American Tech Firms Drive 200 Million AI Chips by 2028 as Spending Nears $1 Trillion
Updated
Updated · The New York Times · Jul 29
American Tech Firms Drive 200 Million AI Chips by 2028 as Spending Nears $1 Trillion
3 articles · Updated · The New York Times · Jul 29
Summary
Hundreds of new data centers from the U.S. Midwest to the Persian Gulf are set to come online in the next few years, pushing global AI chip capacity from about 20 million today to roughly 200 million by end-2028.
That surge is being driven by faith in AI “scaling laws” and an industry arms race: Amazon, Google, Microsoft, Meta and Oracle are projected to spend about $750 billion this year on AI infrastructure, up from roughly $400 billion last year.
The build-out is meant to expand both model training and real-time inference, with researchers and companies expecting more drug discovery, robotics advances and wider use of AI agents in everyday work.
Costs and risks are rising alongside capacity: data centers used 64 gigawatts of electricity globally last year, a figure expected to quadruple by 2030, while communities protest water use, power prices and environmental damage.
U.S. companies already control about 80% of global AI computing power, far ahead of China, deepening geopolitical divides even as economists warn the boom could resemble past infrastructure bubbles before profits fully arrive.
With AI data centers draining global power and water, will environmental limits trigger a sudden collapse in the AI arms race?
When AI facilities consume most global memory chips, how will other critical industries survive this unprecedented semiconductor drought?
As tech giants burn billions on infrastructure in 2026, could this historic spending spree lead to a catastrophic financial bubble?
The 2026–2027 AI Infrastructure Explosion: Capital Expenditures, Chip Bottlenecks, and the Global Resource Crisis
Overview
In 2026, tech giants like Alphabet, Amazon, Meta, and Microsoft are set to spend a record $725 billion on AI infrastructure, causing their free cash flow to drop to decade lows and sparking investor anxiety. As AI workloads shift from training to large-scale inference, these companies are moving from general-purpose GPUs to custom ASICs for better efficiency, with Broadcom and Marvell dominating the custom chip market. Meanwhile, supply chain bottlenecks—such as TSMC’s limited 2nm chip capacity and memory suppliers prioritizing high-bandwidth memory—are driving up costs for consumer electronics. This massive infrastructure buildout is also fueling a surge in global data center electricity and water use, making it harder for tech firms to meet their climate goals.