Ride the Wave: Alphawave
The Price Leaves a Lot of Room for Error
Alphawave is an idea I wrote about in the 2023 Outlook piece. Alphawave is one of the most asymmetric ideas in the semiconductor universe I follow. Shoutout to VIC user agentcooper2120 for helping me get up to speed on this. He spotted this first.
Reminder: Nothing in this newsletter should be construed as investment advice. This is a research service, do your due diligence before you buy or sell a stock, and have it not be limited to this post.
Also, I want to thank Tegus! I didn’t personally conduct any calls on Alphawave, but that’s because the transcript service has a lot of value-adds! This was an example of where I could ramp to conviction because of the vast array of expert calls already done. A massive backlog of calls is being done on Alphawave, and they are a core part of this write-up!
Here are a few great calls on Tegus:
CEO and Founding Partner at Fab Economics
Consultant in Semiconductors and High-Tech
Business Development Manager - IPs worldwide at Ensillica Plc
Principal Director of Data, AI & IoT Cloud Solution Architect at Microsoft
I will do a price raise in two weeks in conjunction with some of my recent posts. My reasoning is keeping up with inflation and charging what I think is this newsletter’s value. This is a friendly reminder that you can get Fabricated Knowledge at the grandfathered price if you subscribe today. You’ll get another heads-up before the final price raise!
The new pricing schedule will be $35 a month and $350 a year two weeks from now. If you subscribe today (and get access to the long writeup on Alphawave), you get Fabricated Knowledge for the current price of $300 dollars.
Anyways on to the long post about Alphawave. This is an exciting situation with many risks, but the price more than compensates for the risk premium.
Here’s a brief outline of the company dive.
History of Alphawave
What does Alphawave Do
Alphawave Investment Thesis (paywall)
Industry Analysis (paywall)
Competitive Analysis (paywall)
Valuation and Model (paywall)
History of Alphawave
The story of Alphawave starts with Tony Pialis. Tony is the CEO and co-founder of Alphawave. He’s a serial entrepreneur in the space. He’s sold two companies, one to Rambus and one to Intel. Importantly Tony was VP of Analog and Mixed Signal IP at Intel between 2012 to 2017, which effectively was the precursor to his work at Alphawave.
Tony and his team delivered the SerDes solutions for Intel at 22 and 14nm. Today, that’s what Alphawave does, SerDes for leading-edge manufacturing, not just for Intel, but anyone wanting to purchase their IP.
Leading-edge development is not easy. It’s notoriously tricky, especially for mixed-signal design. As EDA tools have scaled, designing digital circuits is a tricky proposition but a place where EDA has met that need. The analog design has not quite scaled the same as digital, and what’s worse is mixed signal is even more complex. The crossover from Analog signals (continuous functions) to Digital is a hard place to design, and many digital designers consider it black magic.
So designing in mixed signal is hard; winning in mixed signal is much more challenging. That’s precisely what Tony Pialis and his team have done. Specifically, Alphawave was among the first to get to 112 NRZ/PAM4 on the TSMC N3E process and is often the first to each node. The critical competitor here is Synopsis, the leading IP house in the world. Alphawave doesn’t compete with Synopsis in every business but focuses very specifically on one of the world's most valuable types of IP, SerDes.
Alphawave focused on a new way to make SerDes, focusing on a digital SerDes solution instead of an analog solution. Their solution is much more digital than other digital DSP offerings, and the traditional analog solution is still widely used today. Importantly Alphawave can use the appeal of digital design to offer better power, performance, and area solutions than previous Analog designs. Below are the two comparisons of analog versus DSP solutions. Importantly the digital DSP design can be more flexible to different throughput rates, and scales much better with process shrink.
At a high level, the DSP has many more layers of noise correction, amplification, and other tricks that can scale much better to higher performance levels as silicon scaling continues. This complex set of engineering solutions is more than just traditional analog design but a combination of the mixed signal. In particular, the flattening of the channel, noise reduction, and reproduction techniques expand as more digital processes are added. And the insensitivity to process means standardization of IP and a march to more minor levels of semiconductor production instead of analog designs for each new node.
