On June 10, the World Economic Forum released its 2026 Technology Pioneers list in Geneva. 100 startups from 23 countries were selected. I review this list every year, and the biggest difference this year compared to previous years is that, for the first time, the forum formally carved out a space for "AI agent infrastructure."
Not the AI model itself, not a chatbot, but the underlying capabilities that enable AI agents to act independently in the business world: identity verification, payment settlement, and commercial contracts.
Among the 60%+100 technology pioneers, over 60 companies are directly related to AI.
The forum has carved out a new track
In previous years, the trendsetters on the Technology Pioneers list were "AI models" and "consumer-grade applications," but this year the forum changed its phrasing. The original words of Verena Kuhn, the overall head of the Innovator Community, were: "Artificial intelligence is not only what these companies are building, but also the tool that makes all of this possible."
Let me translate: AI is no longer the goal, but the means. This statement itself is not new, but what the forum did next—dividing the selected companies into two major pillars—carries symbolic significance.
Pillar 1: Basic modules of autonomous AI entities—identity verification, payment, security, enterprise integration.
Pillar 2: Ensuring AI's demands for energy, computing, and storage.
Pillar One is the real focus. The forum is saying: for AI agents to "survive" in the commercial world, having a brain alone is not enough—they also need an ID and a bank account. This is not a metaphor; it is literal.
Skyfire and Paid: The "Interbank Clearing System" for AI Entities
The two companies selected this year caught my attention.
Skyfire US focuses on AI agent identity verification and payment infrastructure. Simply put, it enables AI agents to complete commercial transactions under their own identity, rather than requiring humans to press the confirm button every time.
Paid UK, building AI-oriented commercial infrastructure — pricing, billing, and renewal management. Enabling AI "employers" to pay based on usage, managing the AI workforce like SaaS subscriptions.
Looking at these two together, the logic becomes clear: for the AI ecosystem to conduct business, it needs "who is trading" (solved by Skyfire) and "how to collect payments" (solved by Paid). This is the same principle as the early internet requiring CA certificates and payment gateways.
Early internet CA certificates + early identity verification of payment gateway AI entities + billing management
I checked, and there are currently almost no mature players in this track. SAP and Microsoft have implemented AI entity identity management within their respective enterprise systems, but that is within closed ecosystems. Cross-platform AI entity identity and transaction infrastructure is still a blank space.
10 Chinese companies selected: mainly manufacturing and healthcare
Among 23 countries, the United States has 45 companies, and China ranks second with 10. I sorted out the directions of the Chinese selected companies:
| Enterprise | Direction |
|---|---|
| Deep Wisdom | Automated Machine Learning Platform (E-commerce + Manufacturing) |
| 九科信息 | Large Language Model + RPA |
| Yida Technology | Trade and Intelligent Systems AI |
| Deep Intelligent Control | Physical information integrated with AI for industrial energy saving |
| Tripo AI | AI-generated 3D model |
| Landing Medical | AI medical imaging |
| Phenomenon Innovation | AI innovative applications |
| 千觉机器人 | AI-perceptive robot |
| MicrobeWorks | Synthetic Biology + AI |
| Parallel Bio (China Team) | AI drug discovery |
10 of the companies directly serve the manufacturing industry—automation ML, RPA, industrial energy efficiency, and perception robotics. This proportion aligns with the domestic policy orientation of AI + manufacturing. However, I did not see any Chinese companies working on "AI physical infrastructure" being selected; this area is indeed a gap.
Another signal: Futurum ranks the top ten AI agent vendors
Just 10 days ago (June 1), Futurum Research released a report with a straightforward title: Agentic AI: The Leading Vendors Winning the Enterprise in 2026.
The report lists the top ten vendors in the enterprise-level AI agent track: Microsoft, Salesforce, ServiceNow, AWS, Google, IBM, Oracle, Palantir, SAP, UiPath.
I looked at this ranking several times and found an interesting point — the top ten are all companies making enterprise software and cloud platforms, not a single one is a large model company. None of the model companies like OpenAI, Anthropic, or Google DeepMind made it in.
What does this mean? When enterprises purchase AI agents, they are not buying model capabilities but business process integration capabilities. Whoever has the smoothest-running AI agents within their ERP/CRM/ITSM systems wins. The model is merely the underlying engine, just like databases and operating systems—users will not pay for the engine itself.
Among the top 10 enterprise AI agent vendors in Futurum, the number of pure large model companies
On the same day, the forum also published an in-depth article with an even more striking title: Who Will Control the Enterprise Agentic Workforce? — Who will control the enterprise AI agent workforce? The conclusion is that CIOs are facing a new platform war: Microsoft, Salesforce, ServiceNow, AWS, and Google all want to become the "operating system" for enterprise digital labor, and CIOs need to avoid being locked into a single vendor.
Three practical suggestions for enterprise CIOs/digital leaders
Looking at these three things together — the forum defining AI infrastructure, Futurum ranking vendors, and the top ten vendors all being process platforms rather than model companies — the signal for enterprise digitalization is actually very clear:
First, don’t wait for AI agent infrastructure to mature before taking action. Skyfire and Paid have just been selected as Technology Pioneers, indicating that this track is only just beginning. But your company may already be using products like Copilot and Agentforce—AI agents are already running within your business processes. Now is the time to establish identity management and permission controls for AI agents, even if it starts with something as simple as "who can trigger which AI agent."
Second, when choosing a platform, look at the depth of business process integration, not model parameters. Futurum's ranking has already shown that the key to enterprise AI agents lies not in the model, but in the process. An AI agent running within your ERP is far more useful than an isolated enterprise version of ChatGPT.
Third, monitor the progress of standards for cross-platform interoperability of AI agents. The MCP protocol is currently the fastest-moving, with support from both Anthropic and Microsoft. If your AI agent can only run within a single platform, the cost of future migration will be high.
