Research Note: Pure Storage diversifies further with investment in AI hyper-scaler CoreWeave

Pure Storage, Inc. has announced a strategic investment in CoreWeave, a start-up building AI-focused cloud infrastructure solutions. In parallel, Pure Storage hardware solutions will be available within the CoreWeave Cloud platform as the company looks to expand past traditional data centre sales.

Background

Pure Storage has announced an investment in the AI cloud infrastructure platform CoreWeave. CoreWeave, Inc. is a US company founded in 2017 as Atlantic Crypto, which initially focused on building infrastructure for the cryptocurrency market. The company pivoted in 2018/2019, becoming CoreWeave and renting out its inventory of GPUs.

As the AI boom has taken off, CoreWeave represents an avenue for all manner of IT organisations to evaluate the AI landscape using the same hourly cost model as other public cloud platforms. The opportunity for CoreWeave is to capitalise on the demand for AI solutions, building out entire ecosystems for end-to-end AI development.

Storage

Why would Pure Storage invest in another infrastructure company? The answer is fairly obvious but also validated by other solution providers in the market. AI model development (training) requires access to terabytes or potentially petabytes of data. However, the access profile is more demanding than typical workloads, with high volumes of throughput needed to keep GPUs active.

As GPU clusters scale, centralised storage, either as dedicated hardware or software, is essential to ensure high availability and reliable access to data from many compute nodes in a highly random fashion. The alternative is to build an infrastructure with local storage, such as NVMe drives, where the management overhead (and impact of failure) can adversely affect the time to train AI models.

Domino

At NetApp Insight in September 2024, we met with Domino Data Labs, a NetApp partner. The company uses NetApp hardware to provide customers with reliable access to data using its Enterprise MLOps platform. Predictable storage is a crucial feature of these infrastructure offerings, which typically means using centralised storage. We hope to have a podcast episode coming soon where we talk about the experiences of NetApp and Domino working together.

Investment

It’s easy to see why CoreWeave needs a reliable storage solution, but why Pure Storage? The key is perhaps in the existing widespread use of centralised storage (specifically FlashArray and FlashBlade) across existing enterprises.

In addition to reliability, data mobility is a significant challenge when managing large data sets across hybrid computing environments. This is a topic on which we’ve written extensively since the mid-2010s (see our dedicated Microsite on Data Mobility). Whether IT organisations are using on-premises, public cloud or specialist cloud platforms, moving data efficiently between locations is a challenging task. Content can become out of sync between locations, while the copying process itself can delay useful processing work by GPUs.

By using centralised storage, IT teams gain efficient replication and data movement capabilities, especially when those solutions are available in every location where the data and computing reside. Today, Pure Storage can achieve that goal with on-premises hardware, co-located hardware, virtual appliances in the public cloud (Cloud Block Store) and dedicated infrastructure in specialist clouds like CoreWeave. (Side note: NetApp can also achieve this through its integration into public cloud infrastructure.)

IPO

Rather than issue another funding round (the last one was May 2024), CoreWeave has sold existing shares to new investors, including Pure Storage, Cisco Investments and BlackRock. We view this move in two ways. Firstly, existing shareholders gain some return on their initial investment, while the new investors ratify their belief in the CoreWeave business model. In return, CoreWeave agrees to use Pure Storage hardware, with the benefits we’ve already highlighted. With serendipitous timing, Pure Storage has also announced the certification of FlashBlade//S500 with NVIDIA DGX SuperPOD.

For Pure Storage, there will be monetary value in the investment in the near future, especially as CoreWeave is expected to go public in 2025. The bigger prize, though, is seeding alternative locations outside the traditional data centre with Pure Storage hardware. Pure Storage CEO Charlie Giancarlo has repeatedly highlighted the company expects to sign a hyper-scale partner by “year end” (see this article, which discussed the topic). CoreWeave could be that partner, although the company hasn’t yet confirmed it either way.

The Architect’s View®

The investment in CoreWeave follows on from a previous transaction to invest in ceramic storage company Cerabyte. It is clear that the future of Pure Storage can’t be solely focused on the corporate data centre, while the growth in traditional public cloud will be challenging and a long process (if it happens at all). As such, other avenues are needed. The emergence of specialist cloud platforms is one market to gain additional sales. However, we believe the long-term strategy should be to make the Pure Storage operating system (Purity) ubiquitous wherever large volumes of data are stored and processed. The ease with which Pure Storage can make that process happen will be a measure of the future success of the company.