A few months back Indonesian vice president Gibran Rakabuming announced some initiatives that promoted the use of AI to the younger demographic. One of them is a plan to include AI curricula for the K-12 students. The initiative drew some criticism from tech professionals, mainly due to being perceived as a mostly populist act. However, the interest in AI is beyond the vice president’s personal interest but a national one. The Indonesian Ministry of Communication and Digital Affairs (Komdigi) is said to be working on the Indonesian AI roadmap as of June 2025. As a  member of the public, it is in our interest to know whether we can expect a substantial digital transformation or if this is merely aping the global AI hype. Additionally, if the roadmap succeeds to deliver a thriving local AI scene this can be a blueprint for fellow Global South countries facing similar challenges.

Understandably, it is tempting to consider aiming for homegrown AI supremacy for a few reasons. First, AI is the zeitgeist of post-covid tech industry. We have seen firsthand how technologies like ChatGPT change the way we use the internet. For some people, ChatGPT has replaced Google to become their default place to search for information with a more human-like experience through the chat interface. Secondly, the very recent explosion of AI popularity gives an impression that this is a newly opened green pasture. Unlike high tech industries that rely on physical infrastructure such as manufacturing, the digital industry is deemed to be a more even playing field for the Global South countries to catch up to the more industrial countries. Similar hope is also put on AI as it is considered to be a no man’s land at this point. As Indonesia is experiencing tech winter in the past few years, this article will focus on what Indonesia can do amidst the current challenge to be an active player in the global AI scene.

The Rage about GPT and other AI Products

Before going further into the plan of actively participating in the global AI race, let us dive deeper on what AI really means. In layman terms, current-day AI would most commonly be understood as text-based applications where users can submit prompts expecting to get intelligent responses based on the submitted prompt such as the ChatGPT or Meta AI on WhatsApp. All of them are the extension of what we call the large-language model (LLM). LLM is a type of machine learning model used for natural language processing tasks such as language generation or translation. LLM is distinct compared to the previous model because it “learns” from a huge corpus of texts making it able to “understand” plenty of information it was fed with. The way it could respond to various prompts naturally with answers that seemed to be written by educated humans is the reason for AI's most recent popularity. Despite the human-sounding responses, they are generated by advanced statistical inference algorithms that make the results be as close as natural language. Sometimes the models “hallucinate” by making up incorrect answers, but they work well for most questions that are readily available on the internet; which is good enough for most users.

Aside from LLM-based applications, many software we use daily runs on various forms of AI. Consider face recognition technologies. It goes from automatically detecting faces when taking pictures to grouping photos with the same face in your photo album without any instruction. In other industries such as banking, they have been using traditional machine learning methods to do fraud detection by finding anomalies in consumers’ transaction behavior. The two examples above have very different technicalities, but both make cognitive decisions on behalf of humans. Before any of the systems above existed, we thought that it required a real person to recognize a face and the name of a person associated with it. Another important feature about AI systems is that they are more fuzzy than they are deterministic. In the past, a bank analyst would have to check a person’s transaction history and match it with hard thresholds such as having a transaction above a certain nominal and frequency within a day. With enough training examples, an ML model would provide different thresholds for customers with wildly different habits ranging from housewives in Ternate to gig workers in Jakarta. The ability for AI systems to automate human decision making processes the way factories have automated physical work is why AI companies are deemed to be very valuable right now.

Ingredients for a Leading AI Company

Now, what does it take for any country to have their own OpenAI? The answer is a huge capital and supportive ecosystem. Building a world-class AI company is expensive. Companies like OpenAI and Anthropic have raised north of sixty billion dollars and twelve billion dollars respectively to date and have yet to be profitable. This value is larger than the entire Indonesian unicorns valuation summed in their heyday. The enormous amount of money is required because computation cost and talents are expensive. In 2024 alone, OpenAI claimed that they spent around seven billion dollars on computing costs and one and a half billion dollars on manpowers. Despite its simple interface, every prompt entered via ChatGPT will be processed by some computer in a data center that requires electricity and the subsequent cooling that costs some energy and water. Imagine the scale of computing that OpenAI has to deal with four hundred million active users weekly. AI companies should also stock on graphical processing units (GPU) whose price has been steadily increasing over the past few years due to Nvidia virtually being the monopoly for the AI market. As for the talent cost, research in these companies have been driven by graduates from a handful of select US universities. This makes the supply very limited, leading to a high premium of salary for a good employee.

