Kaiyuan Securities released a research report stating that AWS recently raised the price of its EC2 machine learning capacity blocks by approximately 15%; Google Cloud simultaneously increased multi-method data transfer prices, with rates in North America doubling. AWS and Google firing the first shot in cloud price hikes reflects strong downstream computing power demand, and domestic cloud providers are expected to follow suit. The underlying logic for the price increases stems from recent hikes in CPU and storage costs, coupled with a surge in computing power demand driven by the execution of complex tasks by Agents. Furthermore, general-purpose Agents, represented by Clawdbot, can smoothly execute complex tasks 24/7, demanding even greater computing power, which is expected to further propel the increase in computing demand. The report recommends focusing on the AI infrastructure industrial chain. The main views of Kaiyuan Securities are as follows:
AWS and Google have fired the first shot in cloud price hikes, and domestic cloud vendors are expected to follow. (1) Amazon Web Services (AWS) recently increased the price of its EC2 Capacity Blocks for ML by about 15%, with the hourly cost for a p5e.48xlarge instance rising from $34.61 to $39.80. (2) On January 27, Google Cloud officially notified via email that it will adjust data transfer pricing for methods including CDN Interconnect, Direct Peering, and Carrier Peering effective May 1, 2026, with North American rates doubling. Specifically: North America: $0.08/GB (previously $0.04/GB); Europe: $0.08/GB (previously $0.05/GB); Asia: $0.085/GB (previously $0.06/GB). Kaiyuan Securities believes that AWS and Google's actions signal the start of a trend, and domestic cloud providers are likely to follow. The rationale behind the price hikes is the recent increase in CPU and storage costs, combined with a sharp rise in computing power demand due to complex task execution by Agents.
Clawdbot is gaining sustained attention, and general-purpose Agents are expected to drive high growth in computing demand. The open-source project Clawdbot has recently sparked widespread discussion in the tech community. Created by Peter Steinberger, founder of PSPDFKit, and launched at the end of 2025, it has already garnered over 65,800 stars on GitHub. Clawdbot is an AI Agent under an AGI prototype that can run on local computers, is compatible with various systems like MacOS, Windows, or Linux, and can integrate with common chat tools such as WhatsApp, Telegram, Discord, Slack, Signal, or iMessage. It can also connect directly to productivity tools like GitHub, Vercel, and Google Drive. In addition to built-in tools commonly used by Agents, its capabilities can be extended through plugins and Skills. Kaiyuan Securities argues that general-purpose Agents, exemplified by Clawdbot, which can smoothly execute complex tasks 24/7, impose higher requirements on computing power and are likely to further accelerate the growth in computing demand.
Domestic model vendors are competing to deploy general-purpose Agents, expected to usher in new breakthroughs. On January 27, Kimi released and open-sourced its K2.5 model, introducing new visual understanding, code, and Agent cluster capabilities. Furthermore, the K2.5 model expands Kimi Agent capabilities into daily office work, beginning to master intermediate-to-advanced skills in commonly used software like Word, Excel, PPT, and PDF. On January 15, Alibaba's Qianwen held a product launch event in Hangzhou. The Qianwen App has been fully integrated with Alibaba's ecosystem businesses including Taobao, Alipay, Taobao Flash Sales, Fliggy, and Amap, enabling AI shopping functions like ordering takeout, shopping, and booking flights. The Qianwen App has been upgraded from a standard AI assistant to an Agent entry point for the Alibaba ecosystem. Multi-Agent collaboration has already become an industry trend and is expected to evolve into a productivity tool and life assistant for humans.
Regarding investment targets, the report specifically recommends Sangfor Technologies (300454.SZ). Beneficiary companies include Parallel Processing Tech (920493.BJ), QingCloud Technology (688316.SH), UCloud Technology (688158.SH), KINGSOFT CLOUD (03896), YunSai ZhiLian (600602.SH), Capital Online (300846.SZ), and Tongniu Information (300895.SZ). Risk warnings include technological progress falling short of expectations and the risk of intensifying market competition.
Comments