Software Up 25%, Semiconductors Down 20%: AI Economics Debate Deepens Sector Divergence
nashnova research
Since late June, U.S. software stocks have rallied roughly 25% while the Philadelphia Semiconductor Index fell about 20% — a near-45-point gap that is historically rare and signals a fundamental repricing of where AI profits actually land.
Software vs. semiconductors — why the mirror image?
Since late June, the software sector has climbed roughly 25%; the Philadelphia Semiconductor Index (SOX) has dropped about 20% over the same period — a performance gap approaching 45 percentage points.
This means → capital is rotating out of "the hardware that builds AI" and into "the software that uses AI," repricing the entire AI value chain.
The IGV/SOX ratio, which tracks software's strength relative to semis, has surged to a multi-year high and is now pressing against a key long-term downtrend resistance line — a breakout would signal that software's lead could widen further.
How oversold are semiconductors right now?
SOX has given back most of its earlier "Leopold rally" gains. Its RSI — a momentum gauge measuring near-term overbought/oversold conditions — sits at the most oversold level since the late-July sell-off.
The index has broken below its 100-day moving average but still holds above the 200-day line. In plain terms = the short-term trend has turned negative, but the medium-term floor has not cracked yet.
Nvidia's recent relative strength has failed to lift the broader semiconductor sector — a historically unusual divergence. This reflects a concern that goes beyond any single company: the market is re-examining the bull case for the entire chip complex.
How has the AI economics logic shifted?
Goldman Sachs analyst Privorotsky warned: "More demand plus lower prices — if pricing falls faster than consumption grows, that is not automatically bullish."
This means → unit economics on the AI inference side are deteriorating. Token volume — the basic unit of text AI processes per request — is growing fast, but prices are falling faster. Volume gains may not offset price declines.
In plain terms = AI is being used more and more, but it is getting cheaper and cheaper. Chip sellers do not automatically earn more from that — and this is the fundamental headwind behind the semiconductor sell-off.
Is every part of the chip sector falling in lockstep?
Not quite. Analysts note that the supply-demand picture for training chips, high-bandwidth memory (HBM), and power infrastructure may not be deteriorating in step with the inference side.
This means → semiconductors are splitting internally — companies closer to the "training" link still face tight demand; those closer to "inference" face steeper pricing pressure.
SOX is sitting at what traders view as key support, and semiconductor option implied volatility has collapsed — the cost of expressing a directional bet, long or short, has fallen to historically rare lows. In plain terms = the "ticket price" for betting on semis via options is unusually cheap right now, whether you are bullish or bearish.
市场有风险，内容仅供研究参考，不构成投资建议。