NextFin News - Cybersecurity will likely be the next major use case of artificial intelligence, Nvidia Chief Executive Officer Jensen Huang said Thursday, in the clearest signal yet that the world's most valuable chipmaker is already looking past the current datacenter buildout for its next growth engine.
Huang made the remark at the Goldman Sachs Communacopia + Technology Conference in San Francisco, where Nvidia was scheduled to address investors at 8:50 a.m. Pacific Time on September 10, 2026. The comment reframes the AI investment thesis at a moment when the market is increasingly focused on one question: what comes after the hyperscaler spending wave that has driven Nvidia to a roughly $5.4 trillion market capitalization.
The Setting: From Chips to the Systems That Protect Them
For the better part of three years, the artificial-intelligence capital-expenditure story has been dominated by a single direction of travel: hyperscalers and large enterprises buying graphics processors to train and run models. Nvidia's revenue has ridden that wave, and the stock has climbed more than 31% over the past twelve months, outpacing the S&P 500's roughly 17% gain over the same period.
Huang's cybersecurity framing points investors toward the defensive side of computing. It is not an abstract observation. One week earlier, on September 2, Huang appeared on stage with CrowdStrike Chief Executive Officer George Kurtz at the cybersecurity vendor's Fal.Con 2026 conference in Las Vegas, where he called CrowdStrike Nvidia's "No. 1 cybersecurity partner" and "my No. 1 cybersecurity provider." The two executives used the appearance to unveil SafeMind, a system CrowdStrike describes as the industry's first complete agentic platform for cybersecurity, built with Nvidia's Nemotron open models.
"Our AI researchers love working with yours," Huang told Kurtz on the Fal.Con stage. "Together with our domain expertise — your decade and a half of security data that we can train on — we can create Nemotron, take a frontier model and make it essentially a super AGI that is incredibly good at cybersecurity. And then, you put it into a system called SafeMind, with red-teaming, blue-teaming, the two adversarially working against each other in a digital twin of NVIDIA and testing it against that. I mean, that's completely brilliant."
The sequence of events matters. A chipmaker's CEO does not typically spend his week endorsing a single security vendor's product unless he sees a structural link between that vendor's roadmap and his own company's demand. Huang's comments at the Goldman Sachs conference — aimed at the financial community rather than at developers — suggest Nvidia wants investors to think about cybersecurity not as a cost center for customers, but as a new, compute-hungry workload running on Nvidia silicon.
The Market-Size Math Behind the Claim
The scale Huang is gesturing toward is large enough to move a company of Nvidia's size. Global spending on cybersecurity products and services is forecast to exceed $520 billion annually by 2026, up from $260 billion in 2021, and to reach $1 trillion annually by 2031, according to the 2026 Cybersecurity Market Report from Cybersecurity Ventures. A 2024/2025 McKinsey study cited in that report estimates that AI is expanding a $2 trillion total addressable market for cybersecurity providers.
For context, Huang told an audience at Nvidia's GTC conference in March that he sees "through 2027 at least $1 trillion" in demand for next-generation AI chips. The cybersecurity market he is now highlighting is of a comparable order of magnitude — and it sits directly adjacent to the datacenter business that already powers Nvidia's earnings.
Not all of that spending will flow to accelerated computing. Much of it will go to software licenses, managed services, and headcount. But the portion that requires real-time inference on massive security datasets — the agentic systems that continuously probe, attack, and defend enterprise environments — is inherently compute-intensive. That is the slice Nvidia is positioning itself to capture, and it is the slice that grows fastest when attackers also weaponize AI.
Why Security Is Not Just Another AI Application
Cybersecurity differs from most other AI use cases in one crucial respect: it is an arms race with a built-in escalation loop. When a company deploys AI for customer service or document summarization, the compute requirement is largely a function of its own usage. When it deploys AI for defense, the compute requirement is a function of how hard adversaries are trying to break in — and adversaries are now deploying AI of their own.
This creates a self-reinforcing demand dynamic that most enterprise software categories lack. Better defensive AI forces attackers to build better offensive AI, which forces defenders to buy more compute, which forces attackers to scale again. The workload does not plateau the way a one-time migration or a static analytics pipeline does. It escalates.
SafeMind illustrates the mechanism. The system continuously attacks and deploys protections within a digital replica of an organization's environment, searching for gaps in close to real time. That is not a batch job that runs once a quarter. It is a persistent, always-on inference workload — the kind that keeps GPUs busy around the clock.
"This isn't a copilot that's baked into someone else's intelligence," Kurtz said of SafeMind at Fal.Con. "It is a frontier-class model built and trained by CrowdStrike — on our data — in partnership with Nvidia."
Cyclical or Structural: The Call That Determines the Investment Case
The most important question for investors is whether the AI-cybersecurity link is a cyclical spending wave or a structural shift. A cyclical call would imply a burst of security budgets over the next few quarters, followed by digestion and mean reversion — the pattern enterprise software has followed through every procurement cycle of the past two decades. A structural call would imply a permanent step-up in the compute intensity of security, with demand that compounds rather than reverts.
The evidence points to structural, for three reasons.
First, the threat environment itself has been permanently altered. AI has lowered the cost and increased the speed of attacks — phishing at scale, vulnerability discovery automated by agents, and social engineering generated in real time. Defenders cannot absorb that with headcount alone. They need automated, AI-driven defense, and automated defense needs compute. That relationship does not reverse when the next earnings cycle turns.
