$ALAB Morgan Stanley Technology, Media & Telecom Conference 2026 Notes

Investment Analysis Clubs Saul’s Investing Discussions
1

I have been meaning to get in a few good posts (namely $LITE and $CRDO earnings among others), but have had some other things going on. I did want to briefly touch on $ALAB as their CEO and CFO presented yesterday for Morgan Stanley’s conference. I thought it was a particularly good call and I remain firm in my conviction.

Here are my notes and commentary:

  • Scorpio P TAM ~$4B
  • “fastest-growing product line in the history of the company”
  • “just in 3 quarters it’s already crossed 10% of our revenue”
  • “even more excited about the Scorpio X family which is applicable for scale-up”
  • they project Scorpio X TAM $10B with ramp started this year and expected through the back half of '26
  • UALink “has a bright future”
  • estimate scale-up TAM $20B in 2030 with half of that for UALink (directly addressable by product lines “we are building today”) and plans to address the other half as customer need emerges
  • I thought this was important: “…look at the portfolio, not as copper versus optical…Our customers absolutely want to use copper for as long as possible and as much as possible, simply because it is the most reliable, it is the cheapest and it’s the lowest power technology.” This mimics what $CRDO said on their earnings call and is something I think much of the market is missing. Keep in mind that copper’s benefit is reliability, power efficiency, and cost but its weakness is degradation over distance. Optical can do distance but currently the reliability and cost issues are limiting factors.
  • “…as these clusters scale and data rates go up and you need to connect multiple racks, you do need to go to optical. And we intend to fully participate in this optical opportunity that is in front of us.”
  • Copper may be able to use zero retimers per connection whereas optical needs at least two. These have significant ASP in optical so it is a big opportunity down the road but obviously clients don’t want to pay more for that buildout preemptively. This is extremely bullish for Astera Labs and Credo because it demonstrates that rather than going “all out” for the highest-end equipment, hyperscalers and other high-end clients are systematically deploying their builds in a stepwise fashion. CapEx is already extremely significant and these additional upsells would multiply those numbers (which shareholders of the hyperscalers seem to hate). However, as they begin to monetize these buildouts (e.g., the massive $6B increase in Anthropic run-rate just in February alone!) they will be able to further buildout with upgraded components over time. This paves a long runway for companies such as $ALAB, $CRDO, and $LITE.
  • aiXscale photonics acquisition they see as very accretive as it addresses a sticky part of the optical engine runway and postures them nicely for the next few years of transition
  • Scorpio X will be ramping throughout this year with initial volumes shipped in Q4 '25 but expect a “much more material ramp in Q3 and Q4 as well”; “it will cross over P and Scorpio in total will be our biggest product line in short order”
  • I like the CFOs comments about their current focus as well: “…right now we’re not optimizing to the operating margin. We’re optimizing to seize the moment with the TAM that’s in front of us.”
  • He mentioned their acquisition/acquihire as being to that end and referenced “very sizable” revenue opportunities as they invest significantly in R&D; overall I love to see a company improving its margins to give back to the shareholders. The exception is when further expenditures on the business itself end up being an investment for the future. Given this generational opportunity in AI/data center buildout that most insiders believe is in the early innings, it makes sense Astera wants to posture itself to continue to capitalize on opportunities in a shifting landscape.
  • I really thought Jitendra Mohan’s commentary on scale-up vs scale-out was helpful for me, particularly in the context of Credo’s recent earnings call. He explained that at 200 gig he believes all connections within the rack level (scale-up) will continue to be copper without a need to transition to optical. However, rack-to-rack interconnect (scale-out) may need to go optical as the distance scales above several meters. ALAB seems to have more focus on scale-up whereas CRDO is homed in on scale-out.

I really like Astera’s management and their company focus. Their execution is also phenomenal as evidenced by their ramping revenue, improving margins, and the recent multibillion dollar deal with a hyperscaler (see @wpr101 Feb portfolio post for more details). I continue to hold them as one of my top positions and think they will get a rerate when the market figures it out.

