WEBVTT
1
00:00:00.100 --> 00:00:07.000
[rock music] Hi there. I'm Dan Jones, editor of, uh, Next TMT Networks.
2
00:00:07.860 --> 00:00:24.439
This week I'm talking with Earl Lum of EJL Wireless, and we're talking about data centers, Nokia's AI RAN push, and Elon Musk's small cell ambitions. So let's carry on. Hi there, Earl.
3
00:00:25.640 --> 00:00:41.570
Hi, Dan. Nice to be here, and thanks for inviting me. Good to meet you. Well, I've met you before, but, you know, good to speak to you again. So let's talk about the data center situation right now, particularly- So...
4
00:00:41.620 --> 00:00:57.160
Oh, go ahead. I'm sorry. No, yeah, yeah. Particularly, um, since New York has, has basically set a moratorium of a year against large megawatt and gigawatt data center building.
5
00:00:57.840 --> 00:01:07.160
Do you think that's gonna have an effect on the entire market, particularly as other states could bring similar bans into play?
6
00:01:08.440 --> 00:01:37.020
So th- this whole thing, I guess Executive Order 62 that, uh, the governor of New York signed off, it's almost eerily reminding me of the small cell legislation that happened many years ago where every state had to follow what the FCC was talking about with regards to a shot clock, maximum caps on, uh, the rent and the revenue that could be, uh, given to a town or a state or a city- Right...
7
00:01:37.040 --> 00:01:51.880
with regards to small cell poles and siting and the shot clock about 30 days and nothing longer. So there was one state, I don't remember which one, that was the first one to, uh, put those rules in effect. But- Right...
8
00:01:51.940 --> 00:01:59.140
this could be, New York could be a template, especially the environmental statutes and the requirements that are in there.
9
00:01:59.220 --> 00:02:11.980
So the one-year moratorium, according to the document I was reading, is essentially to be able to do an environmental impact report on exactly what any data center is gonna be able to, to do.
10
00:02:12.000 --> 00:02:29.860
But there's also a lot of requirements going forward for every several years, I think starting in 2028 or 2029, to where a, a certain percentage of the data center's energies have to come from renewable sources, and that ratchets us- Oh, wow...
11
00:02:29.880 --> 00:02:43.160
up to where by 2040, oh, 90-plus percent of the data centers that get approved in New York State specifically have to have renewable energy in process and actually deployed by that point in time.
12
00:02:43.220 --> 00:02:48.060
So it goes from 33% to 66% to then 90-plus percent.
13
00:02:48.180 --> 00:03:00.810
So there has to be a very, uh, strategic plan for the data centers to figure out how they're going to get renewable energy set up and then execute on that plan in order to- Right...
14
00:03:00.820 --> 00:03:04.460
meet, uh, the requirements that the state is going to impose.
15
00:03:04.540 --> 00:03:19.170
I'm not sure what the fines would be if they don't meet it, but, uh, again, uh, the environmental, the environmental side of it is a critical part, uh, as we look forward to potentially other states such as Texas, uh, to look for templates to- What the fines would be...
16
00:03:19.180 --> 00:03:29.880
really focus on essentially the, uh, the issues of the town, the people that are near the data centers, what the environmental impact will be, water usage.
17
00:03:29.960 --> 00:03:40.540
All of these things that everyone's seen on the news are now explicitly outlined in the, the requirements that, uh, New York has set up during this moratorium. Right.
18
00:03:41.280 --> 00:03:58.880
And that could really have an impact on what other mor- moratoriums come out, because, say, like, you wouldn't think that Texas would be, uh, one of the states, but I've seen rumblings around Texas possibly bringing a, a, a ban on, on things as well.
19
00:03:59.000 --> 00:04:09.960
And, and what energy the, the data center uses is clearly a, a, a massive issue in this, in that situation. I mean, absolutely.
20
00:04:10.060 --> 00:04:22.079
Governor Abbott has expressed concerns about the data center growth in Texas, specifically Texas, for most people who aren't aware of it, they have their own grid, so they're not attached to the national grid.
