In the current situation where high-performance computing power and advanced storage are repeatedly emphasized by Microsoft and SK Hynix as “in short supply,” capital has not pressed the brakes; instead, it has continued to increase investment along an axis from the grassroots to application: on one end, Israeli AI security company Zenity has secured $125 million in a Series C funding round, directly targeting AI and AI Agent security and governance; on the other end, Horizon3 in San Francisco has completed $250 million in Series E funding, with its valuation being raised from approximately $650 million a year ago to over $2 billion, betting on automated security testing and attack-defense simulations; extending further into the physical world, software-driven mining startup Mariana Minerals has secured $310 million in Series B financing, led by Khosla Ventures with a16z participating, with AI being viewed as an operating system layer rewriting mining production processes. These three substantial and distinct financing deals, coupled with large tech companies' ongoing expansion of AI infrastructure capital expenditures and the real supply constraints described by CITIC Securities as “not yet showing a significant overall surplus,” constitute the first layer of tension in current AI investments; layered above this are the policy signals from the White House planning to convene multiple AI companies to discuss the “AI framework” on Tuesday, as well as a narrative shift proposed by Google DeepMind’s Chief Strategy Officer that “recursive self-improvement (RSI) may replace AGI, becoming a new target for certain markets,” along with the market response on August 3, 2026, during U.S. stock trading when Meta, Microsoft, Google A, and Oracle all saw their stock prices rise. Capital, policy, and technological expectations are forming a rare triple resonance, and the AI security and AI Agent security sectors represented by Zenity are precisely at the intersection of this resonance.
Zenity Secures $125 Million: Betting on AI Agent Security
At a time when computing power and model narratives have been pushed to the highest level, Zenity has cut into a seemingly “niche” yet actually closest to money's path—AI security and AI Agent security. This Israeli company focuses directly on the AI and AI Agent that enterprises are massively experimenting with: when Agents are granted the ability to automatically access business systems and data, they are no longer merely “smarter chatbots” but semi-automated employees holding the keys to production systems. Once attacked or misused, the potential losses from privilege abuse, sensitive data leaks, and unauthorized operations far exceed those from a string of traditional SaaS account password leaks. What Zenity aims to sell is securing this doorway before this batch of “digital employees” truly enters the core processes of enterprises.
The $125 million Series C financing has raised the valuation of this doorway to a new height. The leading investor Norwest Venture Partners, along with heavyweight institutions such as SoftBank Vision Fund 2, Hitachi Ventures, and LG Technology Ventures, is effectively issuing a priority tag for “AI-native security”: on the path where AI Agents expand into enterprise infrastructure, whoever can first provide a set of security foundations surrounding Agent permissions, access boundaries, and behavior governance will qualify to be included in future IT budgets alongside cloud vendors and model vendors. Zenity securing $125 million at this time does not merely mean acquiring additional funds, but rather being chosen by capital to answer an urgent question: as AI capabilities continue to enhance and deployment scopes expand, who can ensure that these self-service access-capable Agents do not stray from their paths or lose control, and provide enterprises with a visible, manageable, and accountable security order before any real incidents occur?
Horizon3 and Mariana: Dual-line Betting on AI
If Zenity is still addressing the question of “Will AI cause trouble?”, Horizon3 has pushed the question to a more traditional cyber attack-defense battlefield: as enterprises increasingly hand over more systems to models and Agents for decision-making, who will continuously and automatically test how vulnerable these systems themselves are? Headquartered in San Francisco, Horizon3 focuses on automated security testing and attack-defense simulations, recently completing $250 million in Series E funding, led jointly by NightDragon and NEA, with its latest valuation raised to over $2 billion, more than tripling from about $650 million a year ago. Capital expresses its attitude in an extremely direct way: in a world driven by AI, with an exponentially expanding attack surface, a security company capable of making penetration testing and attack-defense drills “perpetual and automated” should be priced as “AI infrastructure” rather than as traditional security tools.
Another clue boldly crosses the cloud and lands in one of the heaviest industries. Mariana Minerals claims to be a software-driven mining company, attempting to rewrite mining production processes through software and AI, and it has just secured $310 million in Series B financing, led by Khosla Ventures with a16z participating—this is not a typical resource stock list but an object of investment from top Silicon Valley funds. In the latest financial reports from Microsoft and SK Hynix, high-performance computing power and advanced storage remain the most constrained segments, with cloud vendors continuing to increase capital expenditures around GPUs and storage. Concurrently, security companies like Horizon3 and traditional resource enterprises like Mariana have simultaneously received substantial financial arrangements. The appearance of this “virtual and physical” combination in capital flow indicates that AI-related investments are overflowing from underlying computing power to security tools and real-world industry applications, as AI is no longer merely a hardware story within data centers but has become an operational lever that can be valued in real assets and production processes.