29 Likes
2

They’re talking both sides of this as Astera continued:

Astera recently bought optical company aiXscale Photonics:

“The transition to AI Infrastructure 2.0 demands purpose-built optical solutions that can handle the complexity and capacity needs of future scale-up networks,” said Sanjay Gajendra, COO and president of Astera Labs. “This acquisition will bring critical talent and advanced photonic technology that, when combined with our fabric switch and signal conditioning expertise, will unleash the full potential of rack-scale AI deployments.”

I’m not a networking expert by any means, but it seems to me there are 4 tiers of AI related connectivity:

  1. Driect Attach Copper (DAC) (A fancy name for a cable as we’ve known it for a century)
  2. Active Electrical Copper (AEC)
  3. Active Optical Connector (AOC)
  4. Optical

The AEC technology includes the famous retimers on which Astera made their initial money. These basically “fix” the signal degradation that copper wiring has. Remember, for AI we’re talking high bandwidth, low latency requirements, so it’s not like old-fashioned stereo cables are suitable.

The AOC technology is newer. Regular optical cables need transceivers to convert the optical signals to electrical signals (as computing chips all work on electrical signals, not light). These consume electricity, and cable termination is apparently still a reliabilty issue. The AOC embeds the transceiver into the cable so it’s all sealed. They are smaller transceivers, so they won’t do really long distances, but for within most data centers, especially for racks reasonably close to one another, they’re fine.

The breakdown for usage appears to be:

  1. DAC for on-board connectivity
  2. AEC for between board connectivity
  3. AOC for rack to rack connectivity in a cluster
  4. Full on Optical for real distance (over 100 meters)

Nvidia has been pushing its NVLink technology. Astera competed with that with its PCIe based AECs using retimers. Heck Nvidia’s Hopper boards had Astera retimers on them! But, that’s changed with Blackwell, which is all NVLink based. And Nvidia convinced Intel to support NVLink, so now it’s no longer a single-source supplier. I believe this relegates Astera’s retimer business to non-Nvidia deployments, such as Anthropic’s use of Amazon’s Trainium ASICs.

Astera had previously demonstrated using AECs to connect racks that were within several feet of each other. I wonder if that market has shifted to AOCs now. I suspect there’s still a cost and energy difference that favors AECs, but as server clusters grow, the distance requirements alone will force AOCs, I believe.

The UALink tech that a number of companies support competes with Nvidia’s NVLink, but where UALink probably shines is with non-Nvidia deployments, such as with AMD-based systems. I think the jury is still out on widespread UALink adoption. Sure the hardware companies trying to grab a slice of the Nvidia-based Ai dominant pie want a slice, but we still have to see what data centers actually end up choosing. And now that companies buying Nvidia are essentially buying versions of the NVL72 reference server architecture that Nvidia defined, someone would have to really want to replace the NVLink and get that requalified, etc.

I still have a position in Astera, but I think Nvidia’s reduction in retimer use combined with a second company, Intel, supporting NVLink, combined with data center footprint growing has limited the performance of ALAB in the past year. I do agree that Astera’s management team impressed me, but I think they have challenges to overcome in the ever-changing AI market.

28 Likes
3

The $AVGO team effectively just came out and reaffirmed what I just shared about companies staying copper-native where they can. $CRDO and $ALAB obvious beneficiaries and stocks flying after hours.

22 Likes
4

If I recall correctly, ALAB CEO has said before that they never had a chance to participate the Nvidia GPU server up until the NVLink Fusion where there is a hybrid between GPU and XPU. ALAB main market is still from XPU server and I think there will be plenty of opportunity for ALAB to grow from XPU because it is cheaper to go with XPU for inference AI workload (unless NVdia dominate in inference market and define their own standard again).

