We’ve heard from a handful of management teams that are direct suppliers to Blackwell over the last four business days, and the tone has shifted to more of a H2 discussion for rack-level systems. There are a few possibilities for this shift in tone, such as these specific suppliers are falling out of favor with Nvidia, but given that a variety of suppliers are echoing one another, it’s more likely that Nvidia’s GB200s are not completely on time to ramp in volume for Q1 and at least for part of Q2 as “mid-summer” and “H2” has come up repeatedly.
This is not a prediction, as reading tea leaves or a crystal ball is not the service we offer. What we offer is analysis – a way of connecting the dots to make more informed decisions. Right now, according to our analysis and earnings calls, the risk to the downside has increased that Nvidia is not shipping on time for the more complex GPU systems. Should the analysis play out, it simply means we will get Nvidia’s stock lower for the next leg higher.
The comments offered below are subtleties, as the last thing a supplier wants to do is get in the crosshairs of Nvidia. Yet, we’ve had a handful of suppliers’ report, and I’m struggling to find the expected strong commentary for the March quarter and there is also an absence of strong commentary for the June quarter (as it currently stands).
I’ve been on back-to-back earnings calls and offer the following excerpts in terms of this conclusion.
Astera Labs: Aries PCIe Gen 6 Tone Changed
It makes sense to start with Astera because they were the most direct in terms of discussing merchant GPUs (i.e., Nvidia) not being the driver for Q1.
Astera stated they are shipping pre-production quantities for its Scorpio P Series and Aries PCIe Gen 6 solutions to maximize GPU throughput. The company stated: “These programs are driving higher dollar content opportunities for Astera Labs on a full rack and full accelerator basis and we expect volume deployments to begin in the second half of this year.”
What this comment implies is both the Scorpio P Series and PCIe Gen 6 will deploy in H2, yet originally, PCIe 6.0 was expected to ramp with support initially offered in the GB200s. Back in March, Astera demo’ed PCIe 6.0 for a wide range of Blackwell products.
There was also indication back in the August call that Gen 6 was confirmed to be used in Blackwell’s GB200, and there were initial shipments: “We have started shipping initial quantities of preproduction orders of our PCIe Gen 6 solution, Aries 6. We ship and support our hyperscaler customers initial program developments that are based on Nvidia's Blackwell platform, including GB200.”
There was further confirmation in the November call that preproduction volumes were shipping: “At the recent 2024 OCP Global Summit, we demonstrated the industry's first PCIe Gen 6 fabric switch, which is currently shipping in preproduction volumes for AI platforms.”
As Nvidia investors, we keep a close eye on these suppliers as they often can flag any issues with unparalleled accuracy. Given we are nose-down on many earnings reports over the past year regarding Nvidia’s AI systems specifically, there’s a notable change in tone for Astera to go from preproduction volumes to now shipping in volume in H2.
The following was discussed in the Q&A section:
“Now, if you look into 2025, we see both contributing growth. The first half of the year will be more predominantly the AI – internal AI accelerator programs. But if you get into the back half of the year, the transition on the merchant GPUs will also be very strong for us. This is where you'll see the custom rack configurations start to deploy and that’s where we see a big dollar increase in our contemporary GPU with Scorpio starting to ramp.”
On one hand this is a positive, but on the other hand, it raises questions as to why merchant GPUs (Nvidia) will be stronger in the back half of the year as this does not match the anticipated timing originally described by Nvidia’s management team with the CEO stating, “Blackwell production is in full steam. I think we're in great shape with respect to the Blackwell ramp at this point.”
There could be an argument made it’s unique to Astera Labs, yet we have seen something similar echoed across a few management teams recently.
Thomas O'Malley Barclays Bank
Super helpful. And then my follow-up was just — I think it was Mike's commentary on one of the first questions here on the call. You talked about kind of the year 2025 on the Aries side, talking about how in the first half of the year, you would see more internal AI efforts followed by the second half of the year being more merchant GPU. That comment was a bit surprising to me given we're going through a big product transition now at the large customer of yours. So is there any change in the way that you see the ramp of 2025 versus where you did before? I would have anticipated maybe the merchant GPU being a bit stronger earlier in the year. Just any reason behind those comments that caught me a little off guard.
Michael Tate CFO
Sure. Yes, so the — first of all, the merchant GPU drives both Scorpio and Aries. So the big incremental piece of the merchant GPUs is the score field content which is all new for us. The designs that we have are complex in nature, they're all new. So the — to get them — to productize and ramp we're looking at that to start off in the back half of the year. Right now, in the first half of the year's preproduction. These are all for custom configuration. So the customization adds a little bit of lead time to the volume rates.
Coherent: Tone on Pluggable Optics Versus Active Cable
In our analysis published last week, it was stated: “As we look at larger AI systems ramping this year, there is some evidence that issues around single-rack Blackwell systems have been resolved yet multi-rack interconnects continue to see issues with overheating. Even though Nvidia has stated they are on schedule with Blackwell, the B300s and next generation Rubin will very likely require more fine-tuning.”
However, there was a sudden change in Coherent’s tone around fiber optics versus copper, which we outlined here:
“Looking beyond traditional pluggable optics, there is an increasing amount of discussion around co-packaged optics (CPOs), which places the optical transceivers directly on the chip package, rather than using separate optical modules. This results in faster data transmission, reduced latency and higher bandwidth. This may be the best of both worlds: the performance of optical yet with reduced power consumption. Tracking this is especially important as since we last covered copper/Semtech, there have been reports that copper is “causing concurrent issues with overheating and glitching” with rumors Nvidia will launch a CPO switch at the upcoming GTC. That could mean Coherent will be a lead supplier for the anticipated CPO switch – we will be monitoring this closely.”
