McKinsey & Company’s report highlights the widespread unease about AI. It points to the staggering sums invested in infrastructure to support it while warning that forecasts for future demand are based largely on guesswork.
The boom of AI investment has accelerated over the last year, to the point that McKinsey estimates datacenters that can handle AI workloads will require $7.9 trillion to keep up with demand by 2030.
The report admits that no one is sure of the level of AI demand. Or, as the report says: “A lack of clarity about future demand makes precise investment calculations difficult.”
McKinsey questions whether hyperscalers will continue to shoulder the cost burden or if enterprises, governments and financial institutions will step in with new funding models. Will the demand for datacenters rise as a result of a surge in AI use, or will it decline as technological advances make AI less resource intensive?
The report leaves out one question: What if AI is only useful for specific tasks and not the magic bullet many corporate leaders think will automate their business processes, allowing them to save money – primarily by eliminating human workers?
A working paper published earlier this month found, for example, that generative AI had not had a significant impact in earnings or recorded hours of any occupation to date, despite billions being poured into the building and training of the models.
The report, which focuses on actual numerical projections, predicts that the global demand for computing capacity could almost triple by 2030. About 70 percent of this demand will come from AI workloads. This depends on two things: whether AI can be used to make a real impact in the business world, and if technology advances can improve compute efficiency.
While the first scenario would increase demand for infrastructure, the second one would tend to reduce it. This is unless, as McKinsey argues, any efficiency gains will be offset by increased usage in the broader AI industry – otherwise known the Jevons paradox.
In order to cover themselves, the company has envisioned three scenarios by 2030, ranging from “constrained demand” to “continued demand” and “accelerated demand.” . The first scenario would add 78 GW for a capex total of $3.7 trillion. The mid-range scenario would add 124 GW for a capex total of $5.2 trillion. (Capex for traditional IT applications would be $1.5 trillion more, resulting in the $7 trillion total expenditure mentioned in the title.) The demand for AI is expected to increase by 205 GW, at a cost $7.9 trillion.
- According to economists, AI does not replace jobs or affect wages in any way
- Tech hires are on hold as AI hype, tariffs and economic uncertainty collide
- Following the Copilot trial, Microsoft’s AI was rated less useful by government staff
- IT decision-makers are not convinced about AI ROI
However, McKinsey says that current investment levels of potential end customers lag behind these projection It claims that dozens of interviews with clients revealed that CEOs were reluctant to invest in computing capacity at the maximum level, as they had limited visibility into future demand. They are unsure if large capital expenditures on AI infrastructure will produce a measurable return in the future.
This does not stop big money from pouring in to datacenter buildouts. DLA Piper’s report last year revealed that 70 percent of financiers and consultants expect AI datacenter funding to continue to increasein spite of growing concerns over the availability of electricity to power these projects.
According to Digital Realty senior vice president Fabrice Coquio, The Register, this month, one reason is because the projected returns for the bit barn sector are favorable in comparison to other areas of economy. This is attracting investors who have no prior experience. Coquio described it as “a typical bubble.”
McKinsey recommends that companies assess AI computing needs as early as possible, anticipate potential changes in demand, and develop scalable investment strategies which can adapt to AI models and use-cases as they evolve. It says that striking the right balance between capital efficiency and growth will be crucial. (r)
