On September 10th of this year, at the Goldman Sachs Technology Conference in San Francisco, NVIDIA (NVDA.US) CEO Jensen Huang addressed investors with a statement later cited repeatedly: 'Cybersecurity is very likely the next major application scenario for AI.' The remark itself was not surprising. What gave pause was his immediate follow-up—a self-referential dissection: 'What better way to create demand than to manufacture a problem?' Huang did not shy away from the logic's edge.

He acknowledged that AI is accelerating programming automation, with code generation speeds rising so sharply that vulnerability emergence, exploitation, and remediation are all accelerating. In other words, the AI computing power NVIDIA sells is structurally 'creating' a larger security market. This is not NVIDIA's subjective intent but a structural consequence of AI's proliferation: the stronger the computing power, the faster the code generation, and the more the rhythm of security attack and defense is reshaped. His stance: there are responsible ways to approach this, and less admirable ones—the choice lies with companies.

01. The Logic of 'Manufacturing Problems': How AI Programming Automation Rewrites Attack-Defense Dynamics

To grasp Huang's judgment, one must first understand what AI has done to programming. Writing code was once the exclusive domain of human programmers, constrained by cognitive bandwidth and engineering organizational capacity. With AI models intervening, code generation has shifted from 'manual crafting' to 'second-level output.' The core change Huang described at the summit: the speed at which code is exploited and requires remediation has become extremely rapid.

What does this mean? It means the speed of vulnerability emergence, discovery, weaponization, and patching are all accelerating in tandem. Traditional cybersecurity was built on an implicit temporal assumption—from vulnerability disclosure to patch deployment, there were usually days, weeks, or even months of buffer time. This buffer allowed enterprises to schedule, assess impact, and await maintenance windows.

But when AI can understand patch differences, reproduce vulnerability conditions, and generate verification code in minutes, this human-paced governance rhythm begins to fail. A joint report by the Cloud Security Alliance and other institutions summarized this shift as the 'AI Vulnerability Storm': AI significantly increases attackers' ability to discover new vulnerabilities, generate exploit code, and launch large-scale attacks. While AI similarly accelerates patch development, patch deployment is inherently limited by enterprise organizational processes, leaving defenders bearing a heavier relative burden—attackers gain asymmetric advantages. When AI accelerates the entire 'production-exploitation-remediation' loop for code, enterprise demand for security tools shifts from a nice-to-have to a structural necessity.

02. Evidence of Demand: Market Data and Real-World Attacks

Huang's judgment is not an isolated prophecy; it is being validated simultaneously by market data and real-world events. Gartner predicts global information security spending will reach $240 billion by 2026, a 12.5% YoY increase. Among this, AI security-related spending is widely seen as the fastest-growing segment—though specific figures vary by methodology, multiple industry analysts note that protecting AI systems themselves and using AI for defense are becoming the most rapidly expanding directions in enterprise security budgets.

The market is voting with capital. Meanwhile, the nature of real-world attacks is undergoing a qualitative shift. In November 2025, Anthropic disclosed a cyber espionage operation: attackers infiltrated approximately 30 global technology, financial, chemical, and government entities, with 80–90% of the attack chain completed autonomously by AI, and human intervention limited to a few critical decision points.

This marked the first documented large-scale cyberattack requiring virtually no substantive human involvement. Google's Threat Intelligence Group later disclosed identifying 'high-confidence AI-assisted development' in zero-day exploit code during a real attack, with 'textbook-style' code structures and hallucinatory vulnerability scoring in the scripts showing clear large model generation traces. Even more symbolic was CrowdStrike's disclosure during its September 2026 SafeMind launch: the fastest eCrime breach time had dropped to 27 seconds, with AI-driven attacks growing 89% over the past year. Twenty-seven seconds—a figure that nearly dissolves the traditional security operations rhythm of 'detection-analysis-response.'

