Potentially this is a serious threat, and it's sometimes touted by critics who argue it's just emerging. Reality is a bit different though, Intel 100G transceivers have been around for some time, from an article more than a year ago:
IDF16 Intel is beginning volume shipments of its fabled Silicon Photonics networking technology, which uses light through thin glass fibers to replace copper wires. This tech can shuttle up to 100Gbps over 2km or more of cable that's just a few millimeters thick, creating a high-speed nervous-system for clusters and data centers. As its name suggests, Silicon Photonics uses patterned silicon to emit and receive precisely aligned data-carrying laser signals. These devices can be fabricated just like normal CMOS chips, meaning they should be relatively cheap and easy to make at high volumes. The dream is to embed this technology into processors and chipsets so they can shift data directly from their internal buses to external networks. That's the ultimate goal. Right now, though, we're just being offered discrete transceivers. You can basically whack a load of Intel's PSM4 and CWDM4 Silicon Photonics modules into a switch, or similar, and use them to network with other Silicon Photonics-equipped boxes, effectively running 100G Ethernet between devices. This technology has been in development for about 16 years.
Light at the end of Intel's Silicon Photonics: 100Gbps network tech finally shipping, sorta • The Register
Despite the fact that these were already shipping a year ago, Applied Opto was able to produce its highest growth during this period.
Here are some of these expressed fears about competition from Intel (and Luxtera):
I think it's fairly well know that Intel has been tanking price on 100G PSM4 with their silicon photonics product. They are expected to have a CDWM4 product this month. Luxtera also has a good silicon photonics line that is used at both MS and AMZN. My understanding is traditional products (AAOI) will have trouble competing with silicon photonics due to power consumption and the resulting operating cost over a 4 year life. Silicon photonics uses one laser vs 4 lasers for an AAOI transceiver. It is a higher power laser, but the power consumption is still significantly less. I believe this is why you see AAOI emphasizing CWDM4 these days. The PSM4 market is even more price aggressive. But if Intel and Luxtera are able to make a CWDM4 product successfully it will cause the same issues. More pricing pressure. I think the pressure AAOI is already seeing at 100G is a symptom of the amount of competition in the market. I think it only gets worse.
Applied Optoelectronics down after analyst price erosion comments - Applied Optoelectronics, Inc. (NASDAQ:AAOI) | Seeking Alpha
From Barrons: But then Murry tells me one of the problems is the overriding preoccupation of Wall Street with the flashiest new technology, such as “silicon photonics,” where optical waveguides can be fashioned on a piece of silicon, rather than as discrete optical components:
There’s a strong feeling among some investors or analysts that there’s some new technology out there that is going to radically change the way optical modules are manufactured, or the cost model. You hear about silicon photonics or various new designs. And I understand why investors look at those things. They are focused on the product itself. There must be some cool new thing that is going to radically alter the way this industry works. I’m a technology guy, I have a background in electrical engineering. These technologies that other companies are talking about, like silicon photonics, these are not new things, we’ve been talking about them for a long time. The way I look at it, these different technologies are useful tools to solve certain problems. Silicon photonics is a good way to make certain kinds of high-speed modules, well in excess of 100 gigabits per second, as long as you keep the distance of those signals very small. But from a cost perspective, it’s not obvious, and to AOI it’s obviously not a cost-competitive solution at 100G. As you go to higher speeds, the calculus can change. The assumption on the part of investors is that new tech always displaces old tech, and that’s not always the case.
Silicon photonics is indeed something being talked about in many corners of the chip and the fiber worlds, and so ask Murry if perhaps he's downplaying the true importance of it.
Investors, responds Murry, over-estimate the connection between silicon photonics and the world of fiber-optic communications:
Moore’s Law has to do with the increasing density of transistors on a silicon substrate. Silicon photonics has nothing to do with transistors. The best I can come up with is, it’s silicon, and everyone knows it’s cheap, and so that’s where the analysis ends. The truth is that silicon has gone a long way because there’s been the technology to make smaller and smaller features. Optics is a whole different animal. For one thing, the features are many orders of magnitude larger. Wave guides are measured in millimeters or centimeters. I try to be very open about it all, but there is no obvious correlation between more and more transistors on silicon, and using silicon waveguides to shape light. Silicon waveguides are not CMOS [complementary metal-oxide semiconductor, the most prevalent material of traditional chips]. They require process steps not involved in CMOS. One can’t just make silicon waveguides like that.