21
00:04:22.180 --> 00:04:31.880
So they have a problem in that it's important that whatever data center projects that are ongoing and proceeding in Texas have their own, uh, source of energy.
22
00:04:32.000 --> 00:04:42.480
And if you look at, I think it was Amazon or whoever's building the world's largest natural gas energy plant there, you, you will have an environmental impact from that.
23
00:04:42.520 --> 00:04:50.340
But it, it is somewhat renewable from the sense of the, it's not being, it's not using up the energy from the grid.
24
00:04:50.420 --> 00:05:00.420
And for Texas, I think specifically, uh, they need to take that into account for all of the mega gigawatt data center projects that are happening over there in that state. Right.
25
00:05:00.430 --> 00:05:15.080
With the megawatt possible ban, bans or moratoriums, does that mean that there's a, a, a place for edge data centers to move in, or?
26
00:05:15.180 --> 00:05:25.060
Well, a- according to the document for New York at least- Right... anything that's a data center, which is a megaw- a megawatt and above would count.
27
00:05:25.780 --> 00:05:36.290
I think there may be some leeway if you're in the 5 or 10 megawatt versus a 20 and above that they highlight specifically as a, quote, "large data center." Yep.
28
00:05:36.360 --> 00:05:51.030
So I, I think that's all going to need to be reviewed by each and every state to figure out w- w- what is the environmental impact for a 5 to 10 mewa- megawatt data center versus the ones that are in the hundreds and, and the gigawatt level.
29
00:05:51.030 --> 00:06:01.300
Mm-hmm. And again, you're talking about hundreds of acres of land versus, uh, maybe 2,000, 5,000 square feet. Now, can you-
30
00:06:02.324 --> 00:06:18.724
Tell me, or maybe you can't, but you said you were working with a, uh, an edge data center, and they were trying to produce a, a data center in a box kind of situation. What, what can you tell me about that? Yes.
31
00:06:18.864 --> 00:06:29.984
So there are many startup companies out there, uh, one of which I'm am working with that- Right... have a different approach to what is going to actually be needed at the edge. Yeah.
32
00:06:30.174 --> 00:06:45.284
A- and these are data centers that would be sub 10 megawatts. Right, right. Uh, and essentially you're talking about a fully- Oh, man... uh, contained data center in a container, a 25 or 40-foot container.
33
00:06:46.004 --> 00:06:54.754
Uh, and instead of using air as the primary method for getting the heat out of the system- Yep...
34
00:06:55.174 --> 00:07:09.824
uh, these systems would be liquid-cooled inside, so they would have zero water impact on the environment for in any location that they would be placed, and the water situation is obviously a major issue in every state and every community that's out there.
35
00:07:10.304 --> 00:07:23.664
Yeah, yeah, for sure. Let's just, um, talk about the Elon Musk small cell plans for just a moment. I mean, do they even make sense particularly? Well, uh, I'm...
36
00:07:23.724 --> 00:07:37.324
You've obviously read some of my quotes from articles that have been out there, but I- Right... guess the issue is for his plan for doing small cells is integrating the small cell with the actual Starlink terminal.
37
00:07:38.024 --> 00:07:53.514
Right. So you would have a small cell transmitting 4G, 5G signal in addition to the transmission and receiving of the signals to the Starlink satellite network, and again, that would cover essentially a home- Right...
38
00:07:53.514 --> 00:08:07.944
uh, potentially, depending on the power levels of that small cell. The issue with that is if that is mounted on a rooftop for that existing Starlink user, how does that user use mobility services outside of their home?
39
00:08:08.864 --> 00:08:18.564
Right. And would there need to be another terminal mounted to their car somehow in order- Yep... to continue to provide that signal as they leave their home?
40
00:08:19.244 --> 00:08:29.304
Or, you know, it's not really a mobile service if they can't leave their house. Yeah. Yeah. It's like, uh, we're indoors with Starlink.
41
00:08:29.824 --> 00:08:35.484
[laughs] E- exactly, and again, line of sight is required for satellite communication services. Yep, yep.
