I’ve been watching NVIDIA’s GPU Technology Conference evolve for years. It started as a developer event for people who worked with graphics hardware. Then AI took over, and GTC became something closer to a de facto AI conference—the place where the industry takes stock of where compute and intelligence are heading. GTC 2026 continued that trend, but with a different undercurrent than previous years.
AI infrastructure is no longer hypothetical. Enterprises are building AI factories and deploying agentic workloads. That shift is creating a security problem that can’t be bolted on after the fact—and the number of security announcements coming out of GTC this year reflected that reality.
CrowdStrike was particularly active at GTC, with several announcements tied directly to NVIDIA’s ecosystem. Taken together, they say something about where enterprise security is going and how tightly it’s now connected to AI infrastructure decisions.
Adversaries Are Already Using AI
Before getting into the specific announcements, one data point from the EY news is worth calling out. In 2025, the average eCrime breakout time—the window between initial access and lateral movement—dropped to 29 minutes. The fastest observed breakout happened in 27 seconds. That’s the threat environment security teams are operating in right now.
Adversaries are using AI to move faster, evade detection, and run operations at a scale that once required far more human effort. Security teams that review alerts manually are racing the clock in a way they weren’t even two or three years ago.
At GTC, CrowdStrike and its partners argued that agentic AI is how security operations keep up. These are AI agents that handle high-volume investigative work on their own while analysts stay in control. Whether that holds up at scale is something the industry is still figuring out, but the benchmarks being published are at least pointing in the right direction.
What the MDR Numbers Actually Mean
CrowdStrike announced an expanded partnership with NVIDIA to advance Agentic Managed Detection and Response. The work uses NVIDIA’s Agent Toolkit—including Nemotron models and NeMo Data Designer—to build specialized security agents inside Falcon Complete Next-Gen MDR.
Early internal testing showed investigations running up to 5x faster and triage accuracy improving more than 3x for high-confidence benign classification. Fine-tuning the Nemotron Nano model hit 96% accuracy in generating investigation queries within Falcon LogScale. These are vendor-reported numbers, so some skepticism is fair—but the specifics are at least more useful than vague performance claims.
“Adversaries are already using AI to move faster and scale their operations,” said Daniel Bernard, Chief Business Officer at CrowdStrike. “The future of managed defense isn’t adding more analysts—it’s embedding AI agents directly into SOC operations to give analysts superpowers.”
David Burg, Global Group Head of Cyber and Data Resilience at Kroll, said the approach has real implications for their client work. “By accelerating investigations and sharpening triage accuracy, it enables our teams to deliver faster, high-quality outcomes for clients around the world,” Burg said.
EY Brings Agentic SOC Services to Market
EY announced at GTC that it selected the Falcon platform to power its Agentic SOC services. NVIDIA AI infrastructure runs the underlying stack. The combination targets enterprises that want agentic security operations but lack the internal capability to build it. EY brings the managed services layer. CrowdStrike provides the platform. NVIDIA supplies the compute.
“The SOC cannot operate at human speed when adversaries move at the speed of AI,” Bernard said. “The Agentic SOC represents a fundamental shift. It elevates defenders from alert handlers to orchestrators of intelligent agents. Those agents think, reason, and act at machine speed—always under security team control.”
Emmett Koen, Senior Director of Cybersecurity Operations and North America Regional CISO at Mondelēz, explained why this matters operationally. “The volume and speed of alerts make manual investigation impossible,” he said. “AI agents continuously analyze activity and surface what matters most. That lets our teams focus on higher-value response and decision-making.”
Testing Security Before AI Goes Into Production
One of the practical problems with enterprise AI adoption is that organizations are building AI factories faster than they’re figuring out how to secure them. CrowdStrike and World Wide Technology tried to address that gap with the launch of the Securing AI with CrowdStrike Lab inside WWT’s AI Proving Ground, which is built on NVIDIA AI factory infrastructure.
The lab gives enterprises a place to test and validate AI security controls before committing to production deployment. Misconfiguration, data exposure and prompt injection risks don’t disappear because you’re excited about what the GPU cluster can do. Having a validated environment to work through those issues before they show up in a breach report is a reasonable approach.
“Enterprises need more than theoretical AI. They need a place to prove it works before investing,” said Chris Konrad, Vice President, Global Cyber, WWT. “By integrating NVIDIA AI Enterprise infrastructure with the CrowdStrike Falcon platform inside our Advanced Technology Center, we’re giving customers a secure, validated path to evaluate and support AI at scale.”
Nebius and the Security-at-Build Question
CrowdStrike also announced a partnership with Nebius, an AI cloud provider built for high-performance AI workloads on dedicated NVIDIA hardware. The integration brings the Falcon platform into Nebius AI Cloud. Organizations can then extend their existing security policies and workflows to AI workloads without rebuilding their security setup from scratch.
This pattern kept coming up at GTC: security vendors partnering with AI cloud and infrastructure providers to put security in place when workloads spin up, not after something goes wrong. “AI companies don’t just need more GPUs—they need infrastructure that performs predictably at scale and fits into how their organizations already operate,” said Mark Boroditsky, Chief Revenue Officer at Nebius.
Getting from Procurement to Deployment
This one didn’t come from GTC, but it’s relevant context. In February, CrowdStrike and Microsoft announced that the Falcon platform is now available on Microsoft Marketplace, with purchases counting toward existing Azure Consumption Commitment balances.
The practical effect: organizations that have already committed cloud spend to Azure can apply some of that toward Falcon without a separate procurement process. It’s a purchasing convenience more than a technical development, but procurement friction is a real barrier to security deployment, especially in larger enterprises. “Adversaries don’t wait for budget cycles, and neither should security teams,” said George Kurtz, CEO and founder of CrowdStrike.
Jay McBain, Chief Analyst at Canalys, noted that cloud marketplaces are increasingly the primary route to market for enterprise software—and that aligning security purchasing with existing cloud spend reduces the gap between deciding to buy something and actually getting it deployed.
What GTC Is Becoming
GTC has always been a good bellwether for where enterprise technology is heading. This year, the signal is clear: AI infrastructure is moving into production, and security is now a first-order concern—not a compliance checkbox at the end.
I’ve covered enterprise security long enough to be skeptical of vendor announcements at conferences. A lot of what gets announced at events like GTC points in a direction without being immediately actionable. But the combination of specific performance benchmarks, named enterprise customers describing real operational challenges, and security built into AI infrastructure from the start suggests this isn’t just marketing. Organizations serious about AI adoption need to think about security at the architecture stage—and that conversation is clearly happening.