FPGA maker Altera declared independence from Intel on February 29, 2024 – Leap Day – and reclaimed its original, storied name in the process. The intended and explicit symbolism was that Altera was making a great leap forward by becoming independent. Intel purchased Altera at the end of 2015 for $16.7 billion and renamed the organization. It became the Intel Programmable Solutions Group (PSG). The acquisition did not turn out to be a marriage made in heaven due to the mismatch between Intel’s top priority, introducing and shipping CPUs, and Intel PSG’s mission: introducing and shipping FPGAs. I worked for Intel PSG’s marketing group during the last few years of PSG’s existence and can attest to this priority mismatch.
The person overseeing Altera’s re-emergence as an independent FPGA vendor is CEO Sandra Rivera. According to her official Intel biography, Rivera previously “led Intel’s Data Center and AI Group, where she guided development of data center products, including Intel Xeon processors, Intel Max and Flex series graphic processors, Intel Gaudi AI accelerators and FPGA products.” Notice how the FPGAs were tacked onto the tail end of Rivera’s bio. I don’t think that’s by accident.
Rivera shouldered a major turnaround when she became Altera’s CEO, and because of the efforts she’s made in the past year, I named her “FPGA Person of the Year” in my January FPGA Awards article. (See “The First Annual FPGA Awards – the Fibbies – celebrating 40 years of FPGAs.”) Not long after that article appeared, I received an email telling me that Rivera would like to speak with me via teleconference. We spoke for 30 minutes on February 6, and the relevant parts of that conversation appear below.
To start, I asked Rivera to review Altera’s past year, since declaring its independence from Intel, and she gave her perspective:
Rivera: “Well, I think it’s gone quite well, actually. We are pretty energized and excited about being smaller and more nimble. The distance between making a decision and executing it is a lot less in a smaller entity than it is in a large complex organization. The time to decision making is faster. We don’t have as many layers and governance. That’s necessary, maybe sometimes a little unnecessary, in a larger organization, but that’s the way of life in a large, complex enterprise…
“You know, the focus has been really from the beginning, ‘How we get back to better execution?’”
Rivera then said that Altera has been innovating well for many years, in her opinion, but said that Altera’s execution was not what it needs to be in recent years.
Rivera: “So we are very focused on a streamlined road map, with a clear approach in terms of a waterfall strategy for high-end, the mid-range, and low-end FPGAs, and we’re very excited to be bringing new products to market.”
I then asked Rivera to elaborate on Altera’s roadmap.
Rivera: “One of the things that we did, more than three years ago now, is we decided that we were going to have a full-stack approach, a waterfall approach, to the roadmap. We built out the Agilex portfolio at the high end of the roadmap as well as Agilex 5, the big middle, and then Agilex 3 [the low end of the Agilex line] …
“When I really came in full time in 2024, we still had a number of projects on the roadmap that I would say were probably more tactical, shorter term in nature, but that didn’t [mesh with] the idea that we would have greater IP reuse and more engineering leverage out of the investments that we make and by getting to a waterfall strategy, which also translates into lower risk, lower cost, and faster time to execution.
“So, the strategy around innovation and execution is tied to a roadmap that is consistent, where we have an approach that leverages IP from top to bottom and is anchored on fewer processes and technologies, where we can use IP in different instantiations of the products across the entire portfolio of the roadmap.”
Because she’d introduced this concept of a waterfall strategy that leveraged IP up and down the product line, I asked Rivera to elaborate on that idea:
Rivera: “So, two things, Steve. What IP reuse means is that we can characterize an IP on a process technology, and through chips and packaging decisions and different SKU options, utilize that Investment across more products and more areas of the road map.
“Then, secondly, the consistency of the software stack is critically important, as you know quite well, and is not so obvious for folks that aren’t [as familiar with FPGAs]. We who are immersed in FPGA technology, as you are, know the FPGA business is really a software business. While the vehicle for monetization is the silicon, it really is the software stack, the tool chain, and the tool flows that customers invest in over periods of time. It becomes so difficult to port over from design to design, and that is one of the areas that we’ve been very, very focused on in terms of the consistency of the portfolio.”