Tony has effectively always been at the leading edge of SerDes. And Intel was excited to design this digital SerDes solution. Still, it was likely not a priority within the bureaucracy of Intel in 2017, so of course, he left to found Alphawave. It’s time now to talk about Alphawave, Tony’s background, and the people surrounding Tony because that’s another exciting story in itself.
It’s likely Tony was excited about the Digital SerDes idea and was willing to all in on another piece of his life’s work; SerDes. He leaves Intel to start it and starts to pitch to potential customers after he develops the technology. One of the companies is the Chinese-based Verisilicon, whose chairman Wayne Dai was brothers-in-law with Weili Dai, the wife of the Marvell co-founder team. Let’s introduce them.
After Marvell’s accounting problems were brought to light, the founders of Marvell, Sehat Sutardja and Weili Dai were ousted and put out to pasture. I wrote briefly about this in my Marvell day overview. This husband and wife duo was aggressive and, while never indicted, clearly wouldn’t have another company management team opportunity available. Wayne Dai introduces the Sutardja family, taking a significant stake in the company. The Sutardja family is a passive stakeholder here, but they are technical and understand what they are investing in, and at this point in time should be considered a slight positive.
So this is how the complicated relationships with the former Marvell co-founder and the Verisilicon relationship began. Alphawave IPOs in the UK, allegedly to be compared to ARM and Imagination technologies and chugs along for a bit. Then the Financial Times drops a hit piece on Alphawave, highlighting their very questionable intercompany relationships. Shares drop an astounding 50%, and everyone assumes Alphawave is a fraud and doesn’t have a real business underneath them.
Fast forward to today. Alphawave is still around; it’s still receiving bookings and revenue, and it recently acquired two companies with the large cash stack they received from the IPO. Alphawave has some concerns, but in light of them winding down their WiseRoad JV, the continued bookings, and revenue growth, I think it is time to ride the wave. Alphawave is a rare opportunity in public markets when facts change and the assumptions surrounding a company don’t.
I get why there is skepticism, but it’s a real company with an actual product and strategy. Additionally, as they start to hit their guidance, shares should rally. It’s priced like a zero, and I firmly believe it isn’t. So what does Alphawave exactly do?
What does Alphawave Do?
I want to spend a moment here explaining precisely what SerDes is and then explain the variety of cores and other businesses Alphawave has bought. The core business of Alphawave is SerDes IP, which is a fancy way to say compressing and decompressing data. SerDes stands for Serializer/Deserializer.
Bits (data) are often transmitted simultaneously on different channels, but sometimes putting all the signals onto a single channel makes sense. Putting the data on a single channel in the form of serial bits. The analogy might be best put as a highway versus local streets.
Usually, the data goes in parallel, communicating to each part of the chip what signal or data needs to be done. But when the data needs to be offloaded, it would be inefficient for each of the sideroads to be connected externally outside the chip.
Instead, connecting one big data highway carrying all the little bits of data into a single channel would make sense. This is Serialization, and in this analogy, it’s when the cars (bits of data) come off the local roads, then get on the highway to move faster and further distances. They are deserialized at their destinations and then move in parallel again.
Serialization is when individual cars go onto the highway, and Deserialization is when the vehicles get off the highway to their local destinations. Highways are often only created for the densest information flow and travel faster, like when all the data from a GPU goes into Memory or vice versa.
As we move toward a world of more connections between different packages, these highways of data are becoming more critical. In the past, when we made monolithic dies, the data would mostly move locally on the chip (a series of densely connected side roads); now, when we connect two dies together, we have to make a highway to join the packages. This is why SerDes is becoming increasingly important.
Alphawave is a leading maker of SerDes IP. Tony Pialis was the leader of SerDes at Intel but decided to go it alone to sell SerDes technology to individual companies so that instead of having to make it themselves, they could leverage off-the-shelf IP and focus on their core design.
And SerDes specialization was quite the right move as things moved to a more heterogeneous future; with chips now needing to move parallel data to serial data off-package, SerDes started to increase. SerDes was the core IP product, and it is growing very fast, given its leverage to advanced packaging.
