AI Debate Map

Does AI concentrate or distribute power?

Training a frontier AI model costs hundreds of millions of dollars and requires thousands of specialised GPUs that only a handful of companies can afford. That sounds like centralisation. But the same technology lets a solo entrepreneur in Penang compete with a Fortune 500 marketing department. That sounds like distribution. Which force wins determines whether AI creates the most democratic economy in history or the most concentrated.

Where They Stand

AI is a centralising force

Yuval Noah Harari has argued in multiple books, lectures, and interviews that AI represents the most powerful centralising force in human history. His core thesis: whoever controls AI controls the flow of information, and whoever controls information controls political power. He has warned that AI could enable "digital dictatorships" that understand citizens better than they understand themselves, making meaningful democracy impossible. Unlike previous technologies that centralised then democratised (the printing press, the internet), Harari argues that AI's resource requirements — massive compute, massive data, massive talent — create natural monopolies that are structurally impossible to democratise. Tristan Harris, through the Center for Humane Technology, extends this argument with a technology-specific lens: the attention economy already concentrated power in five companies, and AI amplifies that concentration by orders of magnitude. His "AI dilemma" presentation argued that AI tools give these companies unprecedented ability to shape human behaviour, opinion, and decision-making at scale. Connor Leahy argues that AI development is converging toward a winner-take-all dynamic: the lab that achieves the most capable model attracts the most capital, talent, and data, which makes the next model even more capable, in a compounding loop that smaller players cannot enter.

Yuval Noah Harari→Tristan Harris→Connor Leahy→

AI is a democratising force

Sam Altman has consistently framed AI as "the great equaliser" — a technology that gives everyone access to capabilities previously reserved for the wealthy and well-connected. He argues that a student in Lagos with ChatGPT now has access to tutoring, legal advice, and business coaching that previously required expensive professionals. OpenAI's pricing strategy (free tier, relatively cheap API) reflects this philosophy, though critics note the company is also building one of the most valuable monopolies in history. Mark Zuckerberg's open-source Llama strategy is explicitly framed as power distribution: by releasing capable models for free, Meta ensures no single company can gatekeep access to AI. His argument is that open-source has always been the great decentraliser — Linux broke Microsoft's OS monopoly, Android broke Apple's mobile monopoly, and Llama will break the closed AI monopoly. Andrew Ng focuses on the developing-world angle: AI enables small teams in emerging markets to build products and services that compete with large Western companies. His DeepLearning.AI and AI Fund investments specifically target making AI accessible to non-tech-native businesses. All three argue that while AI development may be concentrated, AI deployment is inherently distributed.

Sam Altman→Mark Zuckerberg→Andrew Ng→

It depends on policy choices

Mustafa Suleyman argues that AI is neither inherently centralising nor democratising — it's an amplifier of existing power structures. In societies with strong institutions, competitive markets, and distributed power, AI will distribute further. In societies with weak institutions and concentrated power, AI will concentrate further. His book "The Coming Wave" dedicates substantial chapters to the "containment problem": the challenge of ensuring that AI's power is channelled through democratic institutions rather than around them. Fei-Fei Li has emphasised that the trajectory depends on deliberate choices about who participates in AI development. Her work at Stanford HAI focuses on ensuring diverse voices — not just Silicon Valley engineers — shape how AI systems are built and deployed. She argues that public investment in AI research, open datasets, and AI education are essential counterweights to corporate concentration. Kai-Fu Lee, who has built AI businesses in both the US and China, offers a geopolitical lens: AI is centralising power at the NATIONAL level, with the US and China pulling away from every other country. In "AI Superpowers," he argued that countries without sovereign AI capability will become digital colonies of those that have it — a warning particularly relevant for Southeast Asian nations caught between the two superpowers.

Mustafa Suleyman→Fei-Fei Li→Kai-Fu Lee→

Patrick's Take

I see both sides of this every single week. On Monday, I'm training a 5-person startup in Cyberjaya that's using Claude to do the work of a 20-person team — that's democratisation in action. On Tuesday, I'm watching that same startup pay RM 3,000/month to OpenAI and Anthropic — money flowing to American companies for intelligence that used to be created locally. Here's what the power debate looks like from Malaysia specifically: we are a consumer of AI, not a producer. We don't train frontier models. We don't manufacture GPUs. We don't have the talent pool to build an alternative. That means every Malaysian business using AI is, at some level, dependent on decisions made in San Francisco, Mountain View, and Shenzhen. If OpenAI changes its API terms tomorrow, Malaysian businesses have no recourse. But — and this is what I emphasise in my training — the application layer is where the power redistributes. Knowing HOW to use AI effectively is a form of power, and that power is accessible to anyone who invests time in learning. The Malaysian businesses winning right now aren't the ones with the biggest AI budgets. They're the ones whose people understand the tools best. That's why AI training matters — it's the equaliser within the equaliser.

What This Means for Your Business

The centralisation question has a very concrete business implication: vendor dependency. If your business relies on a single AI provider, you are centralised by proxy. Build abstraction layers: standardise your AI workflows so you can switch between OpenAI, Anthropic, Google, and open-source models with minimal friction. This isn't just good engineering — it's strategic resilience. Second, consider where your competitive advantage actually lives. If it's just "we use AI" — that's not a moat, because everyone will use AI. If it's "we use AI plus deep domain expertise in Malaysian [industry]" — that's a moat that AI concentration can't erode. The companies that will be squeezed are the middlemen who add no value on top of the AI layer. Make sure you're not one of them.

What to Actually Worry About

For Malaysian businesses, the power question you should lose sleep over isn't philosophical — it's practical. Malaysia has no sovereign AI model, no GPU manufacturing, and limited AI research infrastructure. That means the AI layer your business depends on is entirely controlled by foreign companies operating under foreign laws. If the US restricts AI exports (as it has with chips to China), Malaysia could find itself cut off or deprioritised. The ASEAN AI governance framework is still nascent. The mitigation is straightforward: diversify your AI toolkit across American, Chinese, and open-source models. Build internal expertise so you're not dependent on any single provider's ecosystem. And advocate for Malaysian investment in AI infrastructure and talent — this is a national competitiveness issue, not just a tech debate.

Featured Minds in This Debate

ThinkerCautious

←Back to AI Minds Directory