Artificial intelligence companies are competing for dominance over the emerging AI stack, and it’s anyone’s game.

The AI market is maturing. SpaceX Corp. wants to raise $75 billion to go public, and Anthropic PBC filed for an initial public offering of stock this week — then claimed that human society isn’t ready for its models. A small contradiction.

The winner will be the first company that can establish itself as a control plane for AI, according to Dave Vellante (pictured, right), chief analyst at theCUBE Research, who described that layer as the “system of intelligence.”

“The harmonization, the context, the ontology, that’s what we call the system of intelligence,” Vellante said. “That ultimately represents this model of the enterprise. I think the thing that’s becoming more clear [is] … you can’t build one, that back end, without the front end because it’s a closed loop where the system of intelligence and the agents … learn from the reasoning traces of the human and then they feed that back to the system of engagement.”

On the latest episode of theCUBE Pod, Vellante and John Furrier (left), executive analyst at theCUBE Research, discussed Snowflake Inc.’s attempt to move itself up the AI stack at its annual summit. They also looked at takeaways from the Cisco Live conference and other industry breaking news.

Snowflake and Databricks rivalry takes a new turn

Snowflake Summit made it clear that Snowflake wants to be seen as more than a data cloud. Whether or not the company can become a control point for the AI stack, or the system of intelligence as Vellante puts it, remains to be seen, but the release of Snowflake CoWork suggests a roadmap.

Snowflake is not alone in its ambitions. Its traditional competitor, Databricks Inc., has Databricks Genie, which similarly enables users to converse with their data.

“Snowflake started out as a big lead,” Furrier said. “Databricks did very well with the neoclouds and all the large AI infrastructure businesses. So, different growth strategies as they both grew to escape velocity. They can both win because the rising tide is so big. There’s so much territory. I don’t think it’s going to be squabbling as much as who’s just got the more durable position.”

Both Furrier and Vellante believe that viewing the competition as Snowflake vs. Databricks is limiting. Model-makers, application and software-as-a-service companies are all trying to establish agentic platforms that allow users and builders to interact with their data and accomplish work.

“It’s not just Snowflake and Databricks,” Vellante said. “They are clearly competing against each other, but … they are now sort of competing with the likes of Google, Microsoft, OpenAI, Anthropic. They’re all trying to sort of converge on the front end, how you talk to your data, not only the business users, but also the developers. And then how that connects to the back end, that system of intelligence.”

New opportunities arise in the AI stack

Cisco Systems Inc. also attempted to declare a place for itself in the emerging AI ecosystem at its Cisco Live conference, where the company announced a unified cloud platform. The opportunity for Cisco is huge, Furrier believes. AI requires both hyperscale infrastructure environments and a distributed edge — Cisco could bridge those two markets.

“There’s no high-performance networking,” he said. “Nvidia owns the GPU category, but who owns networking? Nobody. So, Cisco could really nail this because they have optical. [AI] factories require … high-performance networking, optical interconnect, security, observability, data movement, operational control planes and lifecycle management. This is Cisco’s core competency.”

Nvidia has a flawed scale-up architecture for AI, which could present an opportunity for Cisco, according to Furrier. On the other hand, Nvidia is currently beating networking companies at their own game, Vellante pointed out.

“Nvidia claims … that it’s the largest networking company in the world in revenue terms,” he said. “Nvidia has surpassed Cisco, which is insane to think about in terms of networking revenue. How can that be? They do their own set of custom Ethernet design for AI.”

Amid all the AI-related upheaval, companies are starting to find out just how much models can cost. People have begun tokenmaxxing, or using more tokens than their budgets can support. The consequences could be severe.

“There is a massive problem,” Furrier said. “People are like, ‘Where did these $200 credit card bills come from?’ People are getting whacked. I do not think it’s sustainable with the market right now, given demand and the cost. This is what a bubble burst could look like if it’s not reined in. Token costs is a short-term issue. The worst issue is bad agents going rogue.”

Here’s the full episode of this week’s theCUBE Pod:

You can also listen to us on Apple Podcasts or on Spotify. And for those who prefer to watch, check out our YouTube playlist. Tune in now, and be part of the ongoing conversation.

Photo: SiliconANGLE