The semiconductor scene is going through an intense period of volatility again, but nothing much has changed when it comes to Wall Street analysts who stand by their targets. With the broader semiconductor names under pressure following the single-day plunge suffered by the South Korean memory giants, questions linger as to whether the AI chip bubble has finally begun to show signs of deflating.
JPMorgan (NYSE:JPM | JPM Price Prediction) thinks the latest retreat is worth buying, but not everyone is pounding the table after the latest drawdown. Another bank, Morgan Stanley (NYSE:MS), seems to think that a rotation is underway. But which camp is right remains the hot topic of discussion. Personally, I think there are many ways to play the latest drop in chip stocks without having to step into the blast radius with the memory and storage makers.
Even if the semiconductors stand to benefit from a multi-year structural tailwind, with secular forces still very much in play, a valuation reset (or correction) alongside a rotation might still be on the table. So, instead of subscribing to one bank’s buy-the-dip approach or another’s cautious rotation call, I think it makes sense to expect both scenarios to unfold.
Perhaps the second half of the year is a correction period for the semis while investors rotate their winnings elsewhere, all while the long-term trend stays intact as the AI revolution continues to set a stage for more off-the-charts quarters for the firms sitting comfortably, continuing to sell out of components needed to get the accelerate the AI data center buildout or, better yet, get things running a bit ahead of schedule.
Meta Platforms: A stealth chip winner as custom silicon takes off
With Meta Platforms (NASDAQ:META) kicking off Meta Compute to sell extra capacity to other firms, Mark Zuckerberg and company might have the release valve to completely floor it with the buildout.
Perhaps it makes the most sense to build first and ask questions later about what the right level of AI compute is for a firm’s needs, given the bottlenecks that have popped up from left, right, and center.
From power demands to electrical components, it feels like procuring, building, and selling excess compute, if any, is the most logical move, as the hyperscalers scale up without showing any signs of looking back. While Meta isn’t a traditional chip play, I do think that its custom silicon efforts are being slept on by much of the market as shares sink further into bear market territory. Sure, many firms are getting into custom silicon, so it’s nothing that makes Meta unique.
But what sets the firm apart is its aggressive development cycle (six months rather than one to two years), architectural innovation to get around memory bottlenecks, and optimization for the Mixture-of-Experts (MoE) architecture, which may very well hold the key to next-generation AI that goes beyond large language models (LLMs). Perhaps it’s the MoE optimization that could ascend Meta’s silicon to the next level.
Meta Compute is still underestimated
MTIA is custom-tailored for Meta’s own uses, but with Meta Compute, it might soon become a gold standard as Meta looks to disrupt the neoclouds with not only scale but efficiencies that customers can’t get elsewhere.
So, while some may see Meta as having too much extra compute, I’d be more inclined to view the firm as positioning itself in a way so that it can get really aggressive. More recently, the firm was reported to have plans to spend $13 billion on a massive one-gigawatt (1.0 GW) AI data center in Alberta, Canada. That’s a massive undertaking if true.
With a Street-high price target just north of $1,000 per share (that belongs to Rosenblatt Securities) and widespread hedge fund buying activity in recent quarters, Meta Platforms may very well be an underrated gem as it looks to dominate in all areas it touches, from AI chips (MTIA) to data centers, and models (Muse Spark and Superintelligence Labs).
Rotation or not, Meta already seems set for a big win as it executes on its seriously aggressive AI strategy.
Contact [email protected] for any questions or corrections.