From Jay Daehna in a comment on his own must read article
Questions on Silicon Photonics (SiP). Sometimes I get flack for not discussing Applied Opto's competition enough although sometimes I do to the best of my ability. With hat said, I have some analyst and industry contacts but they not pervasive and I do not cover the optical industry full time. So I am soliciting feedback here from anyone who is "in the know" or technically savvy or in particular works in the industry and can help. No Bull/Bear tug of war here. Let me start with my understanding: My understanding is that as of late last year or earlier this year in the PSM4 (150m - 500M) segment of the intra data center optical transceiver market, Applied Opto was the number one supplier, InnoLight was #2, and Luxtera was #3. Subsequently as 2017 has unfolded, Luxtera has made substantial share gains with a very low price. In the mean time, Applied Opto utilized its somewhat constrained capacity to focus on CWDM, where the pricing was higher, margins better, and less competition.
My understanding is as of late last year or earlier this year in the CWDM market segment, Applied Opto was a clear #1, followed by InnoLight (30% higher cost is my understanding), with Finisar #3. However, Finisar said in a recent call that their CWDM revenue was a pretty small minority of their overall QSFP28 revenue, so I assumed at the time they were a pretty distant #3. So with this knowledge, I have assumed that Applied Opto is in pretty good shape in CWDM, at least for a while. I know that Lumentum and Finisar have cost down versions of CWDM transceivers that they are trying to market in this segment that are pretty fresh and may not be shipping in much volume yet if at all and Oclaro has said they see potential opportunities here, but they are quite small thus far. I also hear that Source Photonics and Color Chip are on the scene on some level. I am told that Color Chip has had a hard time scaling costs down. Not hearing much about Kaiam if they are even still there. Several months ago, the news broke that Fabrinet and MACOM were working with Amazon to develop a "white box" CWDM transceiver for Amazon.
Given the competitive landscape as it is known to me as described above, and considering it is a large and growing segment within the intra data center market, it makes perfect sense as to why Amazon and likely MSFT and FB among others would like to cultivate more high volume sources in CWDM. A while back I was told by an industry analyst who heard from an industry executive close to the laser space that Fabrinet/MACOM were having technical difficulties and the initiative was pushed out six months. An industry consultant confirmed this to me after checking around. It's never been officially announced and I haven't heard much news on it lately. If I assume for a moment that perhaps Amazon isn't totally psyched about how this is going, I ask myself who would be next on deck for them to turn to, which brings me to Intel, which has gotten a lot of press of late. As a side note, Luxtera, the other SiP player, announced a CWDM product at OFC in March, but no one can confirm to me that any have shipped in volume thus far. I have spoken to analysts that have spoken to the company. So. Intel. Industry analysts (sell side people) tell me that they believe Intel has shipped cumulatively 10s of thousands of units - based on 15+ years of work - and most of what they have shipped is PSM4, possibly almost all. Please correct me if I am wrong.
Applied Optoelectronics: Ugly Q3 Miss, Transitory Amazon Weakness - Buy The Dip - Applied Optoelectronics, Inc. (NASDAQ:AAOI) | Seeking Alpha
See next post for the second half
Part 2:
To put this into perspective, my understanding is the intra data center optical transceiver market is roughly 3 million units this year, which includes PSM4 and CWDM. I'm not sure if it includes short reach (<150m). According to Applied Opto, they were looking to ramp capacity to 1 million lasers per month by the end of 4Q this year. At 3 million lasers per quarter, that would be 750K transceivers per quarter (four lasers per transceiver), which I would assume could equate to about $200 million per quarter in revenue at $266 per unit (blended ASP).
LightCounting predicts total UNIT market growth of 100% - 200% in 2018 to 6 million to 9 million units. We know that the 150m to 2km intra data center market is cut into two technology driven segments, PSM4 at 150m-500m and CWDM at 500m-2km. We also know that Luxtera and Intel to a much less degree from a volume shipment perspective have had almost all of their success thus far in the PSM4 150m-500m segment. For CWDM, Luxtera has announced a product but no volume as of yet. Intel has a product announced, and according to posts in comment sections and some analyst reports (BWS/Hallum that I know of) suggest some level of volume CWDM shipments may have started and are poised to ramp sharply essentially now, crushing Applied Opto at Amazon and Facebook, at a minimum, starting in 4Q and maybe 3Q if that impacted the precipitous drop in sales to Amazon.
