We’re coming up on the six-month-anniversary of Astera Labs’ (ALAB) initial public offering (which occurred in March 2024, if you’re reading/watching this installment later than September). Since Chip Stock Investor has yet to do a deep dive (the CSI way is to let the dust settle on hot IPOs for at least six to 12 months before making a decision), it’s time we develop a first opinion.
Spoiler: This could be a key Nvidia (NVDA) supplier that yields valuable data on Nvidia, at the very least.
Astera’s place in the semi supply chain
Astera is a fabless chip designer, founded in 2017. You’ll see later that, for our purposes here, this was a pre-revenue business last year that is now rapidly ramping up production of its chips (manufacturing handled by none other than Taiwan Semi Manufacturing). Once complete, those chips get shipped to companies that integrate Astera’s products into their computing systems (like Nvidia’s Blackwell accelerated computing systems) and data centers.
Astera specializes in networking chips. There is stiff competition. The two big players in this space are the ubiquitous Broadcom (AVGO) and Marvell Technology Group (MRVL). Our Top Chip Stock Buy September 2024 – More AI Progress At Broadcom (AVGO Stock)
Riding the data center accelerated computing boom
How can a company like Astera gain traction when competing against such giants of the semi world? It helps being a designer for one of the hottest trends ever in the history of semiconductors: Accelerated computing and generative AI infrastructure.
Big tech hyperscalers – Amazon AWS, Microsoft Azure, Google Cloud, Meta (Facebook and Instagram), and others – are in a race to cram accelerated computing into their data centers. The TRUTH About the AI Bubble, Alphabet GOOGL Stock, and What It Means For Nvidia
Accelerated computing, of which genAI is but one type of computing workload, involves connecting large clusters of GPUs and other computing equipment together. That means running lots of Ethernet cables and other connectivity tech within the data center to cluster those GPUs into massive supercomputers.
Big tech eats first, but is there room on the plate for a side?
This is the world that Astera competes in. Companies like Broadcom have shared how their customers are now clustering tens-of-thousands of computing chips together to train AI, and generally accelerate the performance of their data centers. Astera is trying to cash in on this trend.
As Broadcom has shared in the following visuals, chips themselves need to be connected. But server racks also need to be connected, and clusters of server racks need to be coordinated and connected to other clusters. It’s a big complicated mess of chips and cabling!
Whether its Broadcom networking hardware, or that provided by Nvidia like with the upcoming launch of Blackwell, one of the many components involved in this tangle of electrical hardware is a small chip type called a retimer. We’ll discuss what that is momentarily. But despite the potentially stiff competition from some very large and very capable designers like Broadcom and Marvell, it seems Astera has found a way to complement the accelerated computing industry story, rather than trying to compete with the established leaders head-on.
ALAB looking to complement and compete within a niche
Here is what Astera’s small business looks like today, according to the company’s own description in its IPO filing from early 2024:
The retimers and the COSMOS software Astera ships with them isn’t an especially large market within accelerated computing… at least not yet. The TAM is currently estimated at about $17 billion. In context, accelerated computing overall is likely worth many tens of billions of dollars in annual spend right now, headed for possibly a couple hundred billion dollars a year by 2027 (by some lofty and optimistic estimates).
Now, we’ve recently made some comments about there being no “defensible moat” in software business models. But that doesn’t mean we believe there is in hardware. True “moats” in the tech world (of which there aren’t many, in our opinion) are usually built with a combination of both hardware and software. Our video last year on business models: The “Infinitely Scalable” Business Model, and Are Semiconductors Really the “New Oil”?
Interestingly, Astera’s management probably understands this. When it began designing its chips, it also began developing that COSMOS software to go with it right from the start. Think of COSMOS as a type of management, diagnostics, and analytics tool that ships with Astera products, used by its system designer (eg. Nvidia) and data center customers.
The ALAB semiconductor portfolio – what’s a retimer?
Back to the question about what a retimer is. Astera has two flavors of retimer, the first one called “Aries” that currently accounts for ~85% to 90% of total revenue.
Taurus is the other retimer product, but in different “packaging,” so to speak. While Aries is sold as a chip, Taurus is sold as an electrical cable or Ethernet cable module, sold to a device company (like Amphenol (APH) or TE Connectivity (TEL) or the like), which they then package into their Ethernet cable and connectivity device products.
