Who speaks for 2 billion Muslims in the age of AI?

Last month, Pope Leo XIV used his first encyclical, “Magnifica Humanitas,” to tell 1.3 billion Catholics what artificial intelligence means for the human person — who it threatens and where its limitations should lie. No single authority can do quite the same for the world’s 2 billion Muslims. Islam has great scholars and institutions, but none are positioned to issue a binding text on a new technology the way a papacy can. Into that silence has stepped an improbable authority: the US government.

This is a cultural story as much as a geopolitical one. The systems now mediating how billions of people search, write, learn, and are in turn monitored and assessed were trained overwhelmingly on English-language data. Arabic, the fifth most spoken language on earth, is chronically underrepresented in that training. Worse still, researchers have repeatedly documented content-moderation systems flagging ordinary text written in Arabic as extremist or dangerous, at rates with no English equivalent. The technology, far from being neutral, encodes a particular worldview, and the Islamic world, with its centuries-deep frameworks, has been given no seat at the table where that worldview is being built.

Further, the table itself is built on hardware controlled by Washington. Last week, the US Commerce Department issued sudden guidance closing a year-old loophole that had allowed subsidiaries of Chinese companies operating outside China to acquire the world’s most advanced AI chips from Nvidia and AMD, including Nvidia’s Rubin and Blackwell chips, as well as AMD’s MI350X chip. Industry estimates suggest that hundreds of thousands of chips may have reached Chinese subsidiaries through this regulatory gap. With a single notice, Washington reasserted what it has long implied: the authority to police where computing power may flow, regardless of whose soil the data center stands on.

Sarawak, the Malaysian state on the island of Borneo, offers one of the clearest illustrations of what that authority means in practice. Malaysia has been one of the major beneficiaries of the hardware loophole, and the state of Sarawak has embarked on an ambitious experiment, powering hyperscale data centers not by gas but by hydropower and solar energy.

Anchored to a homegrown Digital Economy Blueprint and a state AI center, the system is marketed to the region as secure, green, and ethically governed compute. In miniature, it is exactly the kind of self-determination the Muslim world says it wants — sovereign infrastructure on sovereign terms, powered by clean energy. Yet, like every ambitious builder in this field, Sarawak still depends on hardware and platforms designed elsewhere, much of which is now governed by rules written in Washington.

Sarawak is one of a number of examples of Muslim states engaging in this ambitious bid for technological self-determination. The Gulf has set the pace, and Saudi Arabia’s Humain, backed by more than $100 billion in planned investment, is building a sovereign Arabic-language model. The model, named “ALLaM,” is being trained on the largest Arabic dataset ever assembled, in an attempt to ensure the next civilizational technology is engineered in the Kingdom rather than rented from abroad. The UAE’s G42 has also anchored its own full-stack compute and a web of international partnerships.

Turkiye has taken the same idea in a distinctive direction. Ankara now treats AI not only as a sectoral tool but also as core state capacity. Its 2026 presidential program places AI at the heart of public administration, strategic autonomy, and defense planning, embodying the logic of a confident middle power that intends to own its stack rather than rent one. It is funding a sovereign Turkish large language model through its national development plan, with dedicated state grants for homegrown models and domestic AI hardware. The system is being built to think natively in Turkish, with a reach wide enough to serve the larger family of Turkic-speaking peoples. The country that showed the world a middle power could build its own drones and aircraft is now applying the same drive to AI.

None of these are vanity projects. Each is a deliberate refusal to let the next civilizational technology be rented from abroad.

However, the chips that make all of it possible still arrive preconditioned. Successive US rules attach themselves to the hardware, including American data-governance expectations and oversight terms. A sovereign model trained on dependent silicon is sovereign only until the export guidance changes on a Sunday afternoon.

The most forward-thinking governments understand this, and they are already engineering their independence: ensuring federated compute that spreads training across jurisdictions, sovereign clouds that keep data and weights under domestic control, and data-embassy arrangements that borrow the logic of diplomatic immunity for servers.

On this one, Sarawak is ahead of the game. In May, the state’s own AI center and Swinburne’s Kuching campus used federated learning to fine-tune an open model to answer in Sarawak Malay, a local tongue global systems tend to flatten into standard Malay or simply answer in English. The training for the system was spread across a cluster of off-the-shelf gaming GPUs, and neither the data nor the model ever left the campus network. Its authors describe it as the first time the full loop, training and serving alike, has been run end to end on a Malaysian state’s own infrastructure under its own data-sovereignty rules.

Rather than data-center accelerators, Sarawak’s system used consumer cards, showing that a community AI model in its own language is achievable for far less than a $100 billion. This is not a complete escape from using external hardware, but it is a signal that the leverage Washington exercises is understood in capitals from Riyadh to Ankara to Kuching as a problem to be solved rather than a permanent fact of life.

This raises the challenge Washington has not answered, and the one this piece puts to it. While American control over the AI stack is real, it is also finite. Every time the US uses this leverage, it teaches 2 billion people the same lesson: that the infrastructure of their future is held at someone else’s discretion, and that the only durable response is to build their own. The US can keep its hand on the tap. What it cannot do is decide how long the rest of the world will agree to drink from it.

The pope could speak for his flock because the questions AI poses are, at the end of the day, moral questions about dignity, labor, truth, and power. The Muslim world has no shortage of answers to those questions. What it has lacked is the compute, the language data, and the seat at the table to make those answers count. It is now, visibly and expensively, acquiring all three. The only open question is whether Washington grasps that its leverage has an expiry date — and what it will have taught by the time that date arrives.

• Muddassar Ahmed is president of the Concordia Forum and managing partner of Unitas Communications.