Catatonia in Catalonia: MWC 2024
Carrier capex in the doldrums, Qualcomm in a funk, and much more from a great sow
Just a reminder that we operate our newsletters on a paid model. Paid subscribers will get three newsletters a month including our China Deep Tech notes. Paid subscribers will also get early access to the newsletter as well other benefits coming soon. Please subscribe and support our work.
MWC 2024
We attended Mobile World Congress (MWC) last week. It is usually one of the most productive weeks of our year, and 2024 did not disappoint. We only raked up 30 scheduled meetings and 20,000 steps a day, but nonetheless came back with updates across many areas.
Chief among these:
3GPP will start working on 6G next year, which points to deployments coming no sooner than 2028. Sensible, but late enough to make many unhappy.
Mediatek looks good. Qualcomm does not.
ORAN is transforming from a broad initiative to a set of features that will be implemented in a fragmented way.
Chinese companies are back in force at the show, across all hardware categories. Good for conference exhibitors but the marketing sits over a deep layer of unease across the ecosystem.
All of the above X2 for Huawei.
Carrier spending on hardware is down and is likely to fall further this year. Spending on software looks better, but we see more pilot activity this year than actual deployments
IoT for Trees. Seriously. IoT for Trees
Apparently, there is a hot technology called “AI”, that seems to have a few people talking about it.
MWC The Show for Operators
Attendance this year reached 101,000 well above last year’s 85,000, but a bit shy of the pre-pandemic level of 110,000. The show was very crowded the first two days, but there was a noticeable tail-off by day three. Nonetheless, a healthy level.
This comes despite some noticeable absences. Early on we were warned that many companies would have reduced participation. Some of this can be blamed on the pandemic, with many companies still holding back from hosting big booths at trade shows. For example, neither NXP nor Infineon rented show floor space this year. By far, the bigger reason for the biggest attendance head wind comes down to carrier spending. After all, MWC is a show focused on selling things to wireless operators, and operators are not in a spending mood right now. To be clear, the show was busy, but it was also apparent that when it came down to attendance decisions many entities made the marginal decision to send fewer people or not attend. This was most notable among the financial community. Almost none of the couple dozen fund managers we regularly see at the show turned up this year. The sell-side also had a reduced presence. Which then led to many companies not hosting meetings at all. As we get into below, apparently Qualcomm did not even send its investor relations team.
So there was a cascade of reasons for people to not attend this year. And despite this, over 100,000 people did show up. The show is in good shape, and we suspect many more will attend next year.
The 5G Malaise
The root of the slowdown in carrier spending is that 5G has run its course. Most operators, in most countries have already deployed 5G gear and are in no hurry to do much more. As the old saying goes, the operators seem to only make money on even-numbered ‘Gs’, and 5G keeps the pattern going. The not-so-new standard has done nothing to help operators grow revenue - no obvious appeal for consumers, few new services. The hope was that 5G would at least give the operators ways to reduce costs by simplifying their networks. The problem is that to really make that happen requires some significant upfront investments and painful re-tooling of the plumbing of their networks. In many cases, the operators are seeing the cost per bit of running traffic on their network not exactly increasing, but they are definitely not declining much either.
Originally, this year we had expected the industry to instead attempt to shift the focus to 6G. That did not happen because instead of talking 6G, everyone wanted to talk about AI. We will come back to that topic at the bottom of this note, but all you really need to know is that AI is not poised to revolutionize telecom anytime soon, it will bring plenty of ‘under the hood’ improvements, just no exciting ones.
6G was not going to be a good topic anyway. It is nowhere near ready. Most people we spoke to agreed that work on the standard will really begin in earnest next year. The 3GPP, the group that runs the standards process, scheduled a working conference in March 2025, and that is likely to be the starting gun for the technical work required.
With that start date, a likely target for completion of the standard is probably 2028, at the earliest, which likely means no deployments until 2029 or 2030.
This is just fine for the operators. As noted, they are in no hurry to incur more capex. On the other hand, this is not good news for the vendors. Companies like Ericsson, Nokia and Qualcomm have come to rely on new standards to boost growth. Their revenue will not go to zero for the next five years, but winter is coming.
