Is the AI trade over? Bernstein weighs in

The bear case on AI doesn’t sound "particularly compelling", Bernstein analyst Daniel Zhu says, and remains positive longer term, but acknowledged it is hard to disprove near term and that a full recovery in AI-linked stocks could take time.

"We were wrong to be bullish on AI on the short term," Zhu wrote, referring to the firm’s Sept. 29 initiation of coverage of networking stocks. The SOX semiconductor index is down 38% from its June 22 high, the second sharpest drawdown since ChatGPT’s launch, behind only the one after Liberation Day.

Bernstein’s networking coverage has fallen between 12% and 42%, even though all five companies that have reported since the launch issued guidance above consensus. Zhu attributed the drop to multiple compression, not estimates. 

He identified two fundamental bear arguments. The first is concern over the economics of AI labs and the financing of the buildout, after weaker AI annual recurring revenue in September. Zhu called September "a pricing blip rather than a structural deceleration," noting that lab revenue came in stronger in October and token volume growth has stayed strong.

The analyst estimated hyperscalers will have about $1.2 trillion of AI infrastructure online by 2027. At a 15% return hurdle and 24% net margins, that implies roughly $729 billion in end spending, which he called plausible, though "not necessarily a slam dunk."

The second concern is land and power constraints, heightened by Democratic midterm victories. Texas Governor-elect Gina Hinojosa campaigned on a moratorium on approvals. Zhu sees limited impact on 2027, since capacity due that year is largely already in the ground, with only 0.3 gigawatts of credible Texas capacity yet to break ground. He sees more risk to 2028, but believes data centers are generally willing to pay to resolve such issues.

Macro and technical factors also played a role, Zhu noted, including investors selling tech holdings for liquidity ahead of major IPOs, rising rates, fears of renewed war with Iran and pressure on short positions to force liquidations.

He sees potential for a technical bounce, though catalysts look light until January. "In general we would wait to see more evidence that AI lab buildouts are actually under strain... before we’d begin to panic," Zhu said. 

Bernstein left its forecasts unchanged but cut price targets on Lumentum to $635 from $900, Celestica to $340 from $495 and Corning to $120 from $140. It reiterated Outperform on Arista, Ciena, Celestica, Coherent and Lumentum, and Market-Perform on Cisco and Corning.