This morning, SuperAI 2026 opened at Marina Bay Sands in Singapore. Over 10,000 people, 1,500 companies, and more than 150 speakers — tickets had long been sold out.
Walking around the venue, you'll notice that the AI industry chain has become as complete as the internet in the early 2000s—from chips (NVIDIA, AMD, Cerebras, Groq) to models (OpenAI, Anthropic, Mistral) to applications (Salesforce, Palantir), and then to robotics (Boston Dynamics, Unitree Robotics), every link in the industry chain has someone setting up a booth.
Chinese companies have a notable presence. Huawei, Tencent, Alibaba, ByteDance, and Unitree are all on the exhibitor list.
But what truly caught my interest wasn't what was being discussed inside the venue — but three things happening outside it.
Signal 1: 58% of enterprises are already using "Physical AI"
There is a piece of data in Deloitte's 2026 Enterprise AI Report that I think many people have overlooked.
58% → 80% Currently, 58% of enterprises have at least limited use of physical AI (robots, drones, IoT+AI), and Deloitte expects this proportion to reach 80% within two years.
What is "Physical AI"? It's not the kind of AI like ChatGPT that types and chats on a screen, but AI that can interact with the physical world—collaborative robots on assembly lines, inspection drones with automatic responses, robotic arms for automated sorting, and energy systems that can self-regulate.
The report cited several specific application scenarios: collaborative robots on manufacturing production lines are no longer new, automated forklifts and robotic picking arms in the logistics industry are accelerating their penetration, and inspection drones capable of automatically identifying anomalies are being used in the defense and infrastructure sectors. In the Asia-Pacific region, the implementation speed of physical AI has increased by 22 percentage points compared to two years ago.
This trend means a change for peers working on enterprise digitalization: previously, when we talked about "digital transformation," it was basically about tinkering in the digital world—ERP, CRM, OA, data platforms. Now AI is connecting these digital systems with physical production lines, warehousing and logistics, and equipment maintenance. The boundary between the digital world and the physical world is disappearing.
For example. Among the manufacturing cases tracked by KPMG, an electronics manufacturer used AI for intelligent scheduling, reducing production line changeover time by 42%. This 42% is not a "software efficiency improvement" — it is real production time saved. Another auto parts factory increased the defect detection rate of AI visual quality inspection from 92% to 99.7% — again, this is not a number on a screen, but a change in the physical world's yield rate.
Signal 2: "Doubao" enters the car, and not just as a voice assistant
Last night (June 9), Saidou Technology, in which Seres holds a stake, launched a new brand called AIVA in Beijing. Simply put, it is Seres' manufacturing capabilities + CATL's batteries + ByteDance's Volcengine's Doubao large model — the three parties have put together a new automotive brand called AIVA.
The point is not that there is another car brand. The point is their slogan: "AI defines the car, first there is AI, then there is the car."
AIVA President Li Bo's original words at the press conference: AI will change the relationship between people and cars—cars will transform from tools into companions that understand you.
Let me translate the industrial implications of this statement: The Doubao large model in this vehicle is not an "add-on feature," nor is it a voice assistant like "Hello Xiaodi, navigate me to the company." It is the starting point of product definition. The entire vehicle has been designed around AI from the design stage.
This is completely different from five years ago when car companies said, "We have smart cockpits." Back then, the car was built first, and then a screen and a voice assistant were stuffed into it. Now, it's about first defining the capability boundaries of the large model, and then deciding what form the car should take.
For those working on enterprise digitalization, this is a signal — when a general-purpose large model like "Doubao" can enter cars, production lines, and equipment, the data silos in ERP systems will no longer hold. Your customer relationship management, supply chain, and after-sales maintenance could all be reorganized by an AI that understands context and can automatically invoke multiple systems.
Not just cars. Deloitte data shows that finance, aviation, and the public sector have already implemented similar models—AI is no longer a tool placed in the IT department, but has grown into the products and services themselves.
Signal Three: A full venue is just the surface; where the money flows is the key
SuperAI 2026 has two notable figures: the Genesis Startup Competition has a total prize pool of $2.3 million, co-sponsored by OpenAI and Microsoft. The NEXT Hackathon has a prize pool of $200,000, backed by AWS and Vercel.
This means these big companies are not "supporting innovation" — they are competing for the enterprise software gateway of the next decade.
Looking at SuperAI's exhibitor list alongside the trends in Deloitte and KPMG reports, a relatively clear main thread emerges:
| Hierarchy | Representative Enterprise | Ongoing changes |
|---|---|---|
| Computing power layer | NVIDIA, AMD, Cerebras, Groq | The cost of reasoning continues to decline, and the ROI inflection point for enterprises building their own AI infrastructure is approaching. |
| Model Layer | OpenAI, Anthropic, Doubao, Mistral | No longer competing on parameters, starting to compete on the ability to "embed into real business processes" |
| Application layer | Salesforce, Palantir, SAP | From "AI features" to "AI-native" — not adding AI to old software, but rebuilding software architecture with AI |
| Physical Layer | Boston Dynamics, Unitree, Foxconn | Robots and AI are no longer separate; behind every physical action lies a digital decision. |
Last month, the SAP China Summit officially introduced the concept of the "autonomous enterprise" to the Chinese market — 224 AI agents, a knowledge graph-driven business map, and a shift from "recording business" to "running business." I've written about it before, so I won't repeat it.
Interestingly, when SAP talks about "autonomous enterprises," Deloitte talks about "physical AI," Saido talks about "AI-defined cars," and SuperAI talks about "AI meets the world"—these are not matters of the same industry, but their underlying logic is exactly the same: AI is transforming from an "optional feature" into "infrastructure."
Just as companies twenty years ago wouldn't discuss "whether to adopt ERP," today's companies also won't discuss "whether to use the internet"—AI is entering the same stage.
What this means for CIOs
I summarize three practical judgments.
First, re-evaluate your tech stack. If the AI capabilities of your enterprise software (ERP, MES, WMS) are built by stacking "plugins" rather than growing from the architecture layer, the cost will become increasingly significant over the next two years. This doesn’t mean you need to switch systems now, but when making technology selections, "AI-native architecture" should become a hard criterion.
Second, physical AI is not about robots; it's about data. Deloitte says 58% of enterprises are already using physical AI. But KPMG's data reminds you—76% of enterprises are stuck on data bottlenecks. Whether physical AI can run depends first on whether your equipment data has been collected, whether the format is unified, and whether real-time performance is sufficient. Fix the data pipeline on the production line first, then talk about robots.
Third, don't wait for "industry standards". The infrastructure transformation of AI is happening too fast. By the time industry associations release standards or policy documents provide guidelines, your competitors may have already successfully implemented three scenarios. Find a small entry point first, get the data, then expand—this is currently the only proven effective path.