Further complicating matters, no one really seems to have any big ideas for what 6G could include. Until recently, most people talking about 6G tended to focus on new spectrum bands that could be used. And these included some pretty crazy ideas. Terahertz band, for example, which seemed to attract a lot of attention a few years ago, will offer incredible data rates, so long as users are in the same room as the base station and there are no other people blocking the signal. Those kinds of ideas hold the most appeal for the deep technical crowd whose goal is to design cool things. The more…let’s call them ‘commercially minded’ are starting to think about more… let’s call them practical solutions. But as we have noted in the past there are no easy answers for boosting wireless performance for 6G. Expect lots of small tweaks, only arriving by the end of the decade.
And of course, a lot could still happen to delay the standard. Not least is everyone’s favorite spoiler - geopolitics. The 3GPP has been reviewing its voting policies recently. In part, this was spurred by a certain large handset vendor who tried to game the process by signing up legions of its employees to the voting bodies. Another factor, was a certain country all of whose representatives, no matter what company they worked for, tended to vote in a bloc, ensuring leadership positions for members from that country. No reason to be coy - we are talking about Apple and China respectively. Our guess is that the engineering work on 6G will progress fairly steadily despite all this, but we could still end up with delays in the later stages if the politics get unsettled at the upper layers of the 3GPP.
Open RAN
Readers may remember that over last year we have spent a lot of time diving deep into the Open RAN initiative (ORAN). Since then the project has won a lot of headlines and seems to have made great strides. In reality, our sense is that it has largely peaked. More precisely, ORAN’s promise of ‘opening up’ telecom networks is not going to arrive. That does not mean ORAN will go away, but it is role in networks is changing. Instead of providing an architecture upon which operators can design a new type of network, ORAN looks likely to instead become a set of features available in some equipment.
Almost every equipment maker (with one notable exception) has some form of ORAN in their gear. Smaller vendors all include it in their marketing material, but the appeal to operators is not clear. From what we can tell, certain parts of the network will essentially have ORAN access where smaller vendors can plug in their hardware or software offering. But critically, each vendor’s ORAN is likely to be a little different and access to the core interfaces will still be controlled by the major equipment vendors, who will thus continue to control the critical pools of economics.
A major development in ORAN circles came two months ago, when Ericsson won a big contract to supply all of AT&T’s radio gear (i.e. the RAN, that ORAN is supposed to open). As we predicted, Ericsson has used that deal to assert itself in ORAN circles. We heard many examples of Ericsson throwing its weight around in those kinds of meetings. We are still totally convinced they have a printed manual showing employees how to disrupt these kinds of initiatives.
The hope in the ORAN community has long ben to get their ideas written into the 6G standard. That would have been a major boost for the project, but that now seems almost certain not to happen. For starters, any version that could get added to the standard would have been heavily influenced by Ericsson. More importantly, above we mentioned one vendor has never once mentioned Open RAN and seems to deny its existence. That vendor is Huawei. And despite all their travails, they retain a major influence in the market and within the standards bodies. If they refuse to admit that ORAN even exists, it is not going to make it into 6G.
Catatonia in Catalonia: Qualcomm
We try not to play at psychological analysis in our technology analysis, but we have to ask - Qualcomm, is everything ok? Because it definitely felt like something is not right.
Mobile World Congress is one of the company’s premier events to showcase their capabilities, but this year it did not feel like anyone at the company felt much like celebrating.
A few things stood out from the get go. Apparently, their investor relations team did not make the trip. We understand that a lot of investors skipped the show this year, but plenty still made it. At least enough to merit some IR participation.
Their booth is usually a hive of activity. A prime location in the heart of the show has always meant the Qualcomm booth is a good place to meet and so attracts a large crowd, but it felt empty this year. Nor can we blame people for not spending time in the booth - there was not much to see.
MWC is a place for Qualcomm to highlight its new technologies, especially the deeply esoteric ones that only matter to the deep telecom standards crowd who attend the show. There was none of that this year. Or at least nothing significant. Qualcomm usually puts out over a dozen press releases ahead of MWC, this year they had four.
The content in the booth was fairly anemic. About a third of the space was entirely empty. A big corner was dedicated to their automotive products, but there was nothing new on this front at the show. The most animated spot was the team from Humane demo’ing heir AI pin. A good example of Qualcomm technology at work, but what does it say when another company grans the spotlight in your own booth.
As mentioned above, the Qualcomm technical content was light. We actually could not find anyone in the booth capable of discussing the technical elements of their RF products. They just rolled out some interesting integated RF/Modem chips, but no one could actually hold a real conversation about them. The word “Afer Thought” seems to apply here.