42
00:08:35.544 --> 00:08:45.684
So even for the directed device, if you're underground, in a parking lot, in a shopping mall, or any enclosed area, it's not going to work. Well, that's...
43
00:08:45.764 --> 00:09:00.744
I mean, we're clearly gonna be covering that for probably the next year in permutations of, of what actually happens. But, uh, from the, from what I can tell and from what I've researched and all of that,
44
00:09:02.384 --> 00:09:13.904
it's gonna be very difficult to make this, this situation work. Right, a- and I think if, if you talk to any mobile operator in the world- Right... in any country in the world,
45
00:09:15.124 --> 00:09:28.824
small cells are the last thing on the to-do list- Right... always and forever. So to start with that topology doesn't make logical sense, at least for me, as a wireless guy. Right.
46
00:09:28.944 --> 00:09:43.574
I, I think it'd be easier for SpaceX to buy out and renegotiate all of the Dish Wireless leases on all the towers in the US that the FCC set aside a huge fund for, since, uh, they don't wanna pay up, and- Yeah...
47
00:09:43.604 --> 00:09:50.364
they've reneged on all of the leases. So those towers are already out there. The locations are already out there.
48
00:09:51.084 --> 00:10:03.214
It would be easy to repopulate those towers with macro cell equipment from a vendor or from whoever, or Open RAN, whatever it- Right... needs to be, a- and then you would have a footprint.
49
00:10:03.284 --> 00:10:07.664
But that alone does not solve the problem that SpaceX has with their strategy.
50
00:10:07.704 --> 00:10:23.074
They need roaming, because no operator when they're starting off, just like 1&1 in Germany, uh, when they were at Greenfield, you need a national roaming agreement with somebody because your subscribers expect that they can go anywhere and everywhere with their phone.
51
00:10:24.124 --> 00:10:39.844
Right. Right. Including indoors. Indeed. Indeed. Let's also talk about another controversial topic, Nokia's AI RAN and the, and the kind of launch that they're planning fall this year into 2027.
52
00:10:40.764 --> 00:10:52.224
Now, you've talked about NVIDIA as their crucial supplier for their GPUs, clearly. Um, will they be able to supply enough for, for Nokia?
53
00:10:53.334 --> 00:11:08.044
Nokia right now on the baseband side for their silicon partner, uh, is using or is partnered with Marvell. Right. And then for the radios, they are using both Marvell as well as Broadcom. Right.
54
00:11:08.084 --> 00:11:15.684
For, this is for the ReefShark platform of ASICs. Yeah, I mean, basically that, that isn't the...
55
00:11:15.724 --> 00:11:29.984
What, um, the ReefShark stuff that, uh, Nokia has now, right, is not the NVIDIA GPU that it's planning to use for its AI RAN stuff, or not solely that, right? Correct.
56
00:11:30.404 --> 00:11:44.914
And that is, again, an alternative to the existing ReefShark products that Nokia has- Right... and, uh, is meant to be as an upgrade path for the future for mobile operators to slide out one of their cards.
57
00:11:44.944 --> 00:11:56.444
And, you know, I figured show and tell is really good. I actually have one of the Nokia cards that would go- Oh, wow... into the AirScale, and basically s- it's several pounds. And
58
00:11:57.484 --> 00:12:10.794
essentially the card that NVIDIA is going to make, the AVIG for GPU, gonna look something like this. Okay. So again, it's just one card that has the GPU in it.
59
00:12:11.984 --> 00:12:22.044
And the operator will then have to make a choice as to how many of these cards they want to purchase to put into every one of their Nokia Airscale base stations. Right...
60
00:12:22.084 --> 00:12:36.984
depending on their requirements and their capacity needs and what they expect from this AI or this GPU chip to do that's different from the existing chips that Marvell is already supplying. Right.
61
00:12:39.104 --> 00:12:51.264
And I imagine that's probably gonna be a- an extra expense for operators as well because- Well, ob- obviously these cards, they have to pay for them. Right And the question is, is what are they getting?
