April 24, 2024

Insights from AI innovators: Data is driving infrastructure – but the data needs to be better


It was all AI all the time this week, as so many weeks these days seem to be.

Supercloud 6: AI Innovators, SiliconANGLE’s and theCUBE’s signature event series, surfaced a lot of insights about where AI and generative AI are heading, and one upshot was that data is increasingly driving infrastructure architecture. But it’s not yet clear what that architecture will look like and especially who will build it. Check out all the insights below.

Meantime, in the usual yin-yang of late, investors keep piling more money into AI startups — and even established companies are benefiting, as Oracle, UiPath and Foxconn all beat estimates thanks to AI spending — but meantime, governments seem determined to rein it all in.

In fact, they seem inclined to rein in a wide variety of tech companies, led by the House’s bill that could lead to TikTok getting sold or even banned — though the Senate seems less enamored of the idea.

One more thing: Despite some signs of uncertainty lately, some companies in the other perennially hot market — cybersecurity — some companies are managing to keep pulling ahead, including CrowdStrike, Wiz and Zscaler.

Late-breaking Friday: It looks like HashiCorp may be up for sale, Bloomberg reported. More below.

Some of this and other news are topics of discussion in another installment of John Furrier’s and Dave Vellante’s theCUBE Pod, available now on YouTube. And catch Vellante’s weekly deep tech dive, Breaking Analysis, coming out this weekend.

Here’s the top news this week:

The lowdown on AI and data

The latest in SiliconANGLE’s and theCUBE’s editorial event series, Supercloud 6: AI Innovators, highlighted a wide range of startups at the leading edge of AI and data management, along with companies such as Uber and Walmart that are deep into implementing the technologies. A few takeaways:

  • 2024 is the year of implementing the experiments of 2023. “They’re starting to think about, OK, well, if I’m using this model as part of my application, what is the long-term cost going to be of this model? And how am I going to deal with that from a budget perspective?” said Suraj Patel, head of MongoDB Ventures. Companies such as Uber and Walmart are showing the way.
  • The current resource-hogging generative AI technologies could run into a wall, prompting many companies to look for more efficient compute. Some are on the bleeding edge: See the “food for thought” item below. But others are already well underway on a number of fronts, such as Multiverse Computing, which is using quantum computing-inspired software for AI model training, as well as Efficient Compute, Neural Magic, Taalas and, not least, Groq, the current hot thing in chatbots thanks to its ultrafast language-optimized custom processor.
  • Democratization of AI: It’s speeding up as organizations such as Cloud Native Computing Foundation help integrate it into cloud-native environments and companies such as Uber and Walmart make AI tools more available inside their organizations. “You’ll see the CNCF community probably take charge of the generative AI, because as we’ve been saying, the democratization of the data side is not going to be by the data scientist. It’s going to be by the AI itself,” said Furrier. The result, said Alessya Visnjic, co-founder and chief executive officer of WhyLabs Inc.: “The power of creating AI applications is now in the hands of any developer.”
  • Data is driving infrastructure architecture in the AI Age, since it’s too costly to move data at the edge to the cloud for AI processing, so AI compute will be coming to the data instead. That has huge implications for how networks need to be designed, in particular driving more AI, especially inference, into on-premises data centers. “Gen AI applications created a new category of applications that required a new data architecture,” said Venkat Venkataramani, co-founder and CEO at Rockset. Vellante noted Meta has suggested half of Llama deployments will be on-prem, illustrating how AI could lift the fortunes of a lot of infrastructure companies. When it comes to AI going forward, Furrier added, “Infrastructure is where it’s at.”
  • More than ever, you can’t have good AI without good data, so there’s lots more focus on ensuring that — a topic we covered in our recent and ongoing special report, AI’s Next Frontier: Data. For example, says CalypsoAI CEO Neil Serebryany, “We actually have a feature that allows enterprises to, as they get responses from generative AI models, verify if a response is true or not true and build their own internal taxonomy of trustworthy information on the generative AI side of the house.” A corollary: Governance is all the more important to avoid privacy and corporate data protection issues.
  • To support that trend, new data architectures will be needed — something we’ll cover in much more detail at the next Supercloud event in July. But the essence, Vellante says, is that there will need to be unified data sources for AI applications to work well, not just the stovepipes that exist today in multiple databases, data warehouses and data lakes. “Everyone’s trying to build better data structures… that can be ready for AI applications,” said Kyle Weller, director of product at Onehouse.ai.
  • You’ve heard of shadow IT? “Shadow AI is going to become more and more menacing in organizations,” said Zscaler Chief Security Officer Deepen Desai, meaning unsanctioned AI models. “Security professionals will have to leverage AI to fight AI"