Big tech companies are deploying compute clusters with millions of GPUs to train and run AI models. But across the entire continent of Africa — encompassing 54 countries and more than 1.5 billion people — fewer than 1,000 GPUs are available for researchers and developers to train models on local-language datasets.
That disparity illustrates what the United Nations’ top digital envoy calls an “immense concentration of tech power and wealth” in a few zip codes — not just countries or regions, but confined areas, primarily in the U.S., where the companies shaping AI are based.
He didn’t name names, but the point hit close to home for the Seattle audience: 98109 for Amazon, 98052 for Microsoft.
Delivering the keynote address via Zoom at Seattle University’s 2026 Ethics and Tech conference on Friday, Under-Secretary Amandeep Singh Gill called 2026 “especially seminal” for AI governance, as the technology shifts from model capabilities and infrastructure investment to systems that perform real-world tasks autonomously.
Gill pointed to the global response to Anthropic’s Mythos AI model — which the company restricted from broad public release over cybersecurity concerns — as an example of why AI governance requires a comprehensive, international approach.
Here are more of the key messages from his talk.
AI could become a “systemic risk.” Gill said the technology is a “relatively minor risk” now but warned it could soon bypass cybersecurity defenses, accelerate armed conflict, and erode public trust through deepfakes and misinformation. “When we cannot tell the difference between what is true and untrue, what is reality or imaginary, then we lose this shared sense of an understanding of facts,” he said.
Armed conflict could worsen. Gill warned that AI risks “lowering the threshold of conflict, confusing accountability under international humanitarian law, and setting us off on escalation ladders that we cannot control.”
AI’s energy demands are threatening climate goals. The energy required for large language models, agentic systems, and inference is already threatening national net-zero targets, Gill warned. Data center emissions, water consumption for cooling, hardware turnover, and mineral extraction costs are compounding — and falling disproportionately on low-income countries.
AI is both “a potential solution and a stressor” for the environment. It could optimize renewable energy grids and accelerate progress in fusion and batteries, but the short-term costs are mounting. Gill said the UN is examining how to ensure equity and just transitions “over these time horizons.”
The UN is building a scientific panel for AI modeled after the Intergovernmental Panel on Climate Change. Chaired by journalist and Nobel Peace Prize winner Maria Ressa and Turing Award-winning AI researcher Yoshua Bengio, the 40-member panel is deliberately composed of only two members each from China and the U.S., with the remaining 36 from other countries, including seven from Africa, to ensure more countries are heard. Its first report is expected in July 2026.
The UN is putting AI governance conversations under one roof. Conversations about AI previously happened in separate bodies with narrow mandates. Now they’re being brought onto what Gill called a “horizontal platform” where policymakers from all 193 countries can learn from each other and develop common approaches.
Gill called AI governance a “sovereign decision.” The UN won’t tell countries how to regulate AI, but governance frameworks mean little if nations lack the capacity to participate. Gill called for support of community-driven AI projects that invest in local research and innovation ecosystems, allowing people to use these tools to solve their own problems.
He acknowledged the UN is working with limited resources against an enormous challenge, but said the alternative is leaving AI’s trajectory to market forces and geopolitical competition.
The goal, he said, is a world where AI empowers democracies and societies, and creates opportunities not just for “a few billionaires and trillionaires” but for everyone.
Read the full article here
