Countries seeking to benefit from artificial intelligence may need to invest beyond data centres and electricity into laboratories, scientific equipment and robotics, as advances in AI increase demand for physical research capacity.
Jack Clark, co-founder of Anthropic, said the next phase of AI development could be constrained by the availability of facilities capable of testing discoveries generated by increasingly capable AI systems.
Speaking at the 2026 Per Jacobsson Lecture in Bangkok, Clark argued that governments risk repeating the mistakes made during the expansion of computing infrastructure if they fail to anticipate the demand for physical scientific capacity.
He said earlier warnings about the need for more computing power and electricity had not received sufficient attention from many policymakers, contributing to the concentration of data-centre capacity in the United States, China and an expanding group of Middle Eastern markets.
The next constraint, Clark said, could emerge when AI systems increasingly conduct scientific experiments, operate laboratory equipment and work with robots to develop new materials, medicines and energy technologies.
UNESCO estimates cited by Clark put the world’s scientific workforce at about 10 million people. He argued that existing research infrastructure is already insufficient to support the work scientists want to undertake, including materials analysis and the development and testing of new drugs.
The challenge could become more pronounced if AI enables researchers to deploy large numbers of digital research assistants simultaneously.
“An avalanche of proposed new drugs is of little use if only a handful can ever be tested,” Clark said, highlighting the risk that discoveries could accumulate faster than laboratories, clinical trials and production facilities can process them.
That creates an investment opportunity for countries that build scientific testing facilities, expand robotics capacity and strengthen supply chains serving the research economy.
Clark said developed countries with established scientific institutions could use AI to increase researchers’ productivity, while developing economies could participate by supplying raw materials and other inputs or building facilities that combine human expertise with AI-driven research.
The opportunity, however, depends on more than infrastructure spending. Governments must also address the regulatory and practical barriers that prevent discoveries from reaching clinics, factories and farms.
Clark called for measures to accelerate the movement of scientific advances from laboratories into commercial applications and to make those gains accessible across national borders.
His argument points to a possible shift in the competition for AI-related investment. As computing capacity expands, the countries best positioned to benefit may increasingly be those able to turn digital discoveries into tested products, industrial processes and services.
