latest Investor Day materials that AI data center construction is changing the demand structure of the flash industry. The company expects that by 2026, the Flash market size will reach approximately 1.2ZB, with data centers surpassing Edge as the largest source of demand; by 2027, the industry revenue TAM will leap from the historical cycle of about $600 billion to over $3 trillion, approaching $5 trillion.

The core of this narrative lies in AI inference. SanDisk believes that long contexts, RAG, multimodal, and intelligent agent applications will bring about KV Cache demand, elevating SSDs from low-value storage components to a strategic link in AI infrastructure. The company even refers to SSDs as "Token Battery," indicating the storage of completed computation results to reduce redundant calculations, lower energy consumption, and increase throughput.

In terms of business models, SanDisk emphasizes signing multi-year new business model agreements with major customers to mitigate the traditional cyclical fluctuations of the NAND industry. The company stated that there are currently 8 customers involved in such agreements, with an average contract term of over 4 years. By FY2027, approximately half of the bit supply will be included in this model, rising to about two-thirds by FY2028. The total contract value of the agreements calculated at floor price is $93.9 billion, with remaining obligations of $91.1 billion, and $16.5 billion financial guarantee support.

The financial targets are equally ambitious. SanDisk expects full-year revenue for FY2026 to reach $20 billion, a 175% year-on-year growth; Non-GAAP gross margin of 71.6%, with adjusted free cash flow of $8.7 billion. The long-term model provided by the company for FY2028 to FY2030 shows that revenue will maintain high double-digit growth, with a Non-GAAP gross margin of approximately 80%, operating margin of around 75%, and adjusted free cash flow margin of about 50%.

The next and further storyline is HBF High-Bandwidth Flash. SanDisk stated that HBF will address the "memory wall" issue in large-model inference, providing a capacity up to 8 to 16 times that of HBM under similar read bandwidth. The company expects the first HBF inference product samples to appear in 2027, allowing the market to continue reevaluating its long-term space with AI memory architecture changes.