Chip index (SOX) is 34% below its June close while big tech CAPEX keeps climbing. Equivalent of the 1998 dip or 2000 top?

In 2000 the chip index halved before telecom spending even started to fall, and then halved again. On Wednesday, the day after the election, Microsoft rose almost 7% and Amazon 6%, while Micron fell 7%. On Thursday the chip index (SOX) fell 12%, its worst day since March 2020, and the four big buyers of AI capacity fell 2% to 4%. On Friday SOX got back 4% and ended the week 34% below its June 30 close. The selling was put down to fears that data center building slows after Tuesday’s election. A large part of the chip makers’ revenue is capex at Microsoft, Alphabet, Amazon and Meta, so a slowdown hits the sellers straight away, while the buyers get their capacity later than they wanted and hold on to the money until then. Counting from the start of each build-out, telecom equipment spending got to about 1.9x its 1995 level in 2000, and SOX peaked at 8.5x in March 2000. This time the four companies’ capex is already at 4.1x the 2022 level, while SOX peaked at 3.6x in June and is at 2.4x after Friday. In the 90s SOX also fell about 45% in 1996 and about 50% in 1998 while spending was still rising, and both times it went on to new highs. The first leg of the 2000 crash looked just like those two. Telecom spending grew 25% in 2000, its fastest year, and didn’t fall until 2001 (-12%) and 2002 (-23%), when SOX had already halved. I think the building slows down from here, and that it would have slowed without the election too, since spending can’t keep doubling on cash that only grows 10%. Those four spent $202 billion on capex in the September quarter, more than double a year earlier, while the cash their operations brought in rose 10% to $175 billion. It’s the first quarter in at least five years where they spent more than they took in, and Alphabet and Amazon alone borrowed a net $74 billion in those three months. Part of the spending is already signed for years ahead in leases and power contracts, so the build-out most likely carries on at a slower pace, and the sellers’ growth rates were set on the faster one. For the buyers a slower pace means more cash kept today against returns that sit further out and are still only potential, which puts them on the stronger side of a slowdown, even if one of them loses ground to a rival that keeps spending. The price of AI is being pushed down as well, by Chinese models and by Western labs undercutting each other, so each dollar of capex has to earn its return selling something whose price is falling. A lot of the telecom build-out around 2000 was paid for with borrowed money, and the fiber was used for decades while the companies selling into it took some of the hardest falls, Cisco losing more than 80% of its value from its peak in March 2000. This time the borrowing happens with the 10-year Treasury close to 5%, well above the rates most of the last decade’s tech growth was financed at. In 1999, while he believed in the information revolution around him, Warren Buffett said: “The key to investing is not assessing how much an industry is going to affect society, or how much it will grow, but rather determining the competitive advantage of any given company and, above all, the durability of that advantage.”