This being 2024, they of course had an AI demo featuring a phone in airplane mode running stable diffusion. Fine, except we saw essentially the same demo six months ago. And just down the hall Mediatek had a much, much more compelling demo running Stable Diffusion in real time, while Qualcomm’s demo took 20 to 30 seconds to render.
Admittedly, most of this is qualitative and circumstantial, but we have been to enough of these shows to sense a disturbance in the force.
And there were plenty of more substantive problems
First, the company’s marketing effort seemed out of focus At CES, they talked about out cars, not PCs. Despite the launch of their CPUs just months prior, they had no PCs to display. Then at MWC, all they wanted to talk about was PCs. We attended an industry analyst briefing and a good two thirds of the discussion centered on PCs and AI. In fairness they talked a lot about work that they are doing with developers to build the software ecosystem around their Windows PCs. This is important and a big development for the company, but why are we talking about it at a telecom show. Are they going to focus on smartphones at Computex?
The company did have one significant mobile announcement at the show - a new initiative called “Gigabit for Everyone”, and here is where the company seemed to go from depressed to courting self-harm. At heart, this initiative boils down to low pried chips for $99 phones. As we often discuss, for years Qualcomm and Mediatek have existed in a stable competitive equilibrium. Qualcomm sells into high end phones, Mediatek sells into mid and low priced phones, leaving them to mostly compete for phones priced around $300 to $600. Qualcomm now seems to want to disrupt that, pitting them against not only Mediatek but also Unisoc who competes with Mediatek for the cheapest phones. The mere fact that we mention Unisoc and Qualcomm in the same sentence is a problem. Qualcomm has tried an failed many times to get into this market. Admittedly, this time the customer pool is much more concentrated which lightens Qualcomm’s support burden, a key stumbling block back when there were hundreds of customers down in this part of the market.
This strategy has two flaws. First, why pick a fight with Mediatek? That company has gotten a lot better, see above about AI, and so we have to imagine they might respond by giving Qualcomm a tougher run for their money in higher priced phones. Secondly, how much margin can Qualcomm really pick up down here? What would motivate such a move? We have to think that this implies Qualcomm is deeply concerned about its revenue prospects and is now scrounging around in the sofa cushions looking to pick up spare change.
All of this is frustrating because the people we speak with at the company are all working very hard. We do not want to imply that the rank and file are slacking off in any way. But we do now have serious questions about upper management. Where is their head at? It does not seem to be in mobile, the category also known as 80% of Qualcomm’s revenue.
The Limits of Home Grown Silicon?
At last year’s show, we dug deep into the market for the chips going into mobile infrastructure. Last year’s show marked a moment of some intense competitive dynamics as the the chip vendors and their customers the major equipment vendors were poised to make some big decisions. Those questions got resolved in the ensuing months and we think we can learn something reflecting back on the market.
For semis, one of the notable changes that 5G brought to the network comes in the radio access network (RAN). In particular, the latest base stations require a fair amount of compute-heavy processing. We are not going to dig into the details of the architecture here, but instead going to focus on the need for a network processing. There are three major silicon vendors for these sockets - Marvell, Intel and Qualcomm. Ericsson uses a homegrown solution, but worked closely with Intel on the part and uses x86 elsewhere in the gear. Nokia has its own part as well, but they are open about the fact that Marvell helped them design and bring the part to manufacturing. We think Samsung uses Marvell. We do not know what Huawei is doing, no one really wants to talk about it. They could be using an internal part or Marvell chips stored away years ago. Qualcomm has design wins with Fujitsu and Vietel.
Looking at that list, it should be clear that this is really a two-way race between Intel and Marvell. Qualcomm had a new chip for the segment at the show, but this part did not seem to merit its own press release. Qualcomm has always been ambivalent about the market for mobile infrastructure.. It is not really a market they handle well. They tend to enter the market at the start of each new ‘G’, only to quietly walk away a few years later. That history seems likely to play out again this time. As far as we can tell, there is nothing wrong with their part, it actually seems fairly competitive on paper, but they always struggle to win customers. Both Ericsson and Nokia tend to keep their distance from the company, and long tail customers like Fujitsu and Vietel do not provide enough revenue to support the products. Qualcomm seemed to have hoped that 5G deployments would go a different direction - mmWave and more ‘Open’ networks - and when that did not play out their networking products are left a bit stranded.