But capital alone is not enough, and I would argue that a supportive ecosystem is an even more important factor for a competitive homegrown AI company to flourish. The current generation of AI companies were not built in a vacuum. They are part of the Silicon Valley boom that previously launched the mobile revolutions and the dot-com bubble. Looking back, the current AI boom is a natural progression of this trend. It was made possible through the synergy between industry, academia, and investors is what makes this combination possible. Similar ecosystems are hardly replicated anywhere around the world, with China catching up quickly in the past decade. Despite having been a major global IT hub for decades, there has yet to be a leading AI company originating from India. One possible reason is the historical lack of investment in RnD. It leads to there not being enough qualified researchers locally who are able to work on the frontier of AI. Financing is another issue. Even when the talent is available, many AI researchers would opt to work overseas instead for a significant pay increase compared to working in their home country. India is definitely doing better than most other Global South countries in terms of advancement in the digital economy. However, the absence of synergy between the three factors above should become a point of consideration for any other countries trying to get into the AI industry.

Learning Points for Indonesia

Reflecting from the above discussion, there is a lot to learn for Indonesia to be a significant player in the global AI scene. The first lesson is to understand that this is a long game. While many of us are only familiar with the term AI in the past few years, seasoned AI researchers are aware that the field itself has experienced multiple boom-bust cycles previously. There were periods where there were unreasonably high hopes on AI research in the 60s and 80s but failed to live up to the expectation both times. This led to AI winters; periods where funding on AI projects was scarce for multiple years. While it means that we cannot be a major contributor now, this does not mean that we should just sit idly waiting for the next boom. Right now, Indonesia should focus on surviving the tech winter while keeping the technical expertise alive. Naturally many workers from tech-first companies were laid off during this period. Any conventional companies or government-related institutions interested in digitalizing their businesses can now hire experienced engineers previously locked by the tech bubble. On top of that, government support for university students to take internships in technology-related fields can keep the pipeline of junior to senior engineers to solve the anecdotal “hiring entry level with 10 years of experience” problem. Additionally, universities should be encouraged to conduct joint fundamental research with companies to deepen the talent pool. Once we have a robust local tech ecosystem, we will have a workforce ready to strike the opportunity when the next AI boom comes.

The second lesson is to find our own niche to our advantage. Nowadays owning the tech alone is not a competitive advantage anymore. China’s DeepSeek reached comparable performance with its competitors for a fraction of the cost thanks to open source model and some engineering ingenuity. Therefore there are two ways of having a moat in this AI race, via data and/or distributions. The latest AI models are having a problem in that they probably have learned from every text data available on the open internet. As models learn better with more novel data, the absence of new data means that models performance will likely plateau. Indonesia could utilize the twenty seven thousand government-related apps as a starting point. They contain localized, context specific data that major AI models have no access to. Hence, it is possible for a home grown AI product to perform better than the global players for tasks that are suitable to the local context. This includes problems like real time traffic optimization for Jakarta’s traffic jams or disease predictions for telemedicine projects serving remote islands in Nusa Tenggara Timur. Initiatives related to data standardization and interoperability between data sources should take priority if we want to get serious about being a major global AI player. Popular apps that are developed by local companies could also support by being the carrier for these AI products as it is already familiar among local users as well as to keep critical AI infrastructure at home. Additionally, online-to-offline (O2O) channels could be deployed to reach users without strong digital literacy skills to be more inclusive towards Indonesia’s diverse population.

Conclusion

Having a cutting-edge AI company in one’s lawn is a major strategic advantage for any country. The benefit goes from potential cooperation in enhancing national security, increasing economic competitiveness via brain gain, as well as establishing soft power in the tech industry. Hopefully the Indonesian AI roadmap addresses the challenges above, to build a supporting ecosystem where local AI companies can thrive and focusing on our inherent strength in terms of data availability and distribution. Most importantly we need to make sure that the roadmap goes beyond box ticking exercise. While the future is difficult to predict, the roadmap could guide Indonesian companies, government executives, and policymakers to make directionally correct decisions.