We won’t spend much time on Leo, this product is still ramping up and may not begin contributing more meaningful revenue until 2025 or 2026, maybe later. But this product – a memory controller – may sound familiar. Silicon Motion (SIMO) was one of our top small-cap picks of 2024. It specializes in memory controllers too, but primarily for a different device than data centers (SIMO is especially involved in high-performance memory controllers for smartphones and other consumer electronics products). Our last video that included a review of SIMO: 3 Small-Cap Chip Stocks – Are They Still On Course For Big Investment Returns?
Aries and Taurus and other retimers on the market are an important component in the growing complexities of cloud and private data center design and construction. With Ethernet cables and other data signal infrastructure getting longer and more complicated, retimers help clean up and refresh degraded signal (that often happens with long cables), keep data timing (or clock) accurate, and then send on the refreshed signal to the next part of the data center.
Astera management has compared the retimer to a baton being handed off in a relay race. The previous and now tired runner passes the baton (the data) to the next and still-fresh runner.
Here’s what a degraded signal traveling along an Ethernet cable might looks like, and how a retimer cleans things up before sending the data packet onwards to its next destination.
ALAB – a proxy for epic Nvidia Blackwell rollout?
But is Astera really some sort of bet on Nvidia Blackwell? In earnings calls and industry events, it certainly seems like the Aries product is quickly ramping up as Blackwell system fabrication begins to ramp up at Taiwan Semi’s fabs. Astera’s listed use cases for Aries also seem to suggest being one of the ingredients in these computing machine behemoths from Nvidia.
Financials a portent of good things to come?
Which brings us to the financials. If Nvidia and its fab partner TSMC are scooping up retimers as part of the fab and packaging process for Blackwell, this might explain Astera’s rapid rise this year. Astera management has been suggesting, in so many words, it is riding Blackwell’s coattails. Notice the sequential revenue rise in Q2 2024 (ignore the year-over-year growth, Astera generated very little revenue in 2023 prior to the IPO).
And revenue is heating up. The Q3 guidance implies an acceleration in sequential revenue growth from Q2 to Q3 of 27%.
And despite its small size, Astera is making good progress towards GAAP profitability. On an adjusted basis, it’s already profitable. The difference between GAAP and adjusted metrics is almost entirely driven by employee stock-based compensation (SBC, over $40 million worth expected in Q3 once again). That’s not an unusual amount of SBCfor a recent IPO – and especially for a small tech company trying to rapidly expand.
Financial strength – enough worthy of a buy now?
Does this rapid accelerated compute growth make ALAB a buy? The business clearly has a direction it’s headed. And thanks to the IPO early this year, it has a full gas tank (cash and short-term investments) of over $830 million to help it towards its destination.
The valuation is getting more reasonable too. But it certainly isn’t cheap by any stretch of the imagination. On a one-year forward basis, ALAB trades for well over 50x and 70x expected earnings per share and free cash flow, respectively.
Even when backing out the large cash balance (enterprise value), ALAB still trades for a hefty >60x times expected free cash flow.
However, this is a very fast-growing chip design startup, so a premium price can be expected. Here’s one scenario that gets us to fair value (as of this writing/recording, about $38 a share):
The bar is set high. But this isn’t a totally absurd scenario, in our opinion. If the accelerated computing boom continues, spurred on by Blackwell rolling into hyperscaler data centers in late 2024, there’s a chance Astera meets or exceeds the above growth scenario.
Wrapping up – is CSI buying?
After a run through the Chip Stock Investor framework, there’s a lot we like about Astera Labs. As a small startup trading at a premium price, though, we also have some concerns. Here’s a summary:
For investors that want something approaching a “pure play” on Nvidia Blackwell, Astera Labs could yield growth that amounts to that. This makes it an intriguing possible bet, one that we might one day be interested in adding to our “small-but-promising basket” of investments. At this time, ALAB doesn’t provide anything new to our portfolio, though. With Nvidia as our largest position, we already have ample exposure to a potential “Blackwell bump.”
If you’re mulling taking a position in ALAB, our standard message applies: We believe this is a stock that should be kept as a very small position. Let the company prove its worth as a larger portfolio component over time. Start small with a nibble, and if Astera meets your criteria for investing, consider utilizing a dollar-cost average (DCA) plan to buy in increments, like monthly or quarterly.
Nevertheless, Astera is an exciting business with big potential prospects. The fact it has a software component to its business already seems like an especially prescient move by management. We’ll be keeping this one on our radar, keen to see how Blackwell affects the story.