Alphawave also had a huge IPO, raising tons of cash, and then purchased a few other businesses. The core purchases are OpenFive and Banias Labs.
OpenFive was acquired in September of 2022 and was the first significant acquisition that Alphawave pursued. During the acquisition, there was a meaningful CFIUS review, and I think for Alphawave, this is when they began to switch their strategy from Chinese to North American facing.
The CFIUS review and the public criticism from Senator Rubio created a turning point in Alphawave’s story. After this point, they began to wind down the WiseWave relationship and earnestly pursue the North American market. After the CFIUS review, they began to strip out the Verisilicon contract expansion, and some Chinese shareholders likely blew out during this era as well.
OpenFive helped SiFive make custom SOCs based on the RISC-V ISA and was the architecture, design implementation, validation, and manufacturing outsourcer for SiFive.
Some companies wanted to go to SiFive and ask for a custom SOC; OpenFive helped make that a reality. As SiFive started to pursue a closer ARM-like business model, the outsourcing and design verification business made less and less sense. Alphawave decided to pounce on that opportunity.
It makes sense as SiFive offers a customizable but not custom solution. SiFive offers off-the-shelf cores and has a series of SKU options for their customers, not truly custom silicon. (An aside, I like this. It’s pretty cool)
If you’re offering a series of RISC-V cores with customizable but consistent SKUs off the shelf, why keep the entire custom silicon arm? So they sold it to Alphawave.
Under Alphawave, OpenFive has moved away from RISC-V and is now a broad custom silicon provider. This should align with their large IP business, which has a very low number of clients but a very high-dollar business.
OpenFive was a transformative acquisition to get Alphawave from an IP company to a services company. Frankly, I think it’s margin dilutive to the core SerDes IP business, but it likely expands the revenue opportunity as well as the ambition of Alphawave altogether.
I believe the acquisition was a small net positive, as they acquired a new source of potential future revenue, diversifying the company away from a single product company. Additionally, they can leverage their core SerDes and connectivity IP to become a heterogeneous design house.
OpenFive will drive the future custom silicon ambition at Alphawave and is a needed tool to make that happen.
Last, we have Banias. Banias is a meaningful call option but the smaller of the two significant opportunities. Banias is developing a coherent DSP to become a second source competitor to Inphi at Marvell.
Alphawave acquired the company for $240 million in October of 2022, and there are a lot of interesting little leads here. Avigdor Willenz is a key player here and one with an interesting background.
Willenz founded Galileo Technologies, which was sold to Marvell in 2001. He also founded Annapurna, which sold to Amazon in 2015. He’s a serial entrepreneur who usually has an exciting exit, but the companies he sells are actual.
Banias focused on making a coherent optical DSP akin to Inphi's. And in contingency with the purchase, there was a revenue commitment of $300 million in revenue, starting in 2024. Note that it’s non-binding.
Alongside the acquisition of Banias Labs, Alphawave has negotiated a non-binding, multi-year purchasing framework with a leading North American hyperscaler that proposes a multi-year roadmap for Alphawave to develop and sell a portfolio of optical products and DSPs, including coherent DSP technology from Banias Labs, with sales potentially ramping to over US$300 million.
The exciting part is that we now know who that customer is, Amazon. Amazon allegedly was the bridesmaid in the deal, pushing for the sale of Banias to Alphwave. The head of Annapurna likely knew Villenz and wanted to consolidate vendors into another platform they like, which in this case was Alphawave. This also sets up a potential acquisition if they are successful.
Alphawave also most likely does a meaningful amount of their business with the custom Amazon chips. As Tranium, Graviton, and Inferentia ramp, they should benefit from the SerDes chips going into production.
On net, this helps de-risk some of their long-term guides, as a meaningful revenue ramp should occur in 2023 and 2024, partially backed by the likelihood that this revenue ramp will happen. I believe with Amazon as the core customer; this makes it more likely.
Below is a summary of their three big verticals of revenue.
Anyways that’s where I leave you on the free article. I think Alphawave has some hair on the stock, but it’s not a fraud and is priced egregiously wrong. I make my case mostly behind the paywall, where you can see my model price assumptions. I lay out the rest of the case behind the paywall.