The AI Trade Is Finally Cracking | Weekly Roundup

The AI and semiconductor trade showed signs of cracking in a roughly four-standard-deviation “factor implosion,” as momentum reversed beyond mega-cap technology. Quinn said flows into U.S. AI and semiconductor stocks were intertwined with dollar strength against the yen and Korean won, amplifying the reversal. He estimated roughly $100 billion was held in triple-leveraged ETFs, with no clear signal for when that positioning might unwind. With momentum crowded, he warned that relatively few news catalysts could trigger a sharp reversal.

The weakness had already spread across semiconductor stocks in Japan, Korea and Taiwan, which reversed two to three months of gains in days. Quinn said pressure on the Magnificent Seven could extend to AI-related stocks across the region; another speaker called the speed of the reversal characteristic of a bubble. Meta was reportedly considering selling some AI compute capacity, raising Quinn’s question of whether supply could be exceeding demand, though no sale was confirmed. Separately, a widely viewed but unconfirmed post claimed an OpenAI spinout had made a major memory-efficiency breakthrough and expected to announce it soon. Amazon’s announcement also raised concern that large technology companies could cut capital spending to support cash flow. One speaker said hyperscalers were vulnerable because roughly 30% of their income earlier this year came from markups of AI labs’ private valuations on their balance sheets, making the trade reflexive.

The speakers warned that an AI unwind could coincide over the next few months with a reversal in the economic-reacceleration trade. Consumer support, hiring stabilization, manufacturing surveys and capital spending had improved, but some strength in industrials and small caps still relied on AI-related power and buildout activity. One speaker thought growth and inflation were near their peaks, citing weaker ISM new-orders-versus-inventories data as a sign of lower manufacturing PMIs ahead. The jobs report was mixed: unemployment fell, but nonfarm payrolls significantly missed and wage growth showed no revitalization. Falling labor-force participation could explain the lower unemployment rate, while weak hiring, wage growth and labor demand offered little reason for the Fed to hike.

Hawkish Fed sentiment reminded one speaker of June and July 2024, when a hawkish pivot coincided with weakening labor and inflation data. But fewer expected hikes would not necessarily mean strong liquidity or a boom in risk assets, and the case against hikes did not imply inflation would soon return to 2%–2.5%. Forward-looking indicators pointed to meaningfully lower inflation, though the Iran war and other factors had lifted core inflation. With oil back below $70 and recent Treasury-yield increases mostly reflecting real yields, the long end of the curve remained a concern. The speaker favored gold and the SOFR complex to express fading hawkish expectations.

One speaker remained mostly on the sidelines and did not dismiss the multi-year AI compute thesis: efficiency gains could enable more use, while bottlenecks and new use cases could sustain demand. Still, the thesis was under pressure. Fable was difficult to use because compute was unavailable, and questions remained about why Meta might rent excess compute and why OpenAI had delayed its IPO. Another speaker preferred selling rallies to buying dips, expected the Magnificent Seven to remain under pressure, and saw the next two months as especially treacherous given crowded positioning and investors’ risk-on stance.

Bitcoin had moved inversely to the broader unwind and held above $58,000 except for a liquidation wick to about $57,000; roughly $700 million was liquidated. One speaker wanted to watch Strategy for weeks or months without active buying before judging whether natural buyers would return. The fundamental case for the debasement trade remained strong, they argued, given government deficits at 6% of GDP, rising debt and manipulated issuance, but the selloff was a severe positioning rinse, not proof that debasement assets were safe. They preferred those assets to NASDAQ while hoping to buy dips in memory stocks; slowing growth and inflation alongside sticky inflation could renew their relative appeal. Crypto may also see a “Phoenix” emerge over the next year or two, despite many tokens being worthless. Because tokens may offer no equity claim and project structures can leave holders with little or no claim, the key question is where value accrues. The speaker also expected Bitcoin to rise when recurring liquidity risk-offs prompt authorities to add liquidity, a cycle they described as recurring every six, nine or twelve months.