[Note: You can look at the glass half empty or half full here. Half empty means you’re locked into an FPGA vendor by the design tools. Half full means you only need to hike up the learning curve on the tools once. Rivera continued:]
Rivera: “Even when we were looking at our overall AI strategy and how we were going to implement it, we took a very different approach from the competition [specifically AMD/Xilinx] in terms of infusing the [programmable logic] fabric with AI capability that would let you scale up or scale down for as much AI as you wanted, as opposed to having a hardened block with vector engines. We really went with the idea that we’re creating a programmable logic device, a spatial architecture, and the software tool chain around it. That is what we want to provide to customers, so that adding AI/machine learning is similar to adding any other IP with the same tool flows in a Quartus software stack.
“The competition [again referring specifically to AMD/Xilinx] has a different AI tool flow that needs to be integrated with the FPGA flow, which adds more complexity and has more of a hardened block approach. For us, that software consistency and the IP reuse really are the hallmarks of how we’re planning out our roadmap and executing to the strategy.”
I then asked Rivera about singling out AMD/Xilinx as the sole FPGA competitor in Altera’s sights. She replied:
Rivera: “Our vision is to be the number one FPGA solutions provider in the world, and that’s a multi-year, many year journey. We always think about the number one player as our competition. Not that the other players – and clearly you know the folks that are playing at the low end – have done a very good job there. But we very much see ourselves as a full-stack, cloud-to-edge provider, and so we typically look at the number one player in the market as the competition, and we benchmark our performance, our innovation, and our differentiation against them.”
[Note: Here, Rivera acknowledges that FPGA vendors Lattice and Microsemi have done a good job at increasing their market share at the low end and with mid-range FPGAs, but at the same time, by saying Altera wants to be a “full-stack” player, Rivera’s acknowledging that Altera intends to win back some mid-range and low-end business with its Agilex 5 and Agilex 3 FPGA families.]
I then asked Rivera to give Altera a grade for the company as it stands today. She said:
Rivera: “Well, I think that if you ask our customers and ecosystem partners, Steve, they will see us living into our values, which is very much around customer obsession. You know, results driven. And of course, always on a foundation of innovation, but with speed and agility being more of the differentiated capability that we have [now], versus when we were part of a larger organization.
“…We’re hungry and leaning into this mission around being number one. So, we have work to do. Clearly, there’s nothing about the FPGA business that’s fast or transactional in terms of driving growth and success in delivering products to market.
“But there was a lot of momentum as we left the year [2024]. If you listened to the Intel earnings call last week, you can see we had sequential growth quarter on quarter, every quarter last year. We really did outperform relative to the competition in terms of our growth trajectory and gross margin and operating margin…
“And we are bringing exciting new products to market with Agilex 5 and Agilex 3 that have been in planning for quite some time that we’re excited to bring to market to meet our customer requirements.
“We are not perfect. We still have a lot of work to do to get that execution engine going the way that we want, to ensure that our ‘do/say’ ratio with our customers is better than it’s been, meaning that we make commitments, we meet commitments, and customers can count on us to show up when we say we’ll show up. This mantra around delivering the highest quality products on schedule as committed to the customers is something that drives us. But it’s always a journey of never-ending improvement.”
[Because it was a new term for me, I asked Rivera to confirm that the “do/say” ratio was the ratio of things done divided by the number of things you say you’re going to do. She confirmed. That’s what she meant.]
I then asked Rivera about Altera’s big challenges for the next twelve months. She replied:
Rivera: “We have a number of different goals. All of it starts with execution, execution, execution. So the organization is absolutely focused on the things that we can control: our product execution, the quality of our products, the schedules that we commit to, and delivering to those schedules, the support that we provide to customers in applications and designs, the opportunity that we have to support customers’ desire to include and add more AI and machine learning capability to either existing platforms or to new platforms. We think that’s TAM-expansive for the FPGA business.
“So, we have all of the goals and objectives around delivering to the product commitments that we’ve made to customers, and that’s really what the team is mostly focused on. But there’s a lot of other things, of course, that we don’t control such as geopolitics, which are always very interesting.
