SanDisk (SNDK)
The independent flash storage leader powering AI data pipelines with high-speed NAND for training data ingestion, checkpointing, and inference caching.
1. Core Product / Service
SanDisk (NASDAQ: SNDK) is a standalone flash memory and data storage company, spun out from Western Digital (WDC) in early 2025 as a pure-play NAND flash business. The company designs, manufactures, and sells NAND flash memory products used across consumer, enterprise, and data center markets. In the AI context, SanDisk's relevance centers on:
- Enterprise NVMe SSDs: Ultra-high-speed solid-state drives (Ultrastar DC SN series) used in AI training clusters for data ingestion, model checkpointing, and intermediate data caching. AI training pipelines read massive datasets (100s of TBs) repeatedly — flash storage reduces I/O bottlenecks versus spinning HDDs.
- Data Center Storage Platforms: Rack-scale NVMe all-flash arrays that provide high-throughput, low-latency storage for AI data lakes and training datasets. These platforms are critical for the data ingestion stage of the AI token supply chain.
- Consumer/Prosumer Flash: Extreme Pro and other high-performance flash cards used in edge AI applications, robotics, drones, and video production for AI training data capture.
- PCIe Gen5/Gen6 SSDs: Next-generation storage delivering 14-28 GB/s sequential reads, essential for training large language models where GPU utilization is gated by data throughput.
SanDisk was created through the separation of Western Digital's NAND flash and SSD business from the HDD (hard disk drive) business. The standalone SanDisk retains Western Digital's former flash IP portfolio, manufacturing partnerships with Kioxia (Japan), and the all-important "SanDisk" consumer brand.
2. Target Users & Pain Points
Primary users:
- AI hyperscalers operating massive training clusters (NVIDIA DGX SuperPODs, GPU farms) needing high-throughput storage for training data ingestion
- Enterprise AI/ML teams building on-premise or hybrid AI training infrastructure
- Cloud providers (aws, microsoft-azure, google-cloud) offering AI storage services backed by flash
- Edge AI engineers deploying inference at the edge (robotics, autonomous vehicles, drones)
- AI inference server operators needing high-speed caching for model weights and KV caches
Pain points solved:
- I/O bottleneck: GPU utilization in AI training is often gated by storage I/O — GPUs can process data faster than HDDs can feed it. Flash storage eliminates this bottleneck, improving GPU utilization from 60% to 95%+
- Model checkpoint latency: Training runs lasting weeks require frequent checkpointing (saving model state). Flash storage reduces checkpoint write time from hours to minutes, minimizing training interruption risk
- Data shuffle overhead: AI training data needs to be shuffled, batched, and re-read multiple times per epoch. Flash provides consistent low-latency random reads that HDDs cannot match
- Inference caching: LLM inference benefits from KV-cache offloading to fast flash storage, reducing GPU memory pressure
3. Competitive Landscape
| Company | Core Product | AI Storage Role | Key Differentiator |
|---|---|---|---|
| SanDisk (SNDK) | NAND flash, NVMe SSDs | AI data pipeline (ingest, cache, checkpoint) | Brand strength, Kioxia JV manufacturing |
| Micron (MU) | NAND + DRAM + HBM | AI memory + storage full stack | Integrated DRAM + NAND synergy |
| Samsung (SSNLF) | NAND, SSDs, HBM, DRAM | Broadest AI memory/storage portfolio | Vertical integration from fab to SSD |
| SK Hynix | HBM, NAND, DRAM | HBM dominance, secondary NAND | HBM market leader (60-70% share) |
| Kioxia | NAND flash | NAND manufacturing partner | BiCS Flash technology, SanDisk JV |
| Pure Storage | All-flash arrays | Enterprise AI storage platform | Software-defined, direct NVMe over Fabrics |
SanDisk's competitive edge lies in its brand recognition in consumer/prosumer flash markets combined with deep enterprise NVMe SSD engineering inherited from Western Digital's data center business. The spin-off allows SanDisk to focus purely on flash without HDD legacy dragging growth.
4. Unique Observations
Flash storage is the unsung hero of AI training throughput. Much of the AI narrative focuses on GPU compute (NVIDIA), memory bandwidth (HBM/SK Hynix), and networking (InfiniBand/Mellanox). But in practice, a significant fraction of GPU training time is spent waiting for data I/O. Moving training data from HDD to NVMe flash can improve effective GPU utilization by 30-50% — the equivalent of getting 30-50% more compute out of the same GPU fleet.
