AI Chips Run Hot, but Servers in Indigenous Communities Stay Cold

The AI industry is entering an arms race for infrastructure. Data centers, GPUs, power, cooling, networks, and model training costs have become the new battlegrounds for tech giants. Everyone isn't just competing to see whose models are smarter; they're also racing to build bigger server farms, acquire more chips, and burn through more electricity.

Tech companies love to talk about "the cloud." The word sounds light, clean, floating—as if knowledge is no longer bound by the ground. But clouds actually live somewhere. They have server rooms, cooling water, substations, land, noise, carbon emissions, and political negotiations.

AI is even more so. Every seemingly effortless generation connects to chips, memory, data centers, and power systems. The so-called model economy isn't just a software economy; it's a hybrid of hardware, energy, finance, and geopolitics.

The black humor is that we thought AI would make knowledge lighter, but instead it made infrastructure heavier.

This matters especially for Taiwan. Taiwan is one of the global cores of the semiconductor supply chain and is expected to become a hub for AI hardware and servers. But if Taiwan only talks about chip exports, server output value, and the "AI island" vision without addressing public services and cultural infrastructure, a strange picture emerges: we train models for the world but don't prepare our own Indigenous schools with enough digital textbooks and network capacity.

Indigenous regions and communities don't necessarily need the biggest, most expensive, most power-hungry large models. What they really need are tools that can be governed, maintained, localized, and explained.

Language teachers need tools to organize corpora, generate practice questions, and assist with pronunciation annotation. Cultural workers need systems to manage image licensing, tag place names, and map relationships between people. Local media need tools to help verify information, organize meeting minutes, and produce multilingual summaries. Long-term care and medical sites need auxiliary systems that respect privacy, support local languages, and avoid erroneous advice. Young creators need the ability to use AI without having their creative rights drained by platforms.

These needs don't necessarily have to be solved with giant models. They might require small models, edge computing, community databases, offline tools, licensing management, cultural review processes, and training programs. If public AI policy only obsesses over "how many GPUs we have," it's like sending a fighter jet to deliver lunch boxes. It looks advanced but doesn't actually solve hunger.

When governments talk about AI infrastructure, they often place budgets on computing power, industrial parks, talent training, cybersecurity, and data centers. But if AI is truly to enter the public knowledge system, there must be cultural budgets.

Cultural budgets aren't just for events, image films, or having AI generate a few pretty visuals. Cultural budgets should include Indigenous data governance systems, tiered licensing for language and traditional knowledge corpora, maintenance of Indigenous community digital archives, co-design of cultural AI tools, training in AI media literacy for Indigenous teachers and youth, community-controlled backup and data security capabilities, and reasonable compensation for cultural consultants and reviewers.

Many countries are now talking about AI sovereignty. But for Indigenous Peoples, discussing only national-level AI sovereignty isn't enough. Because the state can also become a data concentrator. True cultural data sovereignty must be delegated to communities, allowing Indigenous communities, Indigenous organizations, and rights holders to participate in governance.

If AI infrastructure only makes large corporations earn money faster, that's industrial policy. If AI infrastructure enables primary school teachers, language keepers, cultural workers, local journalists, and Indigenous young people to create and govern knowledge, then it's public policy.

Chips are hot, models are expensive, data centers are impressive. But a society's level of civilization isn't measured by whether it can train the biggest model; it's measured by whether it's willing to include even the smallest voices in its infrastructure.