# AI Raises Questions Over Data Center Economics > Source: > Published: 2026-05-27 16:21:06.434300+00:00 # AI Raises Questions Over Data Center Economics Global Research publishes a critical piece arguing that the ongoing data center buildout for AI carries hidden economic and energy costs. The article reports that compute and token costs make many AI uses expensive, citing claims such as an MIT analysis finding AI cost-effective in about **23%** of basic jobs and alleging heavy-user bills up to **$113,000** (Global Research). The piece also reports that Global Research attributes a statement to an unnamed Nvidia vice president that compute costs for engineering teams exceed employee salaries, and it claims Microsoft has banned AI use by its engineers and that global AI spending has topped **$6.7 trillion** (Global Research). The article frames these items as signs of a potential AI-driven asset bubble. Editorial analysis in this summary highlights that many of the article's high-stakes figures are reported by Global Research and require independent verification. ### What happened Global Research published an opinion piece arguing that the current AI data center expansion hides substantial economic and energy costs. The article reports several high-stakes claims, attributing to Global Research that an "economic analysis by MIT" found AI is cost-effective in about **23%** of basic jobs and that heavy token usage can lead to bills as high as **$113,000**, while other heavy users were said to burn **$500 to $2,000** per month (Global Research). The article also reports that Global Research cites a statement from Nvidia's Vice President of Applied Deep Learning saying compute costs for their AI engineering teams far surpass employee salaries, and it asserts that Microsoft has banned the use of AI by its engineers and that global AI spending has reached over **$6.7 trillion** (Global Research). ### Editorial analysis - technical context Industry-pattern observations: data center growth, higher-performance GPUs, and larger inference and training workloads have driven both capital expenditure and operational energy use across the sector. Companies and research reports outside this article have repeatedly flagged compute and energy as first-order cost drivers for large models. For practitioners, this means cost modeling must include not only GPU-hour prices but also systems engineering, networking, storage, and tokenization behavior when projecting production expenses. ### Context and significance Industry context: public debate about AI infrastructure costs is long-standing, spanning cloud egress, on-prem amortization, and model-efficiency workstreams. If even a subset of the reported figures are accurate, they would raise questions about the marginal economics of deploying large foundation models for low-value tasks. That said, the specific numerical claims in the Global Research piece are reported by that outlet and are not independently sourced within the article; readers should treat the numbers as assertions pending verification from primary sources. ### What to watch Observers should look for independent confirmation of the article's high-impact claims: the purported MIT analysis and its methodology, any public statement or filing from Nvidia or Microsoft that matches the reported assertions, and aggregate industry spending data from financial disclosures or market analysts. Industry-pattern observations: transparency in cloud unit economics, improved model efficiency (quantization, distillation, retrieval-augmented methods), and workload right-sizing are typical mechanisms that reduce the per-task compute burden over time. ### Bottom line The Global Research piece aggregates sharp criticisms and striking figures about AI infrastructure costs. Many of those figures are presented as reported by Global Research and require corroboration. For practitioners, the report is a reminder to maintain rigorous cost tracking and to treat third-party claims as starting points for verification rather than settled facts. ## Scoring Rationale The story raises relevant infrastructure and cost questions that matter to ML operations and budgeting, but it is based on a single opinion piece with unverified numerical claims, so its immediate technical impact is moderate. Practice interview problems based on real data 1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with. [Try 250 free problems](/problems)