Taiwan The Country Making AI for the World With Adoption Ambitions
Taiwan assembles more than 95% of the world’s AI servers. Its businesses rank approximately 20th globally in AI adoption. More than 70% of Taiwanese companies have not integrated AI into their operations.
Business leaders and policy makers need to note that proximity to AI infrastructure is not the same as AI readiness to adopt.
I spoke recently with Sega Cheng, Co-Founder and Chairman of iKala, a Taiwan-based company that has been building AI solutions for enterprise clients across Asia since 2012. Sega came to this work from Google, where he spent five years as a software engineer working on Android, Google Maps, and Google Search, and was among the engineers who introduced machine learning into Google’s products at scale. He holds a Stanford Master’s in computer science. He is not someone who mistakes AI for magic. The conversation covered Taiwan’s hardware paradox, the structural barriers to enterprise AI adoption across Asia, AI sovereignty, and a thesis about the future of software that has direct implications for business leader thinking about where AI value will accrue.
The Hardware Paradox That Every Organisation Should Study
Taiwan’s position in the global AI economy is without precedent. The country manufactures over 90% of the world’s AI chips and assembles over 95% of its AI servers. Companies such as TSMC and Media Tek sit at the centre of the global AI supply chain. And yet, according to a Microsoft AI Economy Institute report Sega cited from earlier this year, Taiwan ranks approximately 20th globally in AI adoption, with a generative AI diffusion rate of 32%.
Sega describes this as a gravitational problem. The hardware industry is so dominant, so resource-intensive in its demand for talent, energy, water, and land, that it has crowded out the conditions in which a software culture could develop. Taiwan, he argues, makes the picks and shovels for the global AI gold rush while still learning how to mine within itself.
80% of AI Adoption Is Data Plumbing, Not Transformation
Sega made an observation that cuts through a considerable amount of AI adoption rhetoric: 80% of the effort required to adopt AI goes into data collection, data cleansing, and data infrastructure. Only 20% touches algorithms, models, or the capabilities that tend to make headlines.
This matters for how organisations frame their AI programmes. The conversation leaders are having in many boardrooms focuses on which AI model to deploy, which vendor to select, which use case to prioritise. Sega’s argument is that those questions are almost always the wrong ones. If the data is not in order, and in most organisations it is not, no model will deliver meaningful results. AI adoption is, in his words, data plumbing. And that work is unglamorous, slow, and expensive.
His practical prescription follows from this. For organisations still hesitant about where to begin, he recommends starting in marketing. Not because marketing is the most strategically significant place to use AI, but because it is the place where results are visible and attributable quickly. A two-point improvement in audience targeting accuracy, a 20% increase in social media engagement: these are numbers an organisation can see and act on. They provide the internal evidence base that makes broader investment in AI adoption credible. This is the argument against transformation as the first move. Transformation without evidence is simply risk. Transformation with a track record of small, visible wins is something organisations can build on.
iKala’s own platform illustrates this approach. Launched in 2018 and now tracking data from over 300 million influencers and creators, the platform was built before generative AI became mainstream. When the technology arrived, iKala faced the same choice every organisation faces: rebuild from scratch or layer AI on top of the existing system. Sega was direct about this. Rebuilding a mature software-as-a-service platform from scratch is not a realistic option for most organisations. The layered approach is the rational first step in an ongoing process. The AI-native version comes later, built in parallel, migrating data and workflow incrementally until the old system is eventually replaced.
Consider Software Soft: The Argument Every Business Leader Needs to Hear
Sega introduced a concept he describes as his working motto: ‘consider software soft.’ The argument is this: AI coding tools, used by engineers across the industry, are collapsing the cost of producing software towards zero. Big tech leaders are publicly reporting that 50% or more of their code is now written by AI. Sega believes software production will effectively be a solved problem within three to five years.
If that is correct, and the trajectory of the last two to three years suggests it is not an unreasonable claim, then the source of competitive advantage in software-dependent industries shifts fundamentally. The value is no longer in the code. It is in the sector-specific application of that code: the deep understanding of what a manufacturing line needs, what a healthcare system requires, what a financial services compliance process demands. Organisations that have spent decades building domain expertise now have a structural advantage, provided they can combine that expertise with the AI capability to act on it.
This is directly relevant to Taiwan’s hardware giants. Sega noted that many of them are actively exploring software and platform businesses to capture what he calls the compounding value of hardware and software together. The barrier is not technical. It is cultural and structural: the OEM manufacturing mindset, which prizes precision, specification, and delivery against exact requirements, can be ill-suited to the iteration, uncertainty, and continuous reinvention that software and AI product development demands. Taiwan’s hardware leaders know this. Closing the gap is a different matter.
Intelligence as Infrastructure: The Sovereignty Argument
The conversation’s sharpest argument came towards the end. Sega framed AI sovereignty not as a geopolitical preference but as an infrastructure necessity. His analogy: water and power are utilities so fundamental that no country or organisation would tolerate indefinite dependency on another entity for their supply. Intelligence, in his view, is the third utility of human civilisation. It is general purpose, increasingly cheap, and increasingly essential to every significant activity that organisations and states undertake.
The implication is that the question of who controls AI infrastructure, who owns the data centres, who manufactures the chips, who controls the models, is not a technology question. It is the same category of question as energy security or water security. Countries and enterprises that treat it as optional are making a category error.
This argument has a direct consequence for organisations in the UK and Europe. The conversation in Westminster, which I have been part of recently, is beginning to engage seriously with sovereign AI capability. The question is whether the pace of that engagement matches the pace at which dependencies are being built.
The central question this episode raises but does not resolve is one that applies equally to nations and to individual organisations: does hardware advantage translate into AI advantage, or does it simply create the infrastructure on which others build their advantage? Taiwan’s position suggests the answer is not automatic. The missing ingredient is not capability. It is the mindset, the culture, and the willingness to operate under conditions of uncertainty that AI adoption at scale demands. The organisations that develop those conditions, regardless of their proximity to the hardware, are the ones that will compound the value.
Listen to the full conversation:
Comments