Why Cisco is interested in AI success story CoreWeave

Cisco is considering an investment in CoreWeave. In doing so, this “AI Hyperscaler” is said to be estimated at a market value of 23 billion. What explains Cisco’s interest?

Bloomberg sources report that a deal is nearing completion between the two parties. It is not reported to be a takeover attempt, but Cisco could well push for such a move in the long run. First things first: a lucrative investment in CoreWeave looms, which can thank the AI hype for its meteoric rise.

CoreWeave CEO Michael Intrator described in April of last year how the “rocket fuel” of the ChatGPT hype has added to the AI infrastructure player’s growth. At the time, Intrator described his sector as niche, but that would be a bit too modest these days. CoreWeave’s strong position did not happen overnight: back in 2019, it already began providing cloud infrastructure based on GPUs.

Million dollar AI strategies packed in this free 3 hour AI Masterclass – designed for founders & professionals. Act fast because It’s free only for the first 100.

Join it here for $0. 🎁

Born into crypto, but more than that

Like many a respectable tech company, CoreWeave began in a garage, albeit a bit more recently than the likes of Google, Apple and HP. In 2017, however, the focus was completely different than it is now, back when it was operating under the original name Atlantic Crypto. Catching on to the first major crypto hype, the company was conceived to launch a GPU-for-rent service to let others mine for cryptocurrencies, mainly Bitcoin. When profits from cryptomining plummeted, a strategic turn was needed. In late 2021, the name CoreWeave was born and a different direction was chosen.

The company’s focus was ahead of its time, knowing what we know now about today’s IT requirements. GPUs, particularly those from Nvidia, are considered the go-to tools to train generative AI models. However, due to scale contraints one often needs a service provider, as buying Nvidia chips yourself is a fast track to bankruptcy for most companies. CoreWeave therefore provides a GPU rental service that it claims is 35 times faster and 80 percent cheaper than like-for-like services from legacy providers. The company’s focus is on North America, but an investment in European expansion followed in June of this year.

Meanwhile, diversification has also seeped into CoreWeave’s offerings. GPUs still rule the roost, but CPU compute is also on offer with Intel Xeons and AMD Epyc chips. As a “native cloud” for Kubernetes, deployment of that technology is also relatively easy. Still, Nvidia’s hardware is CoreWeave’s biggest play. Many parties can call themselves Nvidia partners, but CoreWeave was the first at the Elite level for both Compute and Visualization, demonstrating they’re a cut above the common company.

Nvidia is already an investor. Run:ai, which is being acquired by Nvidia for 700 million, recently partnered with CoreWeave, too, making ties even closer in the future.

Cisco AI deployment takes more shape

We wrote back in June about Cisco’s concrete plans for an integrated AI infrastructure. A turn-key AI stack has been set up with Nvidia and VAST Data. These are known as Cisco Nexus HyperFabric AI clusters. The network is from Cisco, the GPUs from Nvidia and the unified data platform from VAST.

CoreWeave’s infrastructure would be a logical extension to this setup. While AI training for smaller models can be done on an on-prem basis, LLMs with tens of billions of parameters or more require heavier artillery. Training such models is something large players tend to do within their own corporate walls, since they have the capital to buy Nvidia chips en masse.

However, there aren’t that many of them. With Microsoft, Google, Amazon, Oracle, Meta and OpenAI, that’s effectively it. This is also reflected in Nvidia’s figures: almost half of its mammoth revenue is derived from four customers. The “democratization of access to GPUs” as CoreWeave CEO Michael Intrator intends to realize, therefore, does not so much apply to organizations with a few IT professionals, but to all large companies that aren’t one of those lucky few. A Boeing, Rolls-Royce or Siemens may well also wish to build or fine-tune their own models, after which considerable resources are required to deploy such a model as well. But where do you run those? And with whose stack? The CoreWeave-Cisco combination, with on-prem integrations where possible, would possibly be a lot more economical than leaving it to a hyperscaler.

A motivation like this shows where all of Cisco’s recent innovations are focused. For example, it is agitating for an entirely new security architecture in the form of Cisco Hypershield, which, by the way, also caught on to the fact that GPUs are dominating the modern IT stack over the CPUs of yesteryear. On its own, Hypershield is already a logical innovation intended for both on-prem and public cloud, but with a possible investment in CoreWeave, it makes all the more sense. Thus, Cisco is steadily forming a full AI stack for anyone who does not want or cannot build the infrastructure themselves.

Million dollar AI strategies packed in this free 3 hour AI Masterclass – designed for founders & professionals. Act fast because It’s free only for the first 100.

Join it here for $0. 🎁