Beyond this, we have questions about Nokia and Ericsson’s efforts. Both companies have a long history of designing their own silicon going back to the days when they were vertically integrated suppliers of both handsets and network infrastructure. Without the volume that handsets provide, the economics of designing their own chips became more challenging. Yet they persist. Our thesis on homegrown silicon holds that it only makes sense when the chip in question conveys some form of strategic advantage. It is not clear that Ericsson’s and Nokia’s chip provide that. The processing taking place here is not particularly differentiated. It is not easy, but by the same token the differences between Nokia’s chips and Ericsson’s are not major. Operators do not pick one over the other because of these differences. In fact, our sense is that the operators do not like the burden of having to work with different code bases and would greatly prefer some form of common interfaces.
Few very companies can afford the expense of rolling their own silicon. And tot every company should build their own chips. We have to wonder if Nokia and Ericsson, given all their other challenges will continue down this path.
China Inc. Returns in Force
Covid restrictions in China were eased in late 2022, by which time deadlines for booking space at MWC 2023 had long since passed. As a result, very few companies from China made it to the show in 2023, or had greatly reduced presence. This year, China Inc. made up for lost time.
Handset brands Honor and Xiaomi had massive, prominent booths in some of the best real estate at the Fira. Honor, in particular positioned itself right next to Samsung, with a better spot. Always different, Oppo had a booth too, but in the worst real estate in the show - the bottom floor of the last building, but at least they were there.
More significant than the big brands were all the smaller equipment vendors The companies that make antennas and connectors and all kinds of miscellaneous radio boxes. Most notable about this was extent to which these companies have upped their marketing game. These companies all seem to be trying to build brands in way that they have not done in the past.
The best example of this was the Yangtze Optical Fibre and Cable Company (YOFC) who had one of the most visually appealing booths in the show. China’s consumer-facing branded companies have been building big shiny booths for a while, in our archives we have a photo of one Huawei booth with a three story chandelier, but YOFC is a company that makes near-commodity fiber strands. And they were only the most prominent example, many of their peers made big efforts as well.
At one level, this could be seen as an encouraging sign for these companies. They are maturing and developing their capabilities. On the other hand, we also detected a deeper layer of anxiety. China’s economy is not in good shape right now, and the government wants high-tech companies to export the country out of the doldrums. Spending a few million dollars for a splashy booth is not going to change that dynamic. So while YOFC probably was able to attract some new business with its giant booth, they are still just one of a dozen companies selling fiber optic cables, they cannot all win share.
The Return of Huawei
One of the highlights for us this show was the opportunity to reconnect with Huawei’s corporate communications team. Between the 2019 sanctions, the pandemic and the 2022 sanctions, we lost touch with official Huawei channels. Say what you want about them, they remain a significant force in the industry and we think it is important to be able to speak to them directly.
We ended up spending half a day exploring their sizable booth, and got to speak to many people staffing it. They had one of, if not the largest booths at the show, almost a trade sow within the trade show. The booth had dozens of meeting rooms, a print center, close to a hundred interpreters and an outdoor seating area that reminded us of a Mediterranan beach club . We heard one estimate that they spent $100 million on the event and had a 1,000 staff on site. Both sound high, but not implausible.
The public portion of their booth focused on consumer products such as their phones and home networking kit.
Beyond the gates, which were staffed by hosts with impeccable posture, the focus shifted to telecom, enterprise and data center products.
After the launch of the Mate P60 and the HiSilicon Kirin 910s chip, Huawei seemed to flip a switch on their marketing efforts, going from very quiet to everywhere almost overnight. This effort was very much on display at MWC.
That being said, as much as Huawei is trying to paint a picture of a renewed company, undaunted by its challenges, we detected many notes of a deeper anxiety.
For starters, the company seems somewhat uncertain about how the US restrictions are playing out. For instance, we could not find any material talking about their HarmonyOS. This show would have been a good way to get more people familiar with their homegrown operating system, but it looks like the current plan is to restrict Harmony to their domestic market only. Another example, last year we noticed that every singled chip in all of the printed circuit boards in the public part of booth had been taped over. That much effort, individually taping over hundreds of chips was conspicuous. So this year, they did not display a single board anywhere in their booth. None of this sounds like a company with a deep level of confidence.
