Raising millions of dollars in financing, this startup aims to build a "highway" for AI Agents, triggering a delivery revolution worth hundreds of billions.

"AI is entering the next phase: the era of Agent intelligence. It can not only answer questions but also understand context, formulate plans, call tools, and even collaborate with other agents to complete complex tasks." - Jensen Huang, CEO of NVIDIA, clearly pointed out the evolution direction of Agent technology in his speech at the Paris GTC Conference in June 2025.

He further emphasized that infrastructure is the core to support the large - scale implementation of Agents: "We are collaborating with globally leading storage enterprises to build a new type of enterprise infrastructure required for deploying and scaling Agentic AI in hybrid data centers."

Almost at the same time, at the GTC Washington Technology Summit in October 2025, NVIDIA once again clarified the infrastructure blueprint centered around the "AI Factory", defining it as "a brand - new comprehensive computing platform specifically built for the generation, transmission, and service of massive tokens." The consensus of two AI leaders reveals an untapped blue ocean - Agent - Native infrastructure.

Image source: Internet

01 The "Last Mile" Dilemma of the Agent Economy

In 2025, the pace of AI agents moving from demonstrations to production environments is accelerating, but a core contradiction is becoming increasingly prominent: the huge gap between the advanced capabilities of agents and the backward deployment infrastructure.

Industry reports analyze that the agent technology stack is undergoing a key transformation from "model - centric" to "infrastructure - based". Currently, most enterprises still try to deploy agents using traditional DevOps processes designed for web applications, which is metaphorically described by industry experts as "taking off and landing an airplane on a highway" - the tools simply don't match.

Image source: "Research Report on China's AI Agent Industry" by Jazi Guangnian

The 15 - minute execution time limit, stateless design, and lack of session - keeping ability of traditional Serverless architectures (such as AWS Lambda) severely restrict the long - term operation, state persistence, and multi - round interaction capabilities required by agents. The results are: complex deployment, response delay, resource waste, and difficulty in ensuring secure isolation between users.

"We see a huge market gap - the runtime infrastructure exclusive to agents." A partner of Baidu Ventures said when analyzing the track, "This is not only a technical issue but also an economic issue that affects whether agents can be commercially used on a large scale."

02 CoreSpeed: Reconstructing the Infrastructure for Agent Runtime

Against this background, the Silicon Valley startup CoreSpeed has come into the sight of investors. Founded by Milton Yan and Hao Su, this company received millions of dollars in investment from Baidu Ventures and First Heart Capital just half a year after its establishment, and its valuation has reached tens of millions of dollars.

CoreSpeed's positioning is very clear: the world's first Agent - Native Runtime PaaS. Its core innovation solves several key pain points in agent deployment:

Ultra - fast startup in 127 milliseconds: When users initiate requests from any region in the world, CoreSpeed can start the corresponding Agent container within 127 milliseconds, achieving an almost instantaneous response.

Per - User Container architecture: Each user's agent instance runs in an independent container, ensuring complete security isolation and context preservation, and there will never be any crosstalk between users.

Intelligent elastic scaling: The system can dynamically scale from 0 to any size and automatically scale back to 0 after the task is completed, achieving true pay - as - you - go.

"Traditional DevOps was not designed for agents, which has brought the deployment of agents back to the era when operations engineers wrote tens of thousands of lines of Terraform." Milton Yan, CEO of CoreSpeed, explained, "We have removed all components that affect agent performance and reconstructed the entire chain to ensure that each agent can be efficiently launched and run quickly."

03 The "Next.js Moment" of the Agent Development Framework

In addition to its core technology, CoreSpeed has also open - sourced its agent framework Zypher, which is regarded by the industry as a smart ecological layout.

Zypher adopts a plug - and - play component design and is an agent development framework that is composable and can autonomously call tools. Developers only need to register tools - whether local tools or MCP Server - in Zypher's tool list, and Zypher can automatically complete tool orchestration and optimization through prompt words and weight control.