62
00:12:52.473 --> 00:13:04.094
I would assume the performance has to be at least 25% better than the current cards they have to justify the cost of wanting to swap them out and, and do the upgrade. So- Right.
63
00:13:04.164 --> 00:13:14.224
I mean, Nokia had said 100 times better spectrum efficiency at one point, right? Yes, but I don't understand what kind of physics they're using- [laughs] Okay...
64
00:13:14.264 --> 00:13:27.704
as opposed to the physics everyone else is, is using out there. And, uh, a lot of it could be lab simulations. Right And the real question is what happens when you're out in the field?
65
00:13:27.764 --> 00:13:32.214
And I think that's where we'll see some of the data coming out of T-Mobile's- Mm...
66
00:13:32.264 --> 00:13:45.344
field trials by the end of this year, and that will really tell us or not tell us exactly what the performance really looks like on the base stations when they're out in the field with actual traffic- Right... on them.
67
00:13:45.374 --> 00:13:57.634
And then what is the power consumption? Obviously they don't have a B100 or B200 GPU or a more advanced GPU on these cards because the power envelope would not be able to handle that.
68
00:13:57.724 --> 00:14:10.004
Right So they have a much stripped down version of that GPU. But I would argue it's a processor, and ASIC always wins out over a processor, and that was the problem with Open RAN,
69
00:14:10.984 --> 00:14:28.074
in that you were trying to catch up with Open RAN off-the-shelf processors that could never keep up or, uh, pass an Ericsson or a Huawei or a ZTE or- Right. Right. Right... someone else's ASIC. Yeah. So it'll work.
70
00:14:28.484 --> 00:14:42.344
Will it be the most efficient is still the crucial question that needs to be answered. Right. And are operators gonna spend to roll this out in, like, nowhere Wyoming?
71
00:14:43.304 --> 00:14:57.384
I think the question really is for the operator, what kind of AI do they really need- Right... in their RAN? The for, the in, the on, whatever flavor, uh, of that AI is really going to be crucial for them.
72
00:14:57.564 --> 00:15:06.644
I mean, today, if you look, there's already AI being used for energy efficiencies, for putting the radios to sleep- Right... all of these other things.
73
00:15:06.724 --> 00:15:19.774
So there is another layer of AI that could be utilized to improve that radio link. Uh, and I guess according to Ericsson, you can buy their software subscription, which is another way to say- Mm...
74
00:15:19.774 --> 00:15:25.804
that you will have to pay more money- Yep... I'm sure on both the radio and on the base band because they have to work together.
75
00:15:26.544 --> 00:15:34.024
So the cost of that, uh, is again going to be a- another expense for the operator for every radio for every site.
76
00:15:34.124 --> 00:15:45.054
So if you look at a Verizon or, or AT&T with their 95,000 or 100,000 sites and the number of radios, that's going to get expensive. And then the question- Right...
77
00:15:45.204 --> 00:16:02.084
is, is how much money is it really saving or what is it actually doing in your network to justify the cost? Right. I mean, I definitely see operators will use AI RAN, but I see the...
78
00:16:02.104 --> 00:16:18.144
Probably they're gonna roll it out in cities and, and like Beijing, Shanghai, New York, London, Paris. I don't think they're gonna roll it out in the countryside. I just don't, I just don't see that. Think again.
79
00:16:18.224 --> 00:16:23.864
What, what is the reasoning and the purpose that they want to roll out the technology? Is it...
80
00:16:24.474 --> 00:16:50.094
There already is energy efficiencies and savings that are implemented in today's radios for 5G that have some level of machine learning or initial AI in them, so that's already being rolled out and available to any operator today, and the operators are already taking advantage of those sleep modes and everything else regarding energy, 'cause that was low-hanging fruit- Sure...
81
00:16:50.104 --> 00:17:00.614
for the OEMs to input into the radios and into the base stations. So the question is, is to get that next level of functionality and efficiency, where do they want to deploy that?