Now here is the heart of the question, has Intel cracked the code to deliver high performance and 99.9% plus reliability in CWDM at a highly competitive price with gobs of capacity that is ramping vertical right now? If so, if they take over the #1 market share position in CWDM, can they go from zero to hero, meaning 30%/40%+ market share in one or two quarters? If they do, will it all come from Applied Opto or will a big chunk of it come from the other - likely higher cost - competitors in the space? For example, couldn't Applied Opto drop the price and outbid InnoLight as well as Finisar and Lumentum that typically have potentially materially higher costs? Will the market growth be strong enough in 2018 for all players to grow meaningfully?
If Applied Opto is the clear #1 incumbent with a large and scalable factory, decent technology, and low costs, wouldn't it be likely that the big data center customers would keep them in the top 2 CWDM suppliers even if Intel gets a leg up in a rapidly growing market? I did read some stuff from an industry consultant that has been following the industry for decades who is not enamored with Intel's SiP technology, but thinks that its mere presence in CWDM will crush the pricing in that space, negatively impacting Applied Opto (i.e., take margins to 30% but not much market share). He believes Intel has already done this in PSM4, although I suspect Luxtera played an equal or larger role. So there it is. Please "illuminate" me. Did Intel crack the CWDM technology code? Is SiP viable at 150m - 2km in general? Has Intel displaced Applied Opto or any other existing volume supplier of CWDM recently or real time? Is Intel executing a vertical ramp real time? If so, how much share will they have by the end of 4Q or 1Q? Many thanks!
Applied Optoelectronics: Ugly Q3 Miss, Transitory Amazon Weakness - Buy The Dip - Applied Optoelectronics, Inc. (NASDAQ:AAOI) | Seeking Alpha
Big report on Intel's silicon photonics efforts:
The integration of optical data transmission with silicon integrated circuits has been a long-term research objective for Intel, and the first silicon photonics products are due to be announced in 2016. This report evaluates how this will impact IT as we know it today: What current problems are addressed, and what is the potential for the near future? What is still science fiction?
Intel's Silicon Photonics Products Could Change the World of IT
Kushagra Vaid, GM, Azure Cloud HW Engineering from Microsoft joined Intel on stage to reveal that Microsoft is an early adopter of the Intel Silicon Photonics technology and they are using it in Azure data centers to drive scale. Intel silicon photonics PSM4 module will be deployed in Microsoft Azure for switch-to-switch connectivity and it offers up to 100Gbps of data transfer speeds. Intel Silicon Photonics is now in volume production and shipping in the form of 100G optical transceivers. The products, Intel Silicon Photonics 100G PSM4 (Parallel Single Mode fiber 4-lane) and Intel Silicon Photonics 100G CWDM4 (Coarse Wavelength Division Multiplexing 4-lane) are small form-factor, high speed, and low power consumption products, targeted for use in data communications applications, in particular for switch-to-switch optical interconnects in data centers.
Microsoft is using Intel's Silicon Photonics in Azure data centers - MSPoweruser
It replaces copper wires with optical cables and electrons with light, while using the same manufacturing methods companies use to make processors. Today, these laser-based systems are capable of shuffling 100 Gigabits per second over cables up to 2 kilometers long. You’ll struggle to push 25Gbps over a 10-foot-long copper cable, Kushagra Vaid, who oversees infrastructure engineering for Microsoft’s cloud, said from stage at the Intel Developer Forum in San Francisco Wednesday. Vaid was speaking alongside Diane Bryant, executive VP and general manager of Intel’s Data Center Group, who announced availability of the chipmaker’s first two silicon photonics products: 100G optical transceivers for interconnecting data center network switches. According to Vaid, silicon photonics is the future of interconnection inside Microsoft’s cloud data centers. Intel’s announcement is just the first step. The company said the 100G technology will be followed by 400G interconnects for data center switches “within a couple of years.” The first two transceivers simply plug into switch front plates, but the company envisions future-generation products that are embedded in switch motherboards and eventually integrated with switch silicon itself.
Intel's First Silicon Photonics Gear for Data Center Networks Now Shipping | Data Center Knowledge
Intel and Facebook* are collaborating to define the next generation of rack technologies that enables the disaggregation of compute, network and storage resources. Quanta Computer* unveiled a mechanical prototype of the rack architecture to show the total cost, design and reliability improvement potential of disaggregation. The mechanical prototype includes Intel Silicon Photonics Technology, distributed input/output using Intel Ethernet switch silicon, and supports the Intel® Xeon® processor and the next-generation system-on-chip Intel® Atom™ processor code named “Avoton.” Intel has moved its silicon photonics efforts beyond research and development, and the company has produced engineering samples that run at speeds of up to 100 gigabits per second (Gbps).
Intel, Facebook Collaborate on Future Data Center Rack Technologies | Intel Newsroom