Intel Shortlists Suitors for Programmable Chip Arm Altera

[XYang2023]

The problem is that Altera itself is in decline.

Intel bought Altera in 2015 for $16.7 billion, so we can't get the number for Altera's 2015 full year revenue. However, we do know Altera's 2014 revenue was $1.9 billion, and 2013 revenue was $1.7 billion.

Altera's revenue for 1Q2024 was $342 million, 2Q2024 was $361 million, and 3Q2024 was $412 million. Altera hasn't announced its 4Q2024 revenue yet. If we assume 4Q2024 is similar to 3Q2024, then Altera's 2024 revenue will be estimated at about $1.527 billion.

Sadly, under Intel's ownership for nine years, Altera's revenue has shrunk. We haven't considered the inflation factor at all!

From Intel:

"For Altera, revenue increased 14% sequentially and operating profit turned positive in Q3. We also announced the introduction of our new midrange and small form factor products, Agilex 5 and Agilex 3, to serve broad market customers and segments. With an increasingly competitive road map, the business is well positioned to show continued top and bottom-line improvements. Consistent with what we have said previously, we remain focused on selling a stake in Altera on a path to its IPO in the coming years."

Intel (INTC) Q3 2024 Earnings Call Transcript | The Motley Fool

INTC earnings call for the period ending September 30, 2024.

www.fool.com

From TSMC:

"不過，魏哲家預期，無晶圓廠半導體庫存將逐漸恢復健康，晶圓製造2.0產業今年在AI需求以及其他終端市場溫和復甦的帶動下，可成長10%，台積電表現也將再迎強勁成長的一年。"

Translation:

However, Wei Zhejia expects that fabless semiconductor inventory will gradually return to healthy levels. Driven by AI demand and the moderate recovery of other end markets, the wafer manufacturing 2.0 industry is expected to grow by 10% this year. TSMC is also poised for another year of strong growth.

https://www.rti.org.tw/news/view/id/2235107

 

[kingmouf]

Ture, maybe QCOM or BRDCM? Or maybe Altera goes IPO? Intel needs cash and Intel does not need Altera, my opinion.

FPGAs have an increasing part in communications, smartNICs and the rest. So for QCOM and BCOM, Altera can be seen either as an added value or a possibly competing solution. I could argue judging from the push of Xilinx/AMD in the space as well as automotive, that QCOM would be much more interested in Altera than anyone else mentioned here.

 

[Daniel Nenni]

FPGAs have an increasing part in communications, smartNICs and the rest. So for QCOM and BCOM, Altera can be seen either as an added value or a possibly competing solution. I could argue judging from the push of Xilinx/AMD in the space as well as automotive, that QCOM would be much more interested in Altera than anyone else mentioned here.

I agree. Altera and TSMC used to be very close before Xilinx joined them at 28nm. TSMC used Altera's dense test structures to develop and ramp processes. The big switch happened when Apple came and was the lead test and development partner. TSMC then treated Altera and Xilinx equally and Xilinx actually beat Altera to the 28nm node which is a big deal in the FPGA business.

Altera then went to Intel and Xilinx went to AMD and the rest is history. I think breaking out Altera is a good thing. Hopefully Altera can choose the best foundry and support a wider range of markets like they did in the past. At 40nm Altera was the market leader. Their close relationship with TSMC really did matter while Xilinx was floundering with UMC.

 

[blueone]

FPGAs have an increasing part in communications, smartNICs and the rest. So for QCOM and BCOM, Altera can be seen either as an added value or a possibly competing solution. I could argue judging from the push of Xilinx/AMD in the space as well as automotive, that QCOM would be much more interested in Altera than anyone else mentioned here.

I agree. Altera and TSMC used to be very close before Xilinx joined them at 28nm. TSMC used Altera's dense test structures to develop and ramp processes. The big switch happened when Apple came and was the lead test and development partner. TSMC then treated Altera and Xilinx equally and Xilinx actually beat Altera to the 28nm node which is a big deal in the FPGA business.

Altera then went to Intel and Xilinx went to AMD and the rest is history. I think breaking out Altera is a good thing. Hopefully Altera can choose the best foundry and support a wider range of markets like they did in the past. At 40nm Altera was the market leader. Their close relationship with TSMC really did matter while Xilinx was floundering with UMC.

FPGAs are a second-rate solution for production networking chips, and are primarily used by cloud data centers for custom solutions (e.g. Microsoft) for deployments which can't justify the development costs for custom networking ASICs (and they can't find the solutions they really want in the merchant marketplace) due to low volumes, and by adapter companies which haven't mustered sufficient volumes to justify an ASIC.

FPGAs are great for niches that need low latency solutions for low volume deployments where custom ASICs are economically impractical, and they're also awesome for ASIC design verification and concurrent software development for ASIC products, but if a volume-justified ASIC is available an FPGA solution is going to look slow, hot, and unit-cost expensive by comparison. So I've had trouble understanding the notion of FPGAs being a big growth market for many years. As a technical solution reconfigurable logic chips are fascinating, but fascinating does not necessarily equal a big market in the context of high volume chips.