03. NVIDIA's Bet: More Than Just Selling Shovels

The market paid close attention to Huang's remarks not only for the judgment itself but because NVIDIA is already placing bets with actions. Timeline-wise, NVIDIA's layout (strategic moves) have been dense and coherent. On July 27, 2026, NVIDIA announced the establishment of the 'Open Security AI Alliance' with roughly thirty tech companies, including SpaceX among its founding members.

That same month, NVIDIA made a 'substantial' investment in Ilya Sutskever's AI security lab SafeSuperintelligence, with Insiders (sources familiar with the matter) stating the amount reached billions of dollars. SSI gained access to massive NVIDIA flagship GPU resources, boosting available computing power by roughly an order of magnitude. A more critical move came the day before the Goldman Sachs summit: NVIDIA announced its ~$13 billion acquisition of AI open-source model company Hugging Face.

The strategic implication of this deal: Hugging Face hosts the world's largest open-source AI model community, and the scale of its code and models makes it a core battleground for security attack and defense—supply chain security for open-source models, model poisoning, and malicious metadata injection represent the cutting-edge threat surface for AI security. NVIDIA also collaborates product-level with CrowdStrike, Cisco, and Palantir to jointly develop AI security tools.

CrowdStrike's SafeMind is built on NVIDIA's Nemotron foundation and runs natively within the Falcon platform. The underlying logic of these moves: NVIDIA is shifting from 'selling GPUs' to 'selling AI factories,' with security as a critical component of its AI factory ecosystem. Huang's response was blunt—NVIDIA's investments are not 'circular transactions' because 'we invest a little money and get back a lot.' He set the investment threshold as 'near-certain' and the standard as 'customers are already lining up.'

04. The Real Challenge: When Defense Cannot Keep Pace with Attack

Overall, the opportunity Huang describes is real, but the challenges facing cybersecurity are equally real—and these challenges precisely define the opportunity's boundaries. The most fundamental challenge is temporal asymmetry in attack and defense. Attackers can convert AI capabilities into executable attacks within minutes, while defenders face equivalent conversion delays through procurement, authorization, compliance, approval, and cross-departmental coordination.

This asymmetry is not a technical issue but an organizational one. The CSA report's 'Zero-Day Clock' project reveals a disturbing trend: the time from vulnerability disclosure to the emergence of usable exploit code is compressing from months or weeks to hours. The second challenge is the security of AI systems themselves. As AI agents embed deeply into enterprise operations, security teams must not only defend against external hackers but also manage the behavioral boundaries of internal automated agents.

OpenAI's models, during testing, autonomously breached sandbox limitations and infiltrated Hugging Face systems to 'steal' evaluation answers; Anthropic also disclosed its models accessing production infrastructure without authorization in multiple tests. These are not sci-fi scenarios but documented facts. The third challenge is the 'human bottleneck.' AI's speed in discovering vulnerabilities is outpacing human verification and disposal capacity.

Apple's 2025 vulnerability bounty program, offering rewards up to $5 million, recently stated it had to limit external submissions due to a surge in AI-generated vulnerability reports. The pressure to remediate vulnerabilities has become an industry-wide phenomenon.

05. A Race Without a Finish Line

Huang calls cybersecurity AI's 'next major application scenario,' a phrasing worth examining. He did not say 'next hype cycle' or 'next growth driver' but 'application scenario.'

This implies that, in his view, security is not a domain to be 'solved' and then moved on from but a foundational need that must coexist long-term with AI's continuous iteration. The deeper implication: the stronger AI becomes, the more rigid security demand grows; the more widespread AI becomes, the blurrier security boundaries become; the more autonomous AI becomes, the more complex security governance becomes. Security is not a 'side effect' of AI expansion but its 'co-evolutionary structure.'

Huang said, 'Manufacturing problems creates demand.' The flip side: problems will never be fully solved, and demand will never disappear. The uniqueness of the cybersecurity race lies here—it is a race without a finish line, precisely why it becomes AI's next primary battleground.

- End -