英特尔首席执行官陈立武（Lip-Bu Tan）近日表示，人工智能的发展势头远未达到顶峰，其未来影响力甚至有可能超过互联网。然而，在AI快速扩张的过程中，行业正面临一系列关键瓶颈，其中最值得关注的挑战之一是如何确保未来数十亿乃至更多AI代理（AI Agents）不会执行未经授权的操作。
陈立武是在与人工智能基础设施初创公司 Hyperbolic 首席执行官张璟（Jasper Zhang）进行交流时发表上述观点的。此次讨论发生在 Hyperbolic 完成3000万美元融资之后，该轮融资获得了陈立武旗下风险投资机构 Walden Catalyst 的支持。
谈及人工智能行业的发展前景时，陈立武表达了乐观态度。他认为，AI浪潮仍将持续推进，其规模甚至可能超过互联网时代所创造的影响力。不过他同时指出，当前行业面临的核心问题已经不再是需求，而是“扩张速度究竟能有多快”。
在他看来，人工智能基础设施建设正受到多个因素制约。其中包括高性能内存供应不足、GPU短缺、电力资源紧张以及数据中心建设能力受限等问题。随着越来越多企业投入AI领域，对计算资源的需求持续攀升，这些因素正在成为行业发展的现实障碍。
除了硬件供应问题外，融资压力也正在成为新的瓶颈。陈立武表示，建设大型人工智能数据中心需要极其庞大的资本投入，而资金获取能力正在逐渐限制行业进一步扩张。“融资正在成为增长的瓶颈之一。”他指出。
事实上，陈立武此前已经多次对AI基础设施供需失衡发出警告。他曾预测相关硬件短缺可能持续三年时间，并强调只有当投资者能够持续看到明确的投资回报率（ROI）时，大规模资金才会继续流入人工智能产业。
不过，相比硬件和资金问题，陈立武认为另一个更长期、更复杂的挑战正在浮现，那就是AI代理的安全性问题。
随着AI代理逐渐具备自主执行任务、调用工具、访问系统以及协助决策的能力，未来全球运行的AI代理数量可能达到数十亿级别。如何确保这些系统始终处于授权范围内运行，避免执行超出权限的行为，将成为整个行业必须解决的问题。
陈立武此前曾透露，他经常让AI代理在夜间持续工作，并在第二天早晨检查工作成果。这种使用方式与微软CEO萨蒂亚·纳德拉等科技行业领导者的做法类似，反映出AI代理正在逐步进入实际生产环境。
然而，代理系统参与的任务越复杂，其安全风险也会相应增加。陈立武认为，未来行业需要建立更完善的机制，确保AI代理能够在可控边界内运行，防止其执行未经批准的操作或引发潜在风险。
尽管挑战重重，陈立武依然对人工智能的长期前景保持乐观。他认为当前人类正处于AI时代发展的早期阶段，而未来围绕代理式人工智能（Agentic AI）所产生的创新和商业机会仍然十分巨大。
在他看来，未来人工智能产业能否真正释放全部潜力，不仅取决于算力和资本投入，更取决于行业能否建立足够可靠的安全体系，让数十亿AI代理在获得自主能力的同时，始终保持可预测、可控制和可信任的运行状态。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Chief Executive Officer Lip-Bu Tan recently stated that artificial intelligence is far from reaching its peak, and that its future influence is likely to exceed that of the Internet. However, the industry is facing a series of key bottlenecks in the rapid expansion of AI, one of the most noteworthy challenges being how to ensure that billions or even more of future AI Agents do not perform unauthorized operations.
Chen Li Moo expressed this view in an exchange with executive director Jasper Zhang, the founding firm of artificial intelligence infrastructure. This discussion took place after Hyperbolic had completed the $30 million financing, which was supported by Walden Catalest, the risk investment agency under the flag of Chen.
Turning to the prospects for the development of artificially intelligent industries, Chen has expressed optimism. In his view, the wave of AI would continue to advance and might even exceed the impact created by the Internet era. At the same time, however, he noted that the core issue facing the industry was no longer demand but “how fast can it expand”.
In his view, the development of artificial intelligence infrastructure was being constrained by a number of factors. These include such issues as inadequate supply of high-performance memory, shortage of GPUs, power resource constraints and capacity-building constraints in data centres. These factors are becoming a real obstacle to industry development as more enterprises invest in AI and demand for computing resources continues to rise.
In addition to hardware supply problems, financing pressures are becoming new bottlenecks. Chen Li Mu indicated that building large artificial intelligence data centres requires enormous capital investment, and access to finance is gradually limiting further expansion of the industry. “Financing is becoming one of the growth bottlenecks.” He noted.
In fact, Chen has warned repeatedly about the imbalance between supply and demand for AI infrastructure. He had predicted that the hardware shortage could last for three years and had stressed that large-scale capital flows into the artificial intelligence industry would continue only if investors could see a clear return on investment (ROI) on a sustainable basis.
However, Chen believes that another longer-term and more complex challenge is emerging than hardware and funding, namely the security of AI agents.
The number of AI agents operating globally may reach billions in the future as AI agents develop the capacity to carry out their tasks, access tools, access systems and assist in decision-making. Ensuring that these systems remain within their mandate and that actions beyond permission are avoided will be an issue for the entire industry.
Chen Liwu had previously disclosed that he regularly kept AI agents working at night and checked the results the following morning. This pattern of use is similar to that of leaders in science and technology, such as Microsoft CEO Satya Nadra, and reflects the progressive entry of AI agents into the actual production environment.
However, the more complex the proxy system is involved in its tasks, the greater their security risks. According to Chen, the future industry needs to put in place better mechanisms to ensure that AI agents can operate within controlled boundaries to prevent unauthorized operations or potential risks.
Despite the challenges, Chen Zhang Wu remains optimistic about the long-term prospects for artificial intelligence. He believes that humanity is at an early stage in the development of the AI era, and that the future opportunities for innovation and business surrounding the agency AI remain enormous.
In his view, the real release of the full potential of the artificial intelligence industry in the future would depend not only on calculus and capital inputs, but also on the establishment of a sufficiently reliable safety system by which billions of AI agents would remain in a predictable, controlled and trusted state of operation while gaining autonomy.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.