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 used to require significantly more human effort. Security teams reviewing alerts manually are working against the clock in a way they weren’t even two or three years ago.
At GTC, CrowdStrike and its partners were making the case that agentic AI—AI agents that handle high-volume investigative work autonomously while analysts maintain oversight—is how security operations keep up. Whether that argument holds in practice at scale is something the industry is still working out, but the benchmarks being published are at least pointing in a useful direction.
What the MDR Numbers Actually Mean
CrowdStrike announced an expanded collaboration with NVIDIA to advance Agentic Managed Detection and Response, using NVIDIA’s Agent Toolkit—including Nemotron models and NeMo Data Designer—to build specialized security agents inside Falcon Complete Next-Gen MDR.
According to CrowdStrike and NVIDIA, early internal testing showed up to 5x faster investigations and more than 3x higher triage accuracy in high-confidence benign classification. Fine-tuning the Nemotron Nano model achieved 96% accuracy in generating investigation queries within Falcon LogScale. These are vendor-reported benchmarks, so they warrant some skepticism, but the specificity is at least more useful than generic 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, noted that the approach has operational 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 has selected the Falcon platform to power its Agentic SOC services, with NVIDIA AI infrastructure running the underlying stack. The combination is aimed at enterprises that want to move toward agentic security operations but don’t have 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 are moving at the speed of AI,” Bernard said. “The Agentic SOC represents a fundamental shift, elevating defenders from alert handlers to orchestrators of intelligent agents that 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, offered some grounding on why this matters operationally: “The volume and speed of alerts make it impossible for analysts to manually investigate everything. AI agents that can continuously analyze activity and surface what matters most allow our teams to 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 specifically for high-performance AI workloads on dedicated NVIDIA hardware. The integration brings the Falcon platform into Nebius AI Cloud, so organizations can extend existing security policies and workflows to AI workloads without rebuilding their security architecture to fit a new environment.
This is a pattern showing up repeatedly at GTC: security vendors partnering with AI cloud and infrastructure providers to make sure the security layer is present when workloads spin up, not added later when 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 useful bellwether for where enterprise technology is heading. This year, the signal is pretty clear: AI infrastructure is moving into production, and security is now a first-order concern in that conversation—not a compliance checkbox at the end.
I’ve been covering enterprise security for long enough to be skeptical of vendor announcements at conferences. A lot of what gets announced at events like GTC is directional rather than immediately actionable. But the combination of specific performance benchmarks, named enterprise customers describing real operational challenges and security integrations baked into AI infrastructure from the start, suggests this isn’t just marketing. Organizations that are serious about AI adoption need to think about security at the architecture stage—and that conversation is clearly happening.
- The Agentic AI Reckoning Nobody Saw Coming - April 28, 2026
- Cato Networks Redraws the SASE Adoption Map - April 27, 2026
- What Agentic Pentesting Says about the Threat Landscape - April 21, 2026
