SK Chairman Chey Tae-won Warns AI Boom Could Deflate Like a Bubble Without Sustainable Profit Models

Invert the Bank of Korea's ratio and the five companies' first-half capital spending comes to about $330 billion, because $169.15 billion is 51.3% of it ^[1]. JPMorgan's full-year projection is $791 billion ^[4], which leaves roughly $461 billion for the second half, about 40% more than the first ^[2]. Hold the borrowing share flat and that second half needs $237 billion of new bonds ^[3].

The 51.3% is a quotient of two aggregates ^[8]. Money raised in one month pays for whatever a treasury department chooses in another, so the figure establishes that the five came to public credit markets for a sum equal to half their building programme, and not much more than that. Seoul Economic Daily reports the companies cannot fund the investments from internal resources alone and are taking on more debt to pour into AI ^[14].

Chey's own emphasis is the return. "AI investment is being made on a scale 10 or 100 times larger than most investment plans," he said. "Given the enormous resources going in, better business models have to be built and commercialized so the money comes back" ^[6]. He has framed the industry around speed, scale and safety for years and lately added business models and sustainability to the list ^[7]. For manufacturers, he said "Manufacturing AI is not easy without a data platform of scale" ^[8].

SK is buying into the spending it is questioning. The group's data centre with Amazon Web Services in Ulsan's Mipo National Industrial Complex is described as a 7 trillion won facility ^[9], and Chey said of the talks behind it, "We met with big tech companies and made considerable progress" ^[11]. At the original 103 megawatts, 7 trillion won is about 68 billion won per megawatt; spread across 1,003 megawatts, it is 7 billion ^[7]. SK has not restated the cost figure alongside the capacity.

The test Chey named has a number attached to it: second-half issuance. If the five raise less than the $169.15 billion they raised in the first half while capex climbs toward $791 billion ^[5][4], operating cash is taking more of the weight and the bubble argument goes back to being an argument about multiples. The counter is that 51.3% was never a constraint at all: companies with this much operating cash pre-fund while spreads are tight, and the share falls without a dollar of new AI revenue arriving. In my view the reading breaks if second-half issuance rises while announced capex does not, because the borrowing would then be servicing something other than the build.