Palo Alto CEO 警告：代币成本需下降 90% 才能推动大规模 AI 应用

Palo Alto Networks 首席执行官 Nikesh Arora 警告称，为了促进人工智能的大规模应用，代币成本需要下降多达 90%。

“我认为 54% 是个不错的开始，”Arora 周四在 CNBC 的“Squawk on the Street”节目中告诉 CNBC 的 Seema Mody。此前 OpenAI 首席执行官 Sam Altman 对 CNBC 表示，该前沿实验室的最新模型在智能体编程方面的代币效率提高了 54%。“我认为我们可能还需要再来一次。”

Arora 表示，代币效率需要在未来 12 个月内下降至 20%，并在次年再下降 90%。

不断上涨的代币成本已成为企业的主要痛点，并对 AI 预算形成了压力。他说，当前的定价让企业实施 AI 工具变得越来越困难。

“我们需要看到 AI 的定价降下来，”Arora 说道。

Arora 是越来越多推动代币定价下降的高管之一。他们所担心的是，高昂的代币成本为广泛采用制造了主要障碍，导致许多企业无法使用这些工具。

上周，Palantir 首席执行官 Alex Karp 猛烈抨击了 Anthropic 和 OpenAI 使用的代币模式，并认为开放权重模型可能是一种解决方案。

“我不是在指责他们，但有些东西完全错了，”他在 CNBC 的“Squawk Box”节目中表示，“这个国家企业界的基本想法就是我可以躺平，在代币上浪费时间。”

代币问题正促使许多企业转向更便宜的开放权重工具，其中包括正在迅速缩小与美国实验室差距的中国模型。

与此同时，为支持大规模基础设施建设，AI 支出正在加速迈向新高。科技巨头也在寻找新的方式为此类 AI 投资提供资金，SpaceX 上月通过债券发行筹集了 250 亿美元，亚马逊 本周则发行 250 亿美元债券。

Arora 表示，市场将开始承受这一开支水平，否则企业将进行调整以适应市场。随着技术变得更高效，预算也会随之下降。

“重要的是要理解，需求依然是无限的，只要面对的是一条无限的需求曲线，我认为所有这些事情最终都会逐渐趋于合理，”他说。

Palo Alto Networks CEO Nikesh Arora warned that token costs need to drop as much as 90% to promote large-scale artificial intelligence adoption.

"I think 54% is a good start," Arora told CNBC's Seema Mody on "Squawk on the Street" on Thursday, after OpenAI CEO Sam Altman told CNBC that the frontier lab's latest model is 54% more token-efficient for agentic coding. "I think we probably need another turn at it."

Arora said token efficiency needs to drop to as much as 20% over the next 12 months, and 90% by the following year.

Rising token costs have emerged as a major pain point for businesses and put a strain on AI budgets. The current pricing, he said, makes AI tools increasingly difficult for businesses to implement.

"We need to see the pricing for AI come down," Arora said.

Arora is among a growing group of executives pushing for a decline in token pricing. The worry is that high token costs create a major barrier to widespread adoption, preventing many enterprises from using the tools.

Last week, Palantir CEO Alex Karp blasted the token model used by Anthropic and OpenAI, and called open-weight models a potential solution.

"I'm not throwing shade at them, but something has gone completely wrong," he told CNBC's "Squawk Box." "The basic view among enterprises in this country is I'm going to chillax and waste my time with tokens."

The token problem is leading many businesses to implement cheaper open-weight tools, including Chinese models that are quickly closing the gap with American labs.

At the same time, AI spending is accelerating to new highs to power the massive infrastructure build-out. Tech giants are also looking for new ways to fund these AI investments, with SpaceX raising $25 billion last month in a bond sale. Amazon raised $25 billion in debt this week.

Arora said the market will start to come to terms with the spending, or businesses will adjust to the market. Budgets will also decline as the technology becomes more efficient.

"It's important to understand the demand continues to be infinite, and as long as you have an infinite demand curve that you're facing, I think all these things will rationalize over time," he said.