Your ISP Was Not Built for This

I keep noticing a gap between what governments say about AI and what their telecom infrastructure can actually support. They want AI adoption. But the pipes underneath are priced for streaming and browsing, not for what AI work actually demands.

The average US household now uses around 665 gigabytes of broadband per month. That figure is already straining older data cap thresholds. But it was built on the usage patterns of pre-AI internet: video calls, Netflix, social media, file downloads.

AI work looks different. Every time you run a multimodal task, query a large model with image or audio input, or work across multiple AI tools simultaneously, you are pulling and pushing more data per session than most people did in an entire workday five years ago. The average session time on ChatGPT is around 13 minutes. That is fine for text. But voice AI, video generation, and real-time collaboration with AI systems are not text-level in data terms.

The infrastructure is behind the ambition

Governments across the world are pushing AI adoption as a national priority. The UAE launched the world's first Ministry of State for Artificial Intelligence in 2017. Saudi Arabia is building data center capacity at pace. The EU has its own frameworks and investment tracks. The US is spending at a scale that is hard to comprehend.

But none of these strategies address the last mile. They focus on compute: data centers, GPUs, model training. The connectivity question is treated as solved. It is not.

In 2025, Comcast announced it was phasing out data caps on its new national broadband plans. But Comcast is one ISP in one country. And even their move came after years of being behind usage trends, not ahead of them. The FCC's own data shows what caps do to access. The pattern is consistent.

Data caps were designed for a world where the average user browsed, streamed, and checked email. That world is gone. AI productivity is not a browsing pattern.

The mobile problem in the Middle East

This tension is sharpest in markets where mobile is the primary connection. The GSMA estimates that 308 million people in the Middle East and North Africa are connected to the mobile internet. For many of them, mobile is not a backup option. It is the only option.

The UAE leads global AI adoption among working-age populations, with around 64% of them using AI. Singapore is at 61%. These are small, wealthy, highly connected countries. They can absorb the infrastructure question because they have capital to invest in it.

But much of the region is not the UAE. Mobile data packages in large parts of the Middle East and North Africa are still priced on pre-AI usage models. A heavy AI user running several tools across a workday will hit a monthly cap that was designed for someone streaming video in the evenings.

I am genuinely not sure how much AI work costs in data terms in a country where mobile is your primary connection and your plan caps at 30 or 50 gigabytes per month. I have not seen that number calculated anywhere.

This is not a convenience issue

This is not about whether your video call drops occasionally. It is about whether AI-assisted work is economically accessible to workers and businesses in countries that have not aligned their data infrastructure with the shift.

If a professional in a country with restrictive data pricing has to ration AI tool usage to avoid overages, they are not competing on the same terms as someone in a country with abundant, cheap broadband. That gap compounds over time. Productivity gains from AI are not evenly distributed by default. They follow the infrastructure.

There is a version of this that becomes a policy crisis. Countries that push AI adoption in schools, government, and industry without addressing the data cost of AI work at the consumer level will find that the adoption curve flatlines. Not because people do not want to use AI. Because they cannot afford the data bill.

Who is thinking about this?

I read about the G42 Stargate project in the UAE, a planned 1 gigawatt AI campus. Microsoft has committed $15.2 billion to UAE AI infrastructure. MGX is targeting $100 billion in AI assets under management. These are compute investments.

I cannot find the equivalent investment in consumer and small business connectivity infrastructure that is specifically priced for AI workloads. It may exist. But I have not seen it framed that way.

The question I keep returning to is simpler than it sounds. Governments have AI strategies. Telecoms have pricing strategies. Do those two things talk to each other? And if they do not, who pays the cost?