Second, security spending is migrating outside the traditional budget that has historically capped it. Nearly 15% of corporate cybersecurity spending now comes from outside the chief information security officer's office, and non-CISO cyber spending is expected to grow at a 24% annual rate over the next three years, according to the McKinsey study cited by Cybersecurity Ventures. When security budgets escape the CISO envelope — moving into business-unit, cloud, and AI-infrastructure budgets — they are no longer constrained by the historical security allocation. That is a regime change in how the money is approved, not a one-year procurement pull-forward.
Third, the defender's compute requirement is recurring rather than one-time. Training a model is a capital event; running an agentic defense system is an operating expense that grows with the threat level. Recurring, usage-based compute demand is precisely the revenue profile that has supported Nvidia's datacenter business through the AI boom.
The counter-argument is not trivial. Security budgets are still corporate budgets, and corporate budgets are cyclical. If a macro downturn hits in 2027, chief financial officers will scrutinize every software line item, and security — despite its mission-critical status — has historically been cut less than other IT categories but is not immune. A recession-driven compression of cybersecurity spending growth back toward the mid-single digits would prove the structural thesis wrong, at least on the near-term timing. That is the falsifying signal to watch: not whether AI improves security, but whether total security spending keeps growing through a downturn.
The Second-Order Effect the Market Has Not Fully Priced
The first-order reading of Huang's comment is straightforward: more AI security means more chips. The second-order effect is more consequential, and it is the one the market has not fully worked through.
If cybersecurity becomes a major AI workload, it changes the composition of AI demand, not just the size of it. Training clusters are lumpy — they are built in waves, then sit partially idle between model generations. Inference workloads, especially always-on security inference, are smooth and persistent. A shift toward persistent security inference would make Nvidia's demand profile less cyclical and more utility-like, which is exactly the kind of re-rating that supports a higher valuation multiple.
There is also a cross-asset implication. The same escalation loop that drives compute demand for defenders also drives it for attackers — and attackers do not buy Nvidia chips directly. They rent cloud capacity. That means a portion of the offensive side of the arms race still flows back to the hyperscalers, who are Nvidia's largest customers. In effect, the AI security arms race monetizes itself twice: once through the defender's direct GPU purchases and once through the attacker's cloud spend, which funds the hyperscalers' next round of Nvidia orders.
This is why the cybersecurity framing is more than a talking point. It is a description of a demand flywheel in which Nvidia sits at the center, collecting tolls from both sides of the conflict.
Market Reaction: Muted on the Day
Investors did not treat the remark as a near-term earnings catalyst. Nvidia's shares were little changed during the September 10 session, trading around $223 after closing at $225.73 on September 8. The stock's 52-week range sits between $164.27 and $236.54, and the company's year-to-date gain of roughly 20% still outpaces the S&P 500's roughly 11.5% advance.
The muted tape is consistent with the nature of the news. Huang offered a strategic framing, not a product launch, a revenue guide, or a new customer win. Markets reward the latter immediately and price the former gradually, as evidence accumulates. The cybersecurity thesis will be validated or undermined not by conference remarks but by the cadence of security-related GPU orders over the coming quarters.
Who Benefits, Who Is Exposed
If the structural thesis holds, the beneficiaries are clear. Cybersecurity vendors that build genuine agentic defense platforms — systems that run persistent inference workloads rather than bolt chatbot interfaces onto legacy tools — stand to gain budget share. Cloud providers that host those platforms capture the attacker-side of the arms race. And Nvidia, as the supplier of the underlying accelerated compute, collects the toll from both directions.
The exposed parties are the legacy security tools that cannot integrate AI agents without a full architectural rewrite, and the organizations that attempt to defend AI-era threats with headcount and rules-based systems. Huang's comment that the cybersecurity community's ability to work "transparently, working together cohesively" is "the asymmetric advantage the good guys have" is also a warning: vendors that stay outside the emerging security ecosystem risk being left out of the information-sharing networks that make AI defense effective.
What to Watch Next
Three signals will determine whether this thesis is structural or cyclical.
First, whether hyperscalers begin to break out AI-security spending in their capital-expenditure disclosures. If cybersecurity becomes a distinct, growing line item in cloud capex, the market will have a hard number to underwrite the thesis. Second, whether CrowdStrike's SafeMind and comparable agentic platforms gain measurable enterprise traction over the next two quarters — measured in deployment breadth, not press releases. Third, whether the share of cybersecurity spending outside the CISO office continues to climb toward the 24% annual growth rate the McKinsey study projects.
The outlook splits by time horizon. In the short term, the cybersecurity comment is a narrative driver, not an earnings driver — expect volatility to remain tied to datacenter order flow and export-policy headlines. Over the medium term, the key variable is whether security vendors can convert AI pilots into recurring inference revenue. Over the long term, the question is whether AI-driven threats permanently raise the compute intensity of defense, which would make cybersecurity a durable, compounding workload rather than a cyclical budget line.
The Bottom Line
Jensen Huang is not merely identifying a new market. He is describing a shift in the composition of AI demand — from the one-time compute spikes of model training toward the persistent, escalating workloads of automated defense. If he is right, the next era of AI will be built as much on compute for protection as it was on compute for creation.
The first era of AI was built on compute for creation; the next may be built on compute for protection. That is a structural claim, and it will stand or fall on one test: whether security spending keeps compounding when the rest of the corporate budget does not.
Explore more exclusive insights at nextfin.ai.
