expanding market opportunities with new Scorpio Fabric Switches, driving higher content in AI platforms
“business entered new growth phase”, multiple product families ramping across AI platforms from 3rd party GPUs to internally developed AI accelerators
Scorpio which supports backend GPU clustering is already shipping pre-production quantities
adj gross margin 77.8%
adj operating income 36.6M
adj operating margin 32.4%
adj net income 40.3M
adj EPS 0.23 (GAAP EPS -0.05)
Notes from the conference call,
Scorpio works for both scaling out and scaling up networks
P-series Scorpio speeds up data, X series improve efficiency and size of AI clusters
both P and X series have design wins with strong customer and ecosystem interest
will shift towards shorter AI platform refresh cycles
strong execution in all four product lines
new product lines and form factors going to high volume production
introduction of Scorpio smart fabric switches expanded the market opportunity significantly, adding additional multiyear growth trajectories
shipping preproduction on Leo product
interest in Scorpio has accelerated since the launch given its differentiated features
“we are very bullish on the market for our entire product portfolio across third-party GPU based systems with increasing dollar content per GPU on our new design wins”
Scorpio unlocks new market opportunities for design wins in Aries and Taurus with its reach extension
platform specific customization which provide hyperscalers with valuable flexibility
will benefit from new CPU generations that Intel and AMD have
ramping volume production for Taurus across hyperscaler general purpose compute
“Astera Labs has entered a new growth phase”
strong secular tailwinds along with expansion of silicon content in AI and general purpose compute platforms
over 10M smart connectivity solutions widely deployed and field tested across nearly all major AI infra programs globally
we shipped products to all major hyperscalers and AI accelerator manufacturers (note here they said “all”)
recognizing revenue across all four of product lines
Aries is largest contributor, driving up revenue in the quarter
third party GPU based AI platforms did well
strong ramps on new platforms featuring internally developed AI accelerators from multiple hyperscaler customers
cash position of 887M
expect margins to come down closer to 75% next quarter as mix will shift to shipping more hardware
Analyst suggests Astera’s “lead GPU customer” is ramping with Scorpio Switch products (Nvidia maybe?)
ASICs ramping with new SKUs in second half of year
“our visibility is very strong”, both with backlog and design wins
entering new phase of growth where revenue streams are clearly diversifying
good confidence on Taurus will grow nicely in 2025, a lot of products are just starting
software is differentiator with Scorpio, even form factor designs for AI
where Scorpio family sits we have access to a lot more diagnostic information, and collecting data from Aries and Taurus to have holistic view of AI infra to the data center operators
Analyst says COSMOS can do telemetry, linked performance, linked monitoring, tuning up parameters which is “so critical for data center managers”
since public Scorpio release we are getting a lot of incremental excitement for the customer base
“what is really neat about the Scorpio product family is a diversity of designs that we are seeing”
Leo product no only gets higher performance by the added memory, but also significantly better performance than adding additional CPUs - works well for large in memory database applications
TAM for P-series Scorpio is 1B+ today and at least doubling over time
X series will have a bigger TAM than the P series over time
currently X series not used much outside of Nvidia ecosystem, but can customize X to specific requirements of the hyperscalers
fairly diversified in the business today, but in 2025 and beyond customer bases will continue to evolve
expect 2025 to have greater than 10% revenue contribution from Scorpio
COSMOS advantages: management framework, customization framework, optimization type of feature set across all products
customers using Aries today will be easily able to extend the infra to run on top of Scorpio devices, unique advantage of Astera
Analyst, “ASIC side of the business is really ramping up strongly”
COO says public knowledge now that hyperscalers have doubled down on the investment they are doing for their own internal ASIC programs
seeing increased investment from hyperscalers on their own internal programs
design wins in several internal accelerator AI platforms, multiple product lines ramping
“we also get to play in the back end… a fertile land where there are multiple connectivity requirements that we can service with our Aries, Taurus product line and now, of course, with the Scorpio X series”
Scorpio X will contribute revenue towards the back of 2025
hyperscalers are putting a lot of dollars into building their own internal solutions, deploying and customizing through their own infra requirements
try to keep a very lean inventory, very purposefully and thoughtfully using multiple sources for substrate vendors, gaining from the work that has been done to enhance supply chain structure
Taurus will diversify revenue and will increase as a percentage of Q4 revenues, will be one of nice strong growth drivers for company
a lot of knowledge gone into COSMOS, being in trenches past three or four years
COSMOS very rich software, understands what works in the ecosystem
The Scorpio product line seems like a much bigger release than I expected and sounds like demand for Scorpio is off the charts. Hyperscalers which are building out their own chips and data-centers with unique requirements is already benefitting Astera. They say they are selling to every single hyperscaler and AI infrastructure player out there.
Revenue up 47% qoq is pretty amazing, I did not expect them to land as high as 113M on revenue. They guide to 126-130M might be a bit conservative from them given this commentary on the call. The company is also guiding to be GAAP profitable next quarter, so not only is revenue on a tear, the bottom line is improving with scale.
Their biggest competitor is Broadcom (AVGO) which is growing 47% and now has a market cap close to 800B. The analysts didn’t even mention competitors though; it sounds like Astera is defining the innovation within this field.
Incredibly strong earnings and the call was very convincing. Nearly all analysts congratulated the team for their strong quarter. On top of that, the answers of the management team were very convincing, plausible and strong as well. The market seems to follow, too. ALAB is up more than 38% at the moment.
I recommend listen to the call, if you are invested in or plan to start a position.
Thanks to you @wpr101 I invested into ALAB nearly at the bottom enjoyed the run up. My position is now up 123% and it grew into a 17%-ish position in my portfolio. So thank you @wpr101 for bringing this gem to the board.