This later led to Semtech stating they expect sales of CopperEdge products to be lower than the $50 million guidance provided in the earnings call four weeks ago. Our team’s spidey senses were up following Coherent’s report (per the paragraph above), and thus, we respected the stop we provided to our Members in weekly webinars – here and here. The stop was hit on Friday with a trade alert stating the story had changed about 30 mins before market close. The context is important as it communicates that Semtech was also caught off guard as to changes with the server rack configuration with a sudden, unforeseen change in the last four weeks.
Regarding Coherent, if they were the beneficiary of pluggable optics replacing copper, one would assume a strong Q1. Yet when asked about the flat QoQ revenue on the call, the CEO stated it was not due to the AI-related segments: “It's pretty straightforward. We expect datacom and telecom to be up sequentially. And we expect the rest of our industrial related businesses to be down sequentially. And that net at the midpoint to be flat.”
Perhaps AI-related segments are remaining strong enough to not miss guidance, yet why are they not able to offset industrial weakness for sequential growth. That piece was left unanswered.
Read-through for Credo?
One potential read-through is that Credo is in a better position to gain the business following Semtech’s announcement with its AECs for AI backend networks. The company also offers optical DSPs and PAM4 solutions. The possibility of Credo gaining this business is an outcome we have been prepping for by owning a wide array of suppliers. Read more here.
With that said, it’s the sudden change of information from four weeks ago – announced Friday after market close — that marks a notable shift our Members should be aware of.
Power Management Integrated Circuits (PMICs):
This component is where the rubber meets the road as the thermal management issues Blackwell has been rumored to face are addressed with PMICs.
Monolithic Power Systems reported on Friday and had many comments about their ramp being H2 weighted: “Yes. Just to add a little bit of color to how we see the year rolling out, we believe that, we will be off to a slower start in the first half of the year. But as the year develops, the customer base is expected to broaden as hyperscalers launch their new products. We have multiple product ramps with both existing customers as well as with these new hyperscalers. So as Michael just said, we believe, it's likely to be a flattish year, but we believe that from a quality and supply availability perspective, we're in very good shape.”
On one hand, some analysts do not think MPS is designed into the Blackwell systems (Keybanc) whereas Truist is “highly confident” MPS is designed into the systems. According to this report, I lean with Keybanc that MPS is designed into Blackwell Ultra.
“(11/18) KeyBanc lowered the firm's price target on Monolithic Power to $700 from $1,075 and keeps an Overweight rating on the shares. The firm believes Monolithic Power will lose significant market share on Blackwell with the ramp of GB200/B200, as Hopper PMIC overheating issues have persisted on Blackwell. KeyBanc expects IFX to be the primary supplier on Blackwell, with Renesas having secondary share. While Monolithic Power is attempting to requalify, the earliest this likely could happen would be with Blackwell Ultra in the second half of 2025.”
“(12/16) Truist analyst William Stein lowered the firm's price target on Monolithic Power to $762 from $887 and keeps a Buy rating on the shares. The firm reiterated a cautious semiconductor and artificial intelligence sector view, but is more constructive on Nvidia (NVDA) and Monolithic Power (MPWR) while more cautious on Tesla (TSLA). While Monolithic Power's print position at Nvidia is in doubt today, Truist is "highly confident" that Monolithic is designed in to Blackwell, the analyst tells investors in a research note. The firm cut enterprise data revenue estimates for Q1 significantly, but highlights Monolithic's "likely top position across a breadth of AI customers, supporting solid growth over the next few years."
Yet another PMIC supplier that we covered on the Advanced side on Feb 5th, is stating that “We know that we're not the only one players going after this, so the share is still to be determined. The final design, everything is still to be finalized as well, too. So we're working closely with the customer. Again, at this point, we can say that we're targeting for a mid-year launch.”
This company is either discussing Blackwell Ultra with the B300s, or they are implying GB200s are delayed. After hearing from more component suppliers, the likelihood they are actually discussing Blackwell (and not Blackwell Ultra) has increased since we wrote the analysis last week.
Conclusion:
Investors should have a plan for this – ride out any weakness with the understanding there is $300 billion in capex and growing pointed right at AI hardware, or have a more active stance which includes risk controls such as stops on positions, hedging, etcetera. For the IOF, we hedge while having predefined price targets and predefined stops. For the most part, should the analysis we’ve presented above play out, then our goal will be to buy quality suppliers at lower prices. I firmly believe the suppliers that move Blackwell forward with be greatly rewarded, and the hard work we are putting in will be rewarded.
Direct liquid cooling suppliers report today at the bell (SMCI) and tomorrow before market open (Vertiv). Let’s see if they are aligned (or not) with the other suppliers offering a mixed Q1 outlook.
Regarding Nvidia, we recently trimmed 6% to re-allocate to Nvidia AI hardware suppliers. If we break $126 and/or $113, it’s my understanding we will trim more with the goal of loading back up sub-$100 and perhaps even sub-$90. Join the Advanced webinar this Thursday to find out more.
Please note: The I/O Fund conducts research and draws conclusions for the company’s portfolio. We then share that information with our readers and offer real-time trade notifications. This is not a guarantee of a stock’s performance and it is not financial advice. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis. Beth Kindig and the I/O Fund own shares in NVDA at the time of writing and may own stocks pictured in the charts.
Recommended Reading: