Why the NAND Business Intel Sold for $9 Billion Could Be Worth $150 Billion
Solidigm is reportedly weighing a US IPO at a valuation of up to $150 billion. No price is set and no listing is confirmed, but the number reflects a real shift: the AI infrastructure bottleneck is spreading from GPUs and HBM into storage.
Between $9 Billion and $150 Billion
| Metric | Value | Note |
|---|---|---|
| Intel NAND business acquisition price | $9B | 2020 deal |
| Potential IPO valuation (high end) | $150B | Not yet confirmed |
| Potential IPO raise (high end) | $15B | Early-stage review |
On the surface, that's a 16.7x jump. But it would be wrong to say "the business Intel sold for $9 billion is now worth $150 billion." The 2020 deal covered the NAND SSD business, NAND components and wafer operations, and a plant in Dalian, China, and the acquisition closed in stages. Since then, Solidigm has operated as an independent company, and its products, customers, technology and market conditions have all changed.
The $150 billion figure is the high end of an IPO review process reported by Reuters, citing multiple sources. Solidigm has held pitch meetings with investment banks to select underwriters, but SK hynix has said there is "no confirmed specific plan."
Why Storage, Why Now
Looking at AI infrastructure through GPUs alone makes storage easy to miss. But treat a data center as a factory, and the role becomes clear. Compute does the calculating. Memory feeds data to the compute layer at high speed. Storage holds data before and after computation. Network moves data among these three layers.
HBM — Ultra-fast working memory sitting right next to the GPU. Fastest, most expensive.
DRAM — Memory servers use immediately. Fast, high cost.
Enterprise SSD — AI datasets, vector databases, embeddings, cache, warm data. A balance of speed and capacity.
HDD — Long-term storage and cold data. Slow, low cost.
Solidigm is targeting the third layer. Its D5-P5336 offers up to 122.88TB of capacity and is built for read-heavy, large-scale data such as AI data lakes, machine learning and object storage. The company's pitch: high-density QLC SSDs let data centers pack the same storage capacity into fewer racks and less power.
Not Just a Story, Already Showing Up in the Numbers
This shift matters because it isn't purely a future hypothesis. According to TrendForce, combined revenue among the top five global enterprise SSD makers topped $9.9 billion in the fourth quarter of 2025, then rose to $18.46 billion in the first quarter of 2026 and $37.59 billion in the second quarter. In that second quarter, SK hynix Group, including Solidigm, ranked second with $8.63 billion in enterprise SSD revenue.
| Period | Top-5 Enterprise SSD Revenue | Key Driver |
|---|---|---|
| Q4 2025 | Over $10B | AI inference expansion, HDD shortage, high-capacity SSD demand |
| Q1 2026 | $18.46B | AI agents, expanded CSP procurement, sharp price gains |
| Q2 2026 | $37.59B | Higher shipments plus elevated contract prices, spread of GB-series AI servers |
Over the same period, TrendForce cited high-capacity QLC enterprise SSDs as the core strength for SK hynix and Solidigm. That's not the same as saying all of NAND is improving equally. NAND demand for smartphones and PCs remains weak, while enterprise SSDs for AI servers are stronger. The current memory cycle looks less like a broad boom and more like a selective supercycle that splits by product.
Why AI Agents Could Reshape Storage Demand
In training-heavy AI, massive training datasets and checkpoints were the center of storage demand. As inference and AI agents grow, the structure changes. Agents read corporate documents, remember past tasks, search customer records, store vectors and embeddings, and manage the KV cache of long conversations.
In technical material published in 2026, Solidigm laid out an approach that combines high-capacity and high-performance SSDs to push AI inference's KV cache down into the storage layer. Rather than keeping everything in expensive HBM or DRAM, frequently accessed data and bulk data get placed in different tiers.
Multimodal AI pushes in the same direction. Text is smaller than images; images are smaller than video. Move into physical AI, and cars and robots continuously generate camera feeds, video, telemetry and sensor data. Compute can be reused for the next task, but valuable data keeps accumulating. That gap is what creates long-term demand for storage.
Who Benefits, and What the Risks Are
The likeliest beneficiaries are SK hynix, which owns Solidigm; Samsung and Micron, which are aggressively gaining enterprise SSD share; and the broader NAND supply chain. The more AI servers adopt high-capacity QLC SSDs, the more valuable products become for makers that can pack more storage into the same data center footprint and power budget.
The biggest risk is supply. NAND bit supply can scale up faster than HBM through capacity additions and process conversions, and expansion by Chinese producers is a wildcard. Even if AI-driven enterprise SSD demand stays strong, weakness in consumer NAND and rising supply could still weigh on overall prices. Nor does SSD fully replace HDD everywhere. For long-term storage and extremely low-cost cold storage, HDD economics still hold up.
Insight Times Editorial Desk
