PETALING JAYA: Despite all the hype about OpenAI’s new ChatGPT-5 model, some Malaysian business leaders say companies should instead “go local” for artificial intelligence (AI).
Doing so can create jobs, protect Malaysian consumer data, make artificial intelligence more Malaysian and boost companies’ bottom line, according to insights released today by Juwai IQI.
Juwai IQI Group CEO Kashif Ansari explained, “You fall behind if you don’t use artificial intelligence, but costs can grow quickly. By ‘going local,’ as I call it, a typical large Malaysian company can save up to RM1.7 million per year, depending on usage rates. You also get the benefits of data privacy and security, customisation and automation.
“‘Going local’ is my way of saying two things. First, we are using open-source AI models that are offered by their creators for “free” to the larger community. These models are usually not quite as capable as the latest commercially available models on the market but are free of ongoing external usage costs. Second, we are hosting these models on our own servers. That means we never send our data to the big AI companies’ servers.”
He said most of what a typical business in Malaysia does can be accomplished with these open source AI models.
“Yes, we still use GPT-5, amongst others, but only for a fraction of our AI needs. We found that we can do most of what we need here in Malaysia by going local,” he added.
Juwai IQI Group COO and CIO, Nabeel Mungaye, said: “When you use artificial intelligence models from providers like OpenAI or Alphabet, you pay for usage: the more you use, the more you pay. But when you use open-source models hosted on your own servers, there are no ongoing usage fees. This is critical if you have high-volume AI applications, like customer service chatbots or PDF summarisation tools.”
“I’ve run some back-of-the-envelope numbers. A typical large Malaysian company that has embraced the use of AI for things like customer support, PDF summarisation, and marketing could end up paying over RM1.7 million (US$410,000) per year in usage fees from a service like OpenAI. On the other hand, running an open-source model on a local server for those same needs could cost as little as RM63,000 (US$15,000) per year. You would essentially just pay the cost of electricity and maintenance for a powerful server,” he added.
He explained, “Here’s how I arrived at those cost estimates. For a customer service chatbot processing one billion tokens annually, the cost using a paid mini-model would be approximately RM76,000 (US$18,000). PDF document summarisation is actually a very demanding task for AI, and handling some 500,000 documents per month could cost around RM1.65 million (US$390,000) per year. Lastly, the marketing content generation could
reasonably cost about RM10,500 (US$2,500). When you add these costs up, you get RM1.7 million (US$410,000).”
Mungaye said, “Going open source and local also allows us to make the artificial intelligence we use more Malaysian. We can fine-tune or train the model we use to understand local slang and cultural nuances in Malay, Mandarin, English, and other languages.
“We can also teach it industry-specific terminology, so it is more accurate and relevant for the ways that we use it. In sum, by ‘going local,’ Malaysian companies like us can get the benefits of AI at a lower expense while also protecting their costumers’ data and giving them a better service,” he added.