SuperAI Singapore 2026: The Model Wars Are Over. The Real Race Has Just Started

I spent the past two days at SuperAI Singapore, Asia's largest AI conference with ten thousand attendees, 1,500 AI companies from more than 150 countries, all packed into Marina Bay Sands. Balaji Srinivasan, Max Tegmark and Benedict Evans on the main stage alongside leaders from OpenAI, Google DeepMind, Mistral and MiniMax.

Singapore positions itself as the neutral ground where East meets West in AI and this year that framing felt earned. With export controls tightening and sovereign AI strategies accelerating, it is one of the few places where American, Chinese and European AI leaders share a stage and speak candidly.

Here is what they said, and what it means for Australian businesses.

The models are becoming commodities

Benedict Evans put it most bluntly. The big four tech companies are on track to spend around US$700 billion on AI infrastructure in 2026, up from roughly $400 billion last year. Yet every lab is building with the same data, the same chips and the same infrastructure. Benchmark scores for OpenAI, Anthropic, Google, Meta and the Chinese labs are converging. There is no fundamental path to a sustainable lead at the model layer.

Zixuan Li from Z.ai gave the panel version of the same argument. Intelligence is approaching an asymptote. Each marginal point of model improvement will require exponentially more capital, moving from $10 billion to $100 billion per increment. The model layer commoditises and competition shifts to price.

The conclusion both reached: models become infrastructure, and innovation moves up the stack to applications, orchestration and the businesses that actually deploy this stuff well.

If you run a business, this is genuinely good news. You do not need to pick the winning model. You need to get good at using whichever model is cheapest and best for your task, and that capability sits inside your business, not inside a lab in San Francisco or Hangzhou.

Your context is the moat, not your subscription

The frontier models panel landed on a point I have been making to clients for two years. Simply subscribing to a general purpose AI tool is a temporary productivity gain, not a competitive advantage. Everyone has access to the same subscription.

The real moat is embedding your business context, your data, your processes and your standards into how AI works for you. Mistral framed it as the difference between a Swiss army knife and a purpose-built tool. The Swiss army knife is great for testing ideas. The purpose-built tool, loaded with your enterprise context, is what compounds.

Crystal Buchanan from Chat and Build made the same point from a different angle. Every model is trained to produce the most acceptable average output. What makes AI output stand out is taste, meaning your individual judgment, standards and perspective. Before you build agents or write prompts, codify how your business actually wants things done. Without that, every AI output is generic by design.

From chatbots to agents that act

Google Cloud research shared on the 100x Company panel found knowledge workers now interact with roughly 257 different AI surfaces. That fragmentation is producing diminishing returns. The shift underway is from chatbots you prompt to agents you trust to make decisions on your behalf.

Humans make around 4,000 decisions a day. Increasingly, agents will make thousands of those. The enterprise adoption pattern is starting with the easily delegated work: travel booking, calendar management, email drafting and meeting capture. Marketing and sales teams are already among the heaviest agent users, with token usage rivalling engineering departments.

The proactive version of this is coming fast. Buchanan, fresh from an OpenAI launch event the day before her talk, demonstrated agents that monitor your calendar and inbox and act before you ask. The example that stuck: your boss has sent three emails in fifteen minutes and you have not responded, so here is a draft reply.

The 100x company is not about output

The sharpest reframe of the conference came from INSEAD strategy professor Hyunjin Kim. Most people hear 100x and think more output: more documents, more code, more activity. Wrong. A 100x company changes its production function. What can you produce, with what business model, with the same people and capital?

Plaud's CEO Nathan Xu described restructuring 200 engineers into small pods of one designer, one product person and a couple of engineers, each pod running a product with AI doing the heavy lifting. His framing matters for every business owner worried about AI and jobs. The opportunity is not doing the same work with fewer people. It is growing the business 10x while growing the team modestly, and letting people do what they are best at, which is working with other humans.

Investors are watching this shift too. On the East-West panel, revenue per employee is replacing headcount as the key signal of a healthy AI-era company.

The pro-human pendulum is swinging

Max Tegmark of MIT gave the most important talk of the event. He framed AI development as a fork in the road. One path races to replace humans entirely. The other builds controllable tool AI that augments people without wholesale replacement.

The encouraging news is the pendulum is swinging toward the second path. Only around five per cent of Americans support unregulated AI development. The EU AI Act comes into full force in October. The US is moving toward regulating AI like medicine. Even the frontier labs are publicly discussing pausing the riskiest development.

And tool AI, the controllable kind, can still deliver enormous value: safer roads, better healthcare, drug discovery, personalised education and productivity gains across every sector. Dr Daniel Ting's healthcare keynote made it concrete. Physicians using AI now outperform physicians who do not, and AI-assisted diagnosis is four times more accurate than physicians alone. The famous line applies to every profession: AI will not replace you, but people who embrace AI will replace those who do not.

This is the thesis Advancer was built on. People-first AI is not a slogan, it is our winning strategy and the world's leading AI thinkers are converging on it.

What this means for your business

Three things to act on this quarter.

First, stop waiting for the model wars to settle as they already have. The models are commodities and the advantage now sits in how well you deploy them inside your business.

Second, codify your context. Document your standards, your processes and your voice so AI can work to your specification, not the internet's average. This is the unglamorous work that separates businesses getting real AI outcomes from businesses paying for subscriptions.

Third, pick one workflow and move it from chatbot to agent. Meeting capture, email triage and calendar management are proven starting points. Build trust with delegated decisions before you hand over bigger ones.

The race is no longer about who builds the smartest model. It is about who builds the smartest business around the models. That race is wide open, and Australian SMEs can absolutely win it.

Want to know where to start? Book a discovery call with the Advancer team!

Llew Jury is Managing Director of Advancer, Brisbane's AI agency, and co-host of the Zero Shot podcast. He attended SuperAI Singapore 2026 on 10 and 11 June.