The need for $ALAB’s products goes DOWN, the CLOSER the GPUs are to each other and the more densely packed.
The inverse is true: the need for $ALAB’s products goes UP, the FARTHER AWAY the GPUs are from each other and the less densely packed they are.
Right now, it’s impossible to get the GPUs as close, and as densely packed, as the Data Center Engineers would like, because there is insufficient availability of water-cooling
So for the time being, the GPUs have to be FARTHER AWAY, and LESS densely packed, than is optimal. Optimal GPU density pushes air-cooled systems past their physical limits.
Conclusion: at least until water-cooled data centers are widely deployed, there should be huge demand for $ALAB’s products, because they directly address the bottlenecks caused by less-than-optimal GPU proximity/density.
Any thoughts on this? Will this dynamic have a material impact on $ALAB’s revenues for the next few years?
@intjudo I believe that the Aries smart cable allows GPU clusters to be connected on multiple floors because the cables can be wired through the ceiling or the floor, and the cable extends now up to seven meters. This allows data centers to be built more vertically or have more floors in general rather than being very wide.
Aries SCMs extend high-bandwidth PCIe 5.0 and CXL signal reach beyond the rack at 128 GB/s for up to 7 meters to enable larger GPU clusters in a multi-rack architecture
I would add the Aries cable are only one variation of one product, in a family of four products which speed up different aspects such as data and memory. Recalling from previous conferences I believe they said they can optimize down to the silicon and GPU level, and their products range from connecting GPU clusters along with optimzing GPUs themselves. Since some of their deals are under NDA agreements it is kind of hard to figure out the exact nature of amount of “content” in Nvidia products.
I am somewhat amazed that the company came out with Scorpio so fast to become a revenue generator already where they said that 10%+ of revenue next year will be from this product. It seems Astera’s go to market strategy combined with R&D is extremely strong, and their innovations can play in a wide variety of GPU cluster designs. It would be interesting to know what innovations Astera may have in store for liquid cooled solutions.
Understanding ALAB products is really an insider issue. As an example, ALAB’s CXL solution is not as fast as the primary alternative method to address the issue. However, in operation ALAB’s solution actually gets more out of the GPU so that you don’t need as many GPUs. I forget the details but something regarding how it works with the CPU.
In any event, being too enmeshed with the nuances hurts investments. You react to every little opinion on something superior coming along. When in reality watching the numbers that corroborate the narrative is the best way to know who has the product most in demand. ALAB has stated that even with Nvidia their per GPU content will increase as will the pricing as connectivity speeds increase. More silicon content per product.
Completely agree, and with Astera the numbers and narrative line up extremely well. The company said it was going to have more content in the next generation of systems, and their revenue correspondingly shot up 47% qoq.
The bear cases for this company seem to rest upon Nvidia or Broadcom out innovating Astera or taking their market share. However, I believe Astera is a company that Nvidia and Broadcom should be concerned about. Astera is able to get so much diagnostic information about the systems that are built with Nvidia chips and Broadcom components. The other players in the industry simply do not have the same diagnostic capability and software first approach.
The high level of this story is that hyperscalers are all coming to Astera for help connecting and optimzing their Nvidia based systems. Hyperscalers are also building out custom hardware at much more rapid pace now and that trend is massively benefitting Astera.
Thanks for the helpful link; one of the bullet points on that link says:
" AI infrastructure must scale GPU clusters across racks as server racks can only accommodate a limited number of GPUs due to power and thermal management constraints"
I think that corroborates my assertion that $ALAB products are MORE necessary as GPU density DECREASES; would you agree?
It’ll be interesting to see if/how $ALAB’s role in AI infrastructure shifts / adapts / responds as water-cooled systems become more prevalent.
Side note: if anyone here knows of any water-cooled environment in which $ALAB products are deployed, please chime in! It’d be interesting to get a glimpse of the future of $ALAB’s role in the water-cooled AI factories of the future that are just-now being built.
I think that corroborates my assertion that $ALAB products are MORE necessary as GPU density DECREASES; would you agree?
My understanding is that liquid cooling allows for more density or GPU per rack because it improves the two fundamental constraints on racks which are power constraints, and thermal/heat constraints. The liquid cooling lowers the power used and gives off less heat, so therefore more GPUs can be placed in a given rack with less risk.
However, I believe this increased density of the racks with liquid cooling benefits Astera greatly. The newer systems with liquid cooling are not as tested and battle hardened for optimization as the air cooled systems are. Additionally complexity of the designs benefits Astera because its COSMOS software can diagnose the issues as they arise in the design phase before reaching production.
On the other side, Astera’s solutions benefit the air cooled systems or the systems that will be lower density of GPU per rack. These systems will be in service a long time still, sometimes alongside liquid cooled systems and all the GPUs need to be connected some way to build the GPU cluster.
Just an elaboration on CXL as I now recall more details. First, this product is more speculative as it is a paradigm change and CPUs supporting it are just now starting to be delivered.
CXL’s killer app (it has other apps if it becomes accepted on the market) is more efficiently running large databases. It enables lower energy, faster, database access, while using fewer CPUs. No guarantee CXL takes off. But if so, some real upside starting later in 2025 or so.
@wpr101 during yesterday’s $NVDA CC, (…reading between the lines…), imo Huang basically confirmed that there is insufficient supply of liquid-cooled floorspace.
I think that implies that racks of $NVDA GPUs are going to have to be spaced apart from each other to prevent overwhelming the status-quo air handling systems.
That further implies a real demand for $ALAB components to deal with the repercussions at the PCIE layer.
And I think it may not be a coincidence that $ALAB is up 10% so far today.
