CEO: GenAI changes multi-cloud security, network equation

Generative AI presents enterprises with higher cloud costs than traditional workloads, complicating multi-cloud security and networking, according to one industry CEO.

Data gravity and the cost of GPU-based computing infrastructure that's required for generative AI (GenAI) mean many enterprises are choosing to move AI apps to various locations, including back on-premises, according to Doug Merritt, CEO at software-defined networking vendor Aviatrix. In Merritt's view, as told to TechTarget Editorial's Beth Pariseau in an episode of IT Ops Query Season 2: The State of SecOps, these increasingly distributed workloads call for a new approach to centralized network and multi-cloud security management.

"This is a shift probably as dramatic as [cloud computing itself], from the silicon layer all the way up," Merritt said. "It's changing networking, it's changing storage characteristics, and whenever that happens ... you wind up with fragmentation, if you really want to take advantage of GenAI effectively."

Aviatrix pitches its software-defined networking product to address that fragmentation with a centralized control plane where multi-cloud security policies can also be enforced. The company introduced its first security product, a distributed firewall for Kubernetes, in May, and this week rolled out a managed version of its multi-cloud network and security control plane. In the coming weeks, Aviatrix will roll out the first of its own GenAI-powered features to boost incident response and event reduction.

Doug Merritt

Generative AI still needs a lot of time to develop -- as many as 10 years, Merritt estimated. He said Aviatrix has also dealt with hallucinations and cost issues while training large language models, and it will take time before they don't need to be carefully vetted by humans.

"The dilemma that most of us find ourselves in is it's hard to manufacture enough skills, enough skilled people with enough of an edge and capability to keep up with what's occurring," he said. "Right now ... it's almost a mandate to begin to take advantage of [AI]."

Meanwhile, as the former CEO of Splunk, which expanded its log management products to include security observability over the last decade, Merritt said he's seen SecOps practices evolve from a subset of network management to its own discipline. This can create issues of friction and a potentially costly lack of communication among separate teams, so further collaboration would be ideal, but at least some of that separation will likely remain long-term, he said.

"With things like security operations, infrastructure IT operations, DevOps, network ops, there's always this desire to make it one [discipline]," Merritt said. "And then when I think about how I make sure that my [operations are] effective, it's really difficult to have it be one, because the security teams are going to have a whole different competency and passion and focus ... and it's hard to have that same competence as you would on infrastructure optimization."

Beth Pariseau, senior news writer for TechTarget Editorial, is an award-winning veteran of IT journalism covering DevOps. Have a tip? Email her or reach out @PariseauTT.