Moreover, we were not particularly excited by their product offerings. Huawei has a long history of developing interesting products. They had 5G gear on the market before Nokia and Ericsson. In past years, they had a wide range of new technologies and proposed architectures. This year’s products on display seemed largely to consist of modest updates to those past products. Of course, there are obvious reasons why Huawei does not have the ability to lead the market in the same way as before the 2019 restrictions kicked in. The company’s proponents will argue that the fact that they have any gear on display is a victory in its own right. And there is some truth to that.
Our sense is that the company is still on the defensive to a large degree, uncertain about its place in the market. At the same time, it is important to note that their booth was crowded. They drew a huge crowd of operators from around the world.
IoT - For Trees
Over the years, we have heard every conceivable pitch for Internet of Things (IoT) applications. And as much as we believe that eventually many more physical objects will get connected to the Internet we are as weary as everyone else by the degree to which IoT marketing seems to have crept into every category, no matter how absurd or unlikely.
So when someone invited us to the booth for Dryad, marketed as IoT for trees, we had to visit, fully expecting that it would be a struggle to keep a straight face. IoT for Trees seems like a late night brainstorming idea from an overworked marketing team, turning to humor to lighten their load.
And yet, by the end of our meeting with Dryad, we came away convinced. This is actually a pretty good idea.
Dryad is a privately held company based in Germany. They make a small solar-powered smoke detector. Rugged enough to be left in the forest for years. The module has a LoRa radio which connects back to a gateway which has LTE Cat M and satellite links for backhaul to the Internet.
As residents of California, we really like the sound of better detectors for forest fires. The company has over a dozen trials going on around the world in places like Greece, Italy, Australia, Northern California, Canada and Indonesia - all places which have been hit by horrific fires in recent years.
Obligatory AI Discussion
Not surprisingly, the topic of AI came up a lot at MWC. A Lot. However, the people who build telecom networks have to deal with analog circuits and radio waves, and we have noticed that seems to convey a dose of immunity to AI Washing. To be clear, every executive and marketing key note espoused some form of AI, but on the ground and in our meetings, the subject came up rarely.
We encountered many examples of “Telco AI”, from Huawei’s Telecom AI to SK’s Telecom LLM. Microsoft challenged us to “Visualize the future of telco with AI”. That sounds great, except the accompanying demo was just a computer running Co-Pilot image generation of colorful, happy robots.
There are no obvious AI applications in telecom right now. Yes, customer service chatbots and the like, but nothing really differentiated. That does not mean advances in machine learning will not come to telecom, but as we have said those gains will be much less observable. They bring small but meaningful improvements, not revolution.
That being said, we did meet with a company that has a really interesting use case for large language, GPT-style models.
That company is Intraway, a privately held maker of OSS software. OSS is one of those obscure corners of telecoms, a billion dollar market which no one has ever heard of. OSS systems provide administrative functions the operators use to manage their networks, a partner of BSS systems which handle billing.
Amdocs, one of the leaders in the field has always had a booth at MWC in one of the most central locations, but in almost 20 years of attending the show we have never been in that booth. We have not taken a meeting with an OSS/BSS company in ten years. The space is important but also sleepy and proprietary.
Enter Intraway. They started ten years ago with a new approach to OSS, and by new, we mean building a system that used software tools which Internet companies had already been using for a decade - cloud, no code, etc. New for telecom. The company has recently added a machine learning function to its stack.
One of the competitive barriers sheltering the space is the need to tie OSS systems into the networking gear provided by the big equipment makers like Ericsson. Tying software to these systems is time consuming and labor intensive (by design, see above regarding Open RAN). Anyone entering the space needs to first build a mountain of software integrations into all that obscure-by-design gear. Intraway trained an AI model on its own software. Then it feeds configuration manuals for all that telecom gear as a query into the model. The output is an integration module plugging their software features, or cartridges, into the operators’ systems. Months of engineering time done in the amount of time it takes to generate a picture of kittens wearing outfits of the Dutch Masters.
This looks like a very promising way to use LLMs. The incumbents may be able to find ways to block access, but over the years the existing OSS/BSS companies have largely become professional services operations. Adopting these practices cuts directly into their revenue model, charging operators for all those months of bespoke engineering. We are not close enough to the OSS/BSS space to be able to handicap Inraway’s prospects, and we can still one up with a half dozen hurdles. Nonetheless, we think it is an interesting, novel use of AI that could become significant in this one corner of the industry.
If you like this content, you should check out our podcast The Circuit
Thank you for reading D2D Newsletter. This post is public so feel free to share it.