RSI Replacing AGI? AI Capital Rewriting the Narrative
In the past few years, AGI has more often resembled an “endgame myth”—once human-level intelligence is conquered, the world will be reshaped all at once. The RSI (recursive self-improvement) proposed by Google DeepMind's Chief Strategy Officer Jasjeet Sekhon shifts the focus from whether that moment will be “realized,” to how quickly improvements can occur in the process. RSI emphasizes that models can continuously optimize themselves with their own involvement, and the capability curve resembles a long-cycle, compound-interest inclined line. For capital, the AGI narrative corresponds to betting on the success or failure of a certain endpoint, while the RSI narrative corresponds to betting on the system gradually accelerating in the coming years, continuously consuming more computing power and data. Sekhon explicitly states that the unprecedented capital expenditure in AI is largely based on the bet that future systems possess this type of RSI capability; some market participants have already begun to see RSI as a new generation investment target and industry narrative capable of replacing AGI, although this is still distant from industry consensus. When combined with Microsoft and SK Hynix's recent financial reports indicating that “high-performance computing power and advanced storage continue to be in short supply,” this bet seems more like laying the groundwork in advance for a system that “will increasingly spend money” on a real supply constraint.
On top of this technological expectation, the feedback from policies and market prices also resonates within the same time frame. The White House plans to hold a meeting on Tuesday with multiple AI companies to discuss the governance framework for artificial intelligence. Although the list of participants and framework details have not been fully disclosed, this is viewed as a signal that regulators are beginning to set boundaries in advance for more powerful and harder-to-predict AI systems. Almost simultaneously, during U.S. stock trading on August 3, 2026, Meta rose over 6%, Microsoft nearly 5%, Google A about 3.7%, and Oracle around 4.6%, while major tech companies continue to increase their capital expenditures on AI infrastructure. The expectations of RSI in the technical narrative, the proactive layout of governance frameworks by policy levels, and the optimistic pricing of tech stocks regarding long-term investments are being interpreted together by the market: when AI is imagined as a system capable of recursive self-improvement, all seemingly excessive capital expenditure today can be packaged as a long-term option to buy into future compound interest curves in advance.
When Supply is Tight, Security Becomes a New Necessity
The story of computing power is not over. In their latest financial reports, Microsoft and SK Hynix remind that high-performance computing power and advanced storage are still in tight supply, with customers continuously placing additional orders, while they continue to increase capital spending on AI infrastructure; CITIC Securities' research report judges that there is currently no hard evidence of an “overall surplus in computing power.” However, on this expansion curve where the limit of production capacity has not yet been reached, the real shortcoming that has first emerged is security—when enterprises delegate more decision-making and operational permissions to AI systems and even AI Agents, permission management, data leakage, and unauthorized actions immediately become a “red line that must not be crossed,” rather than a feature that can be “added later.”
The financing rhythms of Zenity and Horizon3 are exactly the footnotes for this red line being repriced by the market: the former focuses on AI and AI Agent security and governance, securing $125 million in Series C despite the tight supply environment; the latter concentrates on automated security testing and attack-defense simulations, with its valuation jumping from about $650 million to over $2 billion in about a year. In contrast to external debates about the “computing power bubble,” the synchronized increase in financing amounts and valuations in the security sector indicates that what enterprises are genuinely concerned about is not “buying too many GPUs that won't be used,” but rather “once AI systems scale up and go online, they may not be able to maintain their boundaries.” At the same time, companies like Mariana Minerals, which are reshaping production processes in traditional resource industries using software and AI, also obtained $310 million in financing, indicating that AI security firms and deeply AI-utilizing enterprises are forming a bundled investment strategy: while AI infrastructure remains in a state of tight supply, security and governance have already preemptively become the constraints for implementation. The future shift of AI capital from “stacking computing power” to “focusing on security and governance” will be a decisive variable in whether enterprises dare to truly integrate more powerful AI into their core businesses.
The Next Act: Security Dividends and the Race Against the Bubble
After Zenity, Horizon3, and Mariana Minerals secured substantial funding of $125 million, $250 million, and $310 million respectively, the long-term logic and short-term risks of the current wave of AI security and application boom have been laid out on the table: on one end, the supply constraints of computing power and storage emphasized repeatedly in the financial reports of Microsoft and SK Hynix have made “laying the groundwork for infrastructure before scaling applications” a consensus. This has also supported the proposition that “the stronger AI becomes, the greater the need for security and governance,” which underpins the long-term narrative of companies like Zenity; on the other end, the valuation of Horizon3 being raised from about $650 million to over $2 billion within a year illustrates that as long as one can associate with the AI and security label, the valuation curve can be extremely steep, which also implies that the potential for correction is significant. More subtle risks arise from both the narrative and regulatory ends: the White House’s AI framework meeting has yet to take place, with agendas and policy directions still unclear, and valuations of security and compliance-related companies may be re-priced by new rule systems at any time; the RSI goals proposed by DeepMind’s management are merely bets made by some participants, still quite distant from becoming the mainstream market framework. What should we watch for in the next act? Firstly, whether the supply of AI infrastructure will actually transition from “tight” to “loose”; secondly, whether the rhythm of significant security incidents and regulatory landings will alter corporate risk preferences when moving to the cloud and using models; and thirdly, whether the new narrative represented by RSI can deliver on its promises in product capability and commercial monetization, determining whether security dividends are genuinely a long, thick slope of benefits or a short play that has been overdrawn by bubbles.
Join our community, let’s discuss, and become stronger together!
AiCoin exclusive Hyperliquid benefits: https://app.hyperliquid.xyz/join/AICOIN88
AiCoin exclusive Aster benefits: https://www.asterdex.com/zh-CN/referral/9C50e2
On-chain Telegram community: https://t.me/AiCoinWhaleData
On-chain community: https://www.aicoin.com/link/chat?cid=N6OVMor5g
AiCoin on-chain Twitter: https://x.com/aicoinwhaledata
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