 

[count]

FPGAs are a second-rate solution for production networking chips, and are primarily used by cloud data centers for custom solutions (e.g. Microsoft) for deployments which can't justify the development costs for custom networking ASICs (and they can't find the solutions they really want in the merchant marketplace) due to low volumes, and by adapter companies which haven't mustered sufficient volumes to justify an ASIC.

FPGAs are great for niches that need low latency solutions for low volume deployments where custom ASICs are economically impractical, and they're also awesome for ASIC design verification and concurrent software development for ASIC products, but if a volume-justified ASIC is available an FPGA solution is going to look slow, hot, and unit-cost expensive by comparison. So I've had trouble understanding the notion of FPGAs being a big growth market for many years. As a technical solution reconfigurable logic chips are fascinating, but fascinating does not necessarily equal a big market in the context of high volume chips.

Yeah I agree with this take. However a few years ago, there was a lot of buzz around big datacenter companies implementing algorithms on FGPA, for example Google was talking about using FGPA as a sort of accelerator for some of it's search algorithms. The idea was that they update their algorithms all the time, so an FPGA would be a good platform for them.

The reality is that there are only a handful of companies like google that can benefit from using FPGAs in this way and are actually equipped to do so.

 

[blueone]

Yeah I agree with this take. However a few years ago, there was a lot of buzz around big datacenter companies implementing algorithms on FGPA, for example Google was talking about using FGPA as a sort of accelerator for some of it's search algorithms. The idea was that they update their algorithms all the time, so an FPGA would be a good platform for them.

The reality is that there are only a handful of companies like google that can benefit from using FPGAs in this way and are actually equipped to do so.

Interesting comment about Google. I missed anything from them about FPGAs. Microsoft is the big guy fascinated for several years by them. This is one example of several application implementations they've done.

Microsoft Azure Stack Edge Pro FPGA overview

Describes Azure Stack Edge Pro FPGA, a storage solution that uses a physical device for network-based transfer into Azure.

learn.microsoft.com

 

[kingmouf]

Well you are right and wrong at the same time.

FPGAs were primarily used for a lot of the use cases you mentioned. ASIC prototyping was for several years their most applied use case by far. However, this is changing. The last figures I could find when Xilinx was independent (circa 2020) were that automotive, broadcast and consumer were approx. 20% of its business, wired and wireless approx. 35% and data center approx. 10%. The rest was A&D, Industrial and TME.

Especially with 5G deployments, this shift is even more profound, while AMD/Xilinx have reported many car manufacturers employing their ARM+FPGA integrated solutions (Zynq US+ and subsequent), especially for lidar use. Amazon has just updated their AWS systems and now offers the second version of FPGAs on the cloud (AWS F2 instances). FPGAs are also used in smartNICs, as accelerators for AI/ML, computational storage, etc. FPGAs are heavily deployed in military and space applications.

You are definitely right that if a competing ASIC emerges, then there is no reason to use an FPGA. The volumes for many applications though are not there and for many other applications there are significant issues when the need to update the logic is a must-have.

The other thing is that FPGAs have evolved. In the past, FPGAs were just a flood of gates. Now they are SoCs that the FPGA part is just one of their components. The other components include hardened CPUs, GPUs, ethernet cores, AI accelerator cores, I/O modules, high-speed serial links, encryption engines, etc etc.

 

[blueone]

Well you are right and wrong at the same time.

FPGAs were primarily used for a lot of the use cases you mentioned. ASIC prototyping was for several years their most applied use case by far. However, this is changing. The last figures I could find when Xilinx was independent (circa 2020) were that automotive, broadcast and consumer were approx. 20% of its business, wired and wireless approx. 35% and data center approx. 10%. The rest was A&D, Industrial and TME.

Especially with 5G deployments, this shift is even more profound, while AMD/Xilinx have reported many car manufacturers employing their ARM+FPGA integrated solutions (Zynq US+ and subsequent), especially for lidar use. Amazon has just updated their AWS systems and now offers the second version of FPGAs on the cloud (AWS F2 instances). FPGAs are also used in smartNICs, as accelerators for AI/ML, computational storage, etc. FPGAs are heavily deployed in military and space applications.

You are definitely right that if a competing ASIC emerges, then there is no reason to use an FPGA. The volumes for many applications though are not there and for many other applications there are significant issues when the need to update the logic is a must-have.

The other thing is that FPGAs have evolved. In the past, FPGAs were just a flood of gates. Now they are SoCs that the FPGA part is just one of their components. The other components include hardened CPUs, GPUs, ethernet cores, AI accelerator cores, I/O modules, high-speed serial links, encryption engines, etc etc.

The markets you list are well-known and mostly niches, and even the markets in this category that have higher volumes are mostly in the hundreds of thousands or low millions of units at the very top. The technology which has probably helped FPGAs sales the most, and is pretty much the entire market for AWS F2 instances, is the ability to program FPGA logic in C++. I hear the compiler stacks are still inefficient compared to RTL, but if you really need low latency and determinism they'll still beat any CPU & OS. As for FPGAs becoming SoCs... that's been typical in the FPGA market for years now. (I'm not as much of a dinosaur as some people think.) My point was simply that FPGAs are probably not going to be a big market, in the comparison to CPUs or GPUs.