“My job really is to try to keep the team focused on the things that we control, and showing up for our customers, and making commitments. Good financial performance comes out of that, and then all the rest. My job is to kind of protect the organization from things they just don’t control. There’s not a lot of return on the calories spent ruminating and addressing or responding to rumors.”
Because Altera’s Agilex FPGAs are based on Intel’s semiconductor manufacturing capabilities, and because those capabilities have been in the news quite a lot over the past several months, I asked Rivera if Altera can keep its customer commitments in light of Altera’s heavy reliance on Intel’s manufacturing capabilities. She explained:
Rivera: “We manufacture our products at both TSMC and Intel. So, we are unique in that way, that many of our products are manufactured at TSMC. We have longevity on those products. We expect to be manufacturing, shipping a large number of those devices through 2040. That’s the commitment we’ve made to our customers.
Leibson: “Well, but those were the previous products, not the Agilex products.
Rivera: “Yeah, they’re the previous generation products. All the new generation products are on Intel Foundry, and I can tell you with all certainty, Steve, that there is not a capacity problem. At Intel, there’s plenty of capacity, plenty of capability. And even executing to our plans or even exceeding our plans, we will not move the needle of Intel’s fab capacity in any meaningful way, just because there are so many hundreds of millions of CPUs that go through those fabs. So, we don’t have a capacity problem in any way.
“You know one of the things that you and I may have talked about, or you may have picked this up in in the industry, is that because of the over-purchasing that happened during the supply constraint window, and all of the inventory buildup, some of the sensitivity that we have that we’re trying to work through with our distribution partners or ecosystem partners or our customers is that they do need to signal what their requirements are going to be as they work down and burn through that inventory position that they have. Because of that, the industry is not worried about fab capacity, whether that’s in Intel foundry, or TSMC, or Samsung.
“I think there’s plenty of capacity. They [customers] may believe that they don’t have to signal their requirements, but, of course with fab throughput times, particularly with FPGAs where there’s so many different SKUs and configurations, we do need more visibility into what the requirements are.
“We’ve been working through that. We don’t have a capacity problem, so you don’t need to worry about that, but we may have a planning problem if we don’t get sufficient lead times and insights into what your requirements are as the industry is burning through their inventory. I think this is some of the things that we hear from even my competition, certainly from other distributors. They want a bit more visibility than what we’ve been getting, because customers have not been worried about, ‘Hey, if I place the order, is there going to be supply?’
“Yes, there is [supply] because we don’t have a capacity problem. But the specific products that they’re trying to buy… If we don’t have the signals, we will just have a long throughput time that they’ll have to contend with, where they may be thinking that’s not one of the issues. We just have been working through some of that with our distributors, customers, and ecosystem.”
With four minutes remaining in the interview, I asked Rivera where she thinks things will be twelve months from now. She replied:
Rivera: “I think that you’re going to see continued financial performance of the business. Yes, top-line growth, but importantly, gross margin expansion and operating margin improvement, which will be a testament to product execution, design win conversion, and financial discipline. I think the other thing that you’ll hopefully see when we will be sitting here a year from now, is the kind of momentum that we will have achieved from the introduction of the many different variants of Agilex 5 and Agilex 3 FPGAs into the market.
“So, we’ll see that in the pipeline… We’ll see that in the number of design wins that we’ve been able to close. Then that will translate into revenue and financial performance. That’s for the business. I can confidently say at this point, Steve, that Q1 of 2024 was the low point in terms of the mismatch between end-customer demand and new billings, because there was so much inventory to burn through. It’s not that the end customers ever fell out of love with FPGAs or stopped needing FPGAs. It’s just that we had built up so much inventory that there was just this incongruence between the actual demand and deployment and the billings.
“I think Q1 of 2024 was the bottom, and because of the performance of the business, because we then reported three quarters of sequential growth and discussed what our plans are for this year, because of the size of the pipeline and the design wins that we see, and just the health that we see of Agilex FPGAs being really the fastest ramping products that we’ve ever introduced in our history, a year from now I’ll hopefully have the receipts to show that we’re executing and the customers are voting with their wallets.”
I’m planning to get back to Rivera next year to see if her predictions for Altera come true.
