Big Tech AI Spending Heads Toward $1.3T as Free Cash Flow Falls 84%, Hyundai Flags 2028 Correction Risk

The number that stands out in a new Hyundai Research Institute report is $32.7 billion. That is the combined free cash flow of Alphabet, Amazon, Meta, Microsoft and Oracle over the past four quarters, down about 84% from $201.1 billion a year earlier. The same five companies are on course to spend far more on AI than they do now.

The institute released the report on the 15th under the title "We Must Prepare for a Possible Correction in the US AI Investment Boom." It judged the sustainability of AI investment against five conditions: demand and profitability, financial health, financing, regulation, and infrastructure and supply chains. Its conclusion is that the spending cycle could enter a correction around 2028. It does not say AI demand has peaked. It says the cash and infrastructure needed to sustain the spending are nearing their limits.

The scale of the spending is large. AI-related capital expenditure by the five companies is projected at $789.6 billion this year, about 1,055.1 trillion won. By 2029 it is forecast to reach $1.3079 trillion, about 1,747.6 trillion won, a 65.6% rise in three years.

Whether that spending turns into profit is still unclear. The US Census Bureau reports that only 24.6% of American companies were using AI as of last month. Of those that had adopted it, 57% applied it to three or fewer business functions. Data center and semiconductor investment is running ahead of how broadly companies actually use the technology.

Cash flow is where the strain shows most. Capex has reached 95.5% of operating cash flow, so the companies are putting roughly everything their operations generate back into construction and equipment. In the third quarter, all four companies other than Microsoft posted negative free cash flow.

Outside funding is filling the gap. The five issued $182 billion of corporate bonds from January through October, about double the $93.25 billion issued in all of 2025. Borrowing costs are starting to differ by company. Microsoft's recently issued bonds yield 0.65 percentage points over the 10-year US Treasury. Oracle's spread has more than doubled, from 1.45 points in February to 3.10 points this month. The report reads that as a sign funding conditions are diverging by credit quality.

Physical limits add to the pressure. About 70% of announced US data center capacity has not yet broken ground. Of total planned capacity, 38.4% is concentrated in Texas, Georgia and Ohio, where the prospect of tighter regulation has come into focus since this month's midterm elections. Power shortages and rising prices for components such as memory chips add to the cost of new projects.

The institute does not expect an immediate pullback. Cloud revenue is growing and investment plans are already locked in, so expansion should continue through next year. By 2028, it expects delayed profitability, funding burdens and regulatory and infrastructure constraints to converge. Companies would then postpone or trim new spending case by case. The institute compared this to the way investment in earlier general-purpose technologies such as the internet adjusted to real demand and profitability.

The stakes are direct for South Korea, which depends heavily on semiconductor exports. A drop in US AI spending could ripple through exports, production, facility investment and employment, and a repricing of AI-related assets could reach Korean financial markets. The institute urged preparation while strengthening domestic recovery. It also suggested that spending on stable data center operation and on power shortages will probably continue, pointing to power equipment as an export opportunity. Market watchers say the key moment will be when falling free cash flow and heavy bond issuance turn into actual investment cuts.