We're Lifting Our Price Target on This AI and Networking Chip Play

The positives keep on coming for the Pro Portfolio’s custom AI chip plays.

In mid-April, Meta (META) announced it was partnering with Broadcom (AVGO) to develop multiple generations of custom silicon chips. This past weekend, we learned Alphabet/Google (GOOGL) is in talks with Marvell (MRVL) for a new round of custom silicon chips. Earlier this week, we learned the latest Amazon (AMZN) tie-up with Anthropic includes Anthropic using Amazon’s Trainium chips, another positive for Marvell.

Today, Google unveiled the eighth generation of its Tensor Processing Units (TPUs), consisting of two chips dedicated to AI training and inference workloads. Both of those chips will be available later this year.

Also, this morning, we are learning that SK Hynix (HXSCL) is planning to invest 19T won (~$12.85 billion) ‌in a new manufacturing facility in South Korea for advanced packaging to meet the growing demand for AI memory. That helps put some additional context around new survey findings from the Global Electronics Association about how AI is eating up memory chip capacity:

62% of manufacturers report constrained availability or extended lead times

82% report rising prices, including 33% citing significant increases

Just 14% expect conditions to improve in the next six months

While 94% of companies cite they can still source memory, most face limitations that complicate production planning and increase costs

Aggregating these developments, we can surmise a few things:

Demand for AI silicon and memory remains strong, and industry capacity remains tight, a positive for our shares of Applied Materials (AMAT) .

The bullish multi-year comments by Broadcom and Marvell earlier this year have become far more credible.

Hyperscalers continue to line up capacity to meet expected capex demands for not only this year but the next few as well. While the focus now is on AI chips, as we saw with quarterly results and big bookings from GE Vernova (GEV) this morning, the follow-through to come is for power, networking and other aspects of the AI data center buildout.

In response, we are raising our MRVL price target to $160 from $150 and bumping up our panic point to $100 from $80. As we make these adjustments, we will also note that with an RSI reading above 85, MRVL shares are deeply overbought, and members should not commit fresh capital them at current levels.

Meanwhile, we are reviewing our AVGO price target and will have more to say on this as we get quarterly results from the hyperscalers, which could bring additional capex comments. The same goes for Arista Networks (ANET) and Eaton (ETN) , both of which are set to report on May 5. All three of these shares are also overbought at current levels, and Nvidia (NVDA) is not far behind.

While we see further upside as AI adoption and usage rise, driving demand for AI and data center infrastructure, the sharp move higher in all of these stocks in a relatively short period of time has us on the sidelines following the very profitable trimming we did earlier this month.

More Pro Portfolio:

At the time of publication, TheStreet Pro Portfolio was long AMAT, ANET, AVGO, ETN, GOOGL, META, MRVL, and NVDA.