"Zypher is to CoreSpeed what Next.js is to Vercel." An early user commented, "The former defines the development standard, and the latter provides the cloud execution environment. Together, they form a complete closed - loop of Agent - Native infrastructure."

This metaphor accurately captures CoreSpeed's business logic. As an open - source framework, Zypher is quickly building a developer ecosystem: DeckSpeed, built on Zypher, topped the Product Hunt list in May; Zypher has become a 3 - credit CapStone course project at the Tandon School of Engineering of New York University; Meta engineers reproduced Lovable in an afternoon based on Zypher...

04 The Rise of the Agent Infrastructure Track

CoreSpeed is not an isolated case. With the explosion of the agent economy, a number of startups focusing on infrastructure are being favored by top VCs.

CoreSpeed: The Pioneer of Agent - Native Runtime

As the leader in the track, the CoreSpeed team has both technical and business genes in its background. CEO Milton Yan has a computer science background from Carnegie Mellon University and continuous entrepreneurship experience, and CTO Hao Su has been deeply involved in the field of distributed systems for many years. Its 127 - millisecond startup and Per - User Container architecture have become industry benchmarks.

Lindoon: Infrastructure for Agent Collaboration Network

Lindoon focuses on solving the communication bottleneck in multi - agent collaboration and provides a secure communication protocol and state - synchronization framework between agents. Founded by former Google and Figma engineers, this project recently received a $9 million Series A financing led by Sequoia Capital.

Cline: Agent Development Toolchain

Cline positions itself as "the GitLab of the agent era" and provides a complete agent development pipeline from coding, testing to deployment. The founders are from the core teams of GitHub and Vercel and have a deep understanding of the developer ecosystem. After completing a $12 million seed - round financing led by a16z in March 2025, Cline is rapidly iterating its cloud - based IDE and collaboration features.

DeepInfra: Agent Inference Optimization Platform

DeepInfra focuses on providing optimized infrastructure for large - model inference, especially optimizing long - term inference tasks for agents. The team consists of alumni from the Stanford AI Lab and former AWS engineers, and its dynamic batching and memory optimization technologies can reduce agent inference costs by 60%. This project has received $15 million in investment from Benchmark Capital.

Wuwen Xinqiong: Infrastructure Agent Swarm

This Chinese startup's "infrastructure agent swarm" realizes intelligent scheduling of cluster resources and fault self - healing through multi - agent collaboration. This solution has helped customers increase the iteration speed by 5 times and reduce resource consumption by 30%.

Image source: Tuoduan

05 The Future Explosion of Agents Driven by Infrastructure

"We are now on the eve of the explosion of various vertical delivery - level agents, and CoreSpeed will be an important participant in this explosion." Milton Yan is full of confidence in the company's future.

Industry analysts generally believe that 2025 - 2026 will be a critical period for the maturity of agent infrastructure. Just as the explosion of the mobile Internet awaited the maturity of the iPhone and Android, the full take - off of the agent economy is also awaiting the improvement of its exclusive infrastructure.

When the deployment of agents becomes as simple as deploying a website and resource costs change from fixed expenditures to pay - as - you - go, we may witness a more disruptive transformation than the large - model itself - millions of dedicated agents permeating every industry, every workflow, and every life scenario.

Infrastructure innovation is never as glamorous as application - layer innovation, but it is the most solid cornerstone in every technological revolution. While developers are excited about another new agent application, Milton Yan and his team are buried in container scheduling, resource allocation, and network optimization - these less - glamorous but crucial technical details.

"We believe that when the evolution paces of infrastructure, collaboration networks, and the application layer tend to be consistent, the Agent economy will enter a real acceleration phase." Milton Yan said, "As the underlying driving force of this transformation, Agent - Native infrastructure will not only become the core driving force in the agent era but also build a long - term moat for the next decade."

In this just - started marathon of agents, CoreSpeed is building a faster, more stable, and smarter track for all participants.

This article is from the WeChat official account "Silicon Rabbit King" (ID: gh_1faae33d0655), author: Silicon Rabbit King, published by 36Kr with authorization.