by Tan Aik Keong (AK)
Over the past few decades, a handful of infrastructures have completely changed how we live. When you turn on a tap, you don't think about which reservoir the water came from. When you flip a light switch, you don't think about which power plant is generating the electricity. When you go online on your phone, you don't think about which cell tower your data passed through. For most people, there's really only one thing that matters: when you need these resources, they're there — stable, convenient, and reasonably priced.
I think artificial intelligence is heading the same way. Today, when people talk about AI, they still mostly picture ChatGPT, Claude, Gemini, or one generative AI app or another. But over the next decade, AI may stop being a collection of separate software products and gradually become a resource you draw on the way you already draw on electricity, the internet and water. In other words, what we're really buying won't be a particular AI product anymore — it'll be intelligence itself.
The AI Token Factory
Under this shift, a new concept starts to matter: the AI Token Factory. If electricity is measured in kWh, water in cubic metres, and telecom data in GB, then one of the most important units of measurement in the AI world is the Token. Every time you ask an AI a question, upload a file, have it analyse data, generate a report, write code, or even control a robot, you're consuming tokens.
That points to a fundamental shift in how the AI industry will make money. Software used to sell licenses, then moved to SaaS, where businesses paid per user, per month. In the AI era, businesses are increasingly likely to pay for the intelligence they actually consume. A company won't need to own a large AI model or buy its own fleet of GPUs, any more than a restaurant builds its own power plant just because it needs electricity. What a business actually needs is to be able to draw on intelligence the moment it's needed, and pay only for what it uses.
Seen this way, an AI Factory looks a lot like a future "intelligence power plant." A traditional data centre mainly stores and processes data; an AI Factory is far more active. GPUs, AI models, enterprise data, networking, power and cooling combine to take in questions, instructions and data, and output answers, code, predictions, designs, recommendations — or directly execute actions. A power plant converts energy into electricity; an AI Factory converts compute, power and data into intelligence that people and machines can use.
Tokens produced matter more than GPUs owned
That means the value of a future AI infrastructure may no longer be measured by how many GPUs it owns. The more important question is how many tokens those GPUs can produce each day, and how much real economic value those tokens create. Owning a lot of compute isn't the same as owning a lot of productivity — what matters is whether that compute can actually be put to work by businesses, governments, hospitals, schools, factories and ordinary consumers.
AI as the fourth utility
I'd go as far as saying AI could become the fourth major utility, after water, electricity and telecoms — an "intelligence utility." When you wake up, your personal AI agent may already have read your email, organised your day and filtered the important news. By the time you get to the office, AI has analysed yesterday's sales data, flagged unusual orders, and told management what to prioritise today. When a doctor sees a patient, AI can help organise medical records and reference material; when an engineer designs a product, AI can simulate multiple options at once; when a developer builds a system, more and more of the code may be generated directly by AI.
In future, a person won't have just one AI assistant — they may have a dozen AI agents running at once: one for finances, one for scheduling, one for health, one for learning, one for shopping. It's even more pronounced for businesses: a mid-to-large company may one day run hundreds or thousands of AI agents in the background, continuously handling customer service, HR, finance, sales, procurement, legal and operations. At that point, the number of tokens each person and each company consumes every day will climb fast.
Token-as-a-Service
That will give rise to a new business model: Token-as-a-Service. What a business buys may no longer be a fixed piece of software, but a pool of "intelligence credit." A company might buy a billion tokens a month, then allocate them across customer service, finance, sales, development and internal AI agents according to each department's needs. We may even see tiered tokens: cheaper models for simple tasks, and more expensive, more capable models called on for complex legal analysis, software development, scientific research or major business decisions.
It's a bit like how we use electricity and telecoms today — we don't run everything on the highest tier of network, and we don't leave every device drawing maximum power at all times. Future AI systems will similarly learn to automatically pick the right cost tier of intelligence: cheap tokens for easy tasks, premium tokens only for the hard problems. What businesses will really need to manage isn't just an IT budget — it's an "intelligence budget."
Intelligence security
If this trend holds, it could even reshape how nations compete. Today, a country takes energy security, food security, cybersecurity and data security seriously. In future, we may need to start talking about a new concept: Intelligence Security. If government, banking, healthcare, manufacturing, transport and even defence systems come to depend heavily on AI, then how much AI compute and how many tokens a country can reliably access every day could become part of its national competitiveness.
That's also why the world is pouring investment into GPUs, data centres, sovereign AI and AI Factories. For Malaysia, this trend is particularly worth watching. We already have growing data centre investment, along with strategic location, power infrastructure, multilingual talent, and proximity to ASEAN markets. But the next question shouldn't just be "how many data centres does Malaysia have" — it should be "how much genuinely valuable intelligence can these data centres produce for Malaysia?"
More importantly: do those tokens actually translate into business productivity, jobs, new industries and GDP? If most of this AI infrastructure ends up serving overseas companies while local businesses can't fully use that intelligence capacity, then what we have is infrastructure — not a complete AI economy. The real opportunity is making it cheaper and easier for Malaysian businesses to access AI intelligence, and to apply it across manufacturing, finance, healthcare, retail, logistics, education and government services.
The bill of the future
Twenty years ago, it was hard to imagine that today each of us would buy tens or even hundreds of gigabytes of mobile data every month — "buying internet data" sounded like a strange concept back then. A decade from now, buying AI tokens may feel just as ordinary. Households will have their own AI agents; businesses will run large fleets of them; cars, robots and factories will be calling on AI continuously.
Water lets cities survive. Electricity lets machines run. The internet lets the world connect. AI tokens may be humanity's first way to access intelligence at scale, and cheaply.
Once intelligence can be drawn on the moment it's needed, switched off when it isn't, and paid for strictly by usage — that's when the real AI revolution begins.
One day, the monthly bill that already includes water, electricity and phone may well have one more line item: AI Token.
Part of the AK AI Corner column. Originally published in Oriental Daily (东方日报) on Sep 3, 2026.