16 Likes
5

No, as Astera itself wrote:

Astera Labs and NVIDIA’s deep collaboration extends over multiple generations of NVIDIA platforms, with Astera Labs’ Aries PCIe/CXL Smart DSP Retimers deployed in volume across NVIDIA Hopper and NVIDIA HGX platforms

And as seen in this Tom’s Hardware article:

Astera Labs has already had a big win as the company’s previous-gen retimers power Nvidia’s HGX GPU systems. Here, we can see an existing HGX H100 system with eight fire-breathing H100 GPUs in the OAM form factor placed on a universal baseboard

This system now powers much of the world’s AI infrastructure, so much so that it remains in a state of shortage, and we can see the eight rectangular Astera Labs PCIe 5.0 retimers on the left of the motherboard in the second image. Another eight retimers also reside on the right side of the motherboard, for a total of 16 chips per board, boosting the signals between the GPU and the CPUs that reside on a connected motherboard (not shown here). The connected motherboard (diagrams in the slide deck below) also houses nine retimers as well, so there is plenty of Astera silicon in every shipping Nvidia HGX system.

As Semi-Analysis later reported, however, Nvidia designed its newer boards to place connected components closer together, which eliminated some/all of the Astera retimers built into those boards. Here’s an article referencing the Semi-Analysis article if you’re interested in the details:

A report by SemiAnalysis has revealed that Nvidia’s Blackwell Ultra B300 chip will not use Astera Labs’ retimers out of the box.

It added that some hyperscalers could still choose to use Peripheral Component Interconnect Express (PCIe) retimer switches, but it’s unclear how many will opt for it.

If Astera is now relegated by Nvidia to just the NVLink solution, then those using just Nvidia in their data centers may not be purchasing new Astera Labs components.

12 Likes
6

That’s what I said though, that ALAB never had a chance until NVLink Fusion.

Here’s the quote from the CEO during Needham conference

https://seekingalpha.com/article/4859758-astera-labs-inc-alab-presents-at-28th-annual-needham-growth-conference-transcript

”NVLink Fusion was a new announcement that caused a little bit of a stir in the market to begin with. So NVLink Fusion is where you are able to use the nice ecosystem that NVIDIA has curated with the NVL 72, including the power components and the liquid cooling and all the nuts and bolts and everything and the NV switches in particular, and use that with your own compute rack.

So not only does it present a new opportunity for Astera, previously NVLink was really off limits. Now we have an ability to build a solution that attaches 1:1 with XPU. For every XPU, you need this solution. This is a new TAM. So we are very excited about it. We have one hyperscaler customer that’s deploying it. Hopefully, we’ll see other customers deploy it as well over time.”

My point is, they are never big in GPU from the start, except for Aries the discrete retimer chip on the board. Their big revenue was from hyperscalers that want to go with XPU. Now they have more chance for GPU as well, for hybrid model where customers put both XPU and GPU in the system.

17 Likes
7

Aries was the primary driver of Astera’s revenue in 2023 and accounted for most of its 242% YoY growth in 2024, as Astera themselves announced:

The revenue growth in 2024 was largely driven by Aries PCIe Retimer products…

Which is why some investors worried as Nvidia redesigned to avoid retimers as much as possible. However, Astera smartly moved into new areas of connectivity.

As for the CEO’s comment on NVLink Fusion:

That hyperscaler is clearly Amazon, which as of 11Feb owned $43M of Astera shares, with a new warrant announced enabling them to own up to $466m of ALAB, based on Amazon buying $6.5B worth of future Astera products.

AI build outs are complicated.

NVLink Fusion is a great move for Nvidia, but it’s really competition to UALink, which is a Scale-Up technology that Astera has been pushing. So, it’s good that Astera is a player in both technologies.

The question for us investors is how often Astera will be chosen over others for their NVLink Fusion implementations. Astera’s software-defined architecture (which they call COSMOS) may mean they’ll have products to market sooner, especially as many of these heterogenous data centers may need custom tailored solutions.

And while I’m positive, it’s not a slam dunk for Astera.

27 Likes