In the AI token supply chain, SanDisk operates at the data movement layer:
- Data Ingestion: Raw training data (text, images, video) stored on flash arrays for high-speed streaming into GPU memory
- Checkpoint Storage: Model state (weights, optimizer state, gradients) written to flash during training runs
- Inference Cache: Model weights and KV caches loaded from flash during inference serving
- Edge Data Capture: Training data generated at the edge (robotics, autonomous vehicles) captured on high-endurance flash cards
The Aschenbrenner 13F angle is critical: SanDisk is Aschenbrenner's largest combined bullish position ($1.113B total, 8.14% of portfolio), structured as a LONG $724.36M (5.30% of book, +ADD 8.15%) + CALL $388.76M (2.84% of book, 365,700 call contracts). This is a sophisticated stock+call structure — a bullish bet on storage for AI. Most AI infrastructure investors chase GPU stocks (NVDA) or cloud providers (MSFT, AMZN). Aschenbrenner's SanDisk position suggests a thesis that:
- AI training data volumes will grow faster than GPU compute capacity (more training data per unit of compute)
- The storage bottleneck in AI infrastructure will drive flash revenue growth
- The WDC spin-off unlocks value by creating a pure-play flash company
Key risk: NAND flash is a commodity market with significant price cyclicality. SanDisk's revenue can swing violently with NAND pricing cycles, independent of AI demand. The stock+call structure hedges some downside, but if NAND overcapacity emerges alongside AI storage demand, margin compression could offset volume growth.
5. Financials / Funding
13F Position (Q1 2026) — Aschenbrenner Fund
- LONG: $724.36M — 5.30% of book — +ADD 8.15% (increased from Q4 2025)
- CALL: $388.76M — 2.84% of book — 365,700 call option contracts
- Total SNDK exposure: ~$1.113B — second-largest bullish bet in the portfolio (8.14% combined)
- Stock price at Mar 31, 2026: $635.34
- Conviction: HIGH — the largest single-stock bullish positioning in the portfolio
- Thesis: AI-driven flash demand exceeding market expectations; structural storage bottleneck in AI training pipelines
Company Financials
- Spin-off date: Completed separation from Western Digital (WDC) in early 2025
- Revenue (Q3 FY2026): $5.95B, EPS $23.41 — demonstrating strong demand from AI data ingestion workloads
- Revenue (FY2025 est.): ~$14-16B, with ~30-40% from enterprise/data center markets and growing share from AI storage
- Market cap: ~$30-35B (May 2026)
- EBITDA margin: ~25-30% in enterprise SSD segment; consumer flash margin more volatile
- Debt: Carries moderate debt from spin-off (~$2-3B), investment-grade rated
- Manufacturing: Joint venture with Kioxia (Yokkaichi and Kitakami fabs in Japan) for NAND wafer production; uses BiCS 8th-gen 3D NAND (218-layer) and developing BiCS 9th-gen
- Key financial metric to watch: Enterprise SSD revenue growth rate and NAND ASP (average selling price) trends
Sources: Western Digital earnings releases pre-spin-off; SanDisk investor materials at https://www.westerndigital.com/ (2026-05-21); SanDisk brand site at https://www.sandisk.com/ (2026-05-21); SEC 13F Q1 2026 (Accession 0002045724-26-000008); DaveManuel.com 13F analysis 2026-05-19; EliteCurrenSea 13F update May 2026.
6. People & Relationships
- David Goeckeler — CEO (formerly CEO of Western Digital, led the spin-off). Previously EVP at Cisco driving networking/security.
- Wissam Jabre — CFO. Formerly finance lead at Western Digital, managed the balance sheet separation.
- Kioxia joint venture — Critical manufacturing relationship. SanDisk and Kioxia jointly operate NAND fabs in Japan (Yokkaichi, Kitakami). The JV structure provides manufacturing scale without full capex burden.
- NVIDIA partnership — SanDisk SSDs are qualified for NVIDIA DGX SuperPOD and DGX H100/B200 reference architectures. Flash storage certified for AI training pipelines.
- AWS / Azure / GCP — SanDisk (formerly WD) SSDs are used in hyperscaler storage infrastructure for AI workloads.
- Key investor note: Aschenbrenner holds SNDK Stock+Call as the largest combined position ($1.113B, 8.14%: LONG $724.36M + CALL $388.76M), signaling conviction that AI-driven flash demand will exceed market expectations.