82
00:17:00.664 --> 00:17:12.384
And like you said, I would agree that they're more worried about, uh, the efficiencies in a metro market, right? Sure. Like New York City or Boston or San Francisco or LA, uh, or Dallas.
83
00:17:12.764 --> 00:17:19.084
Those are the cities that have massive MIMO primarily- Yeah... which is where you want, you're going to want to deploy this.
84
00:17:19.164 --> 00:17:33.064
But if we look at a European market, the European market is vastly different than the United States or North America- Yeah... in that massive MIMO only constitutes roughly 20 to 25% of the total network over there.
85
00:17:33.204 --> 00:17:47.534
So the rest of it is just macro radios where you may not necessarily need to deploy, uh, unless you're doing it for energy, in which case you care about every site because it all adds up for you as operational expenses as an operator.
86
00:17:48.434 --> 00:18:03.744
Right. Right. That brings the AI RAN stuff into perspective at least a little bit. We should, uh, wrap it up here, I think, for the time being. Thank you so much for your time, Earl. Thank you for having me, Dan.
87
00:18:03.824 --> 00:18:16.934
That was Earl Lum from EJAL Wireless. This is nextTM 10x Rocks. We come to you every Tuesday with a podcast. Tune in next time. We'll be here. [upbeat rock music]
1
00:00:00.100 --> 00:00:07.000
[rock music] Hi there. I'm Dan Jones, editor of, uh, Next TMT Networks.
2
00:00:07.860 --> 00:00:24.439
This week I'm talking with Earl Lum of EJL Wireless, and we're talking about data centers, Nokia's AI RAN push, and Elon Musk's small cell ambitions. So let's carry on. Hi there, Earl.
3
00:00:25.640 --> 00:00:41.570
Hi, Dan. Nice to be here, and thanks for inviting me. Good to meet you. Well, I've met you before, but, you know, good to speak to you again. So let's talk about the data center situation right now, particularly- So...
4
00:00:41.620 --> 00:00:57.160
Oh, go ahead. I'm sorry. No, yeah, yeah. Particularly, um, since New York has, has basically set a moratorium of a year against large megawatt and gigawatt data center building.
5
00:00:57.840 --> 00:01:07.160
Do you think that's gonna have an effect on the entire market, particularly as other states could bring similar bans into play?
6
00:01:08.440 --> 00:01:37.020
So th- this whole thing, I guess Executive Order 62 that, uh, the governor of New York signed off, it's almost eerily reminding me of the small cell legislation that happened many years ago where every state had to follow what the FCC was talking about with regards to a shot clock, maximum caps on, uh, the rent and the revenue that could be, uh, given to a town or a state or a city- Right...
7
00:01:37.040 --> 00:01:51.880
with regards to small cell poles and siting and the shot clock about 30 days and nothing longer. So there was one state, I don't remember which one, that was the first one to, uh, put those rules in effect. But- Right...
8
00:01:51.940 --> 00:01:59.140
this could be, New York could be a template, especially the environmental statutes and the requirements that are in there.
9
00:01:59.220 --> 00:02:11.980
So the one-year moratorium, according to the document I was reading, is essentially to be able to do an environmental impact report on exactly what any data center is gonna be able to, to do.
10
00:02:12.000 --> 00:02:29.860
But there's also a lot of requirements going forward for every several years, I think starting in 2028 or 2029, to where a, a certain percentage of the data center's energies have to come from renewable sources, and that ratchets us- Oh, wow...
11
00:02:29.880 --> 00:02:43.160
up to where by 2040, oh, 90-plus percent of the data centers that get approved in New York State specifically have to have renewable energy in process and actually deployed by that point in time.
12
00:02:43.220 --> 00:02:48.060
So it goes from 33% to 66% to then 90-plus percent.
13
00:02:48.180 --> 00:03:00.810
So there has to be a very, uh, strategic plan for the data centers to figure out how they're going to get renewable energy set up and then execute on that plan in order to- Right...
14
00:03:00.820 --> 00:03:04.460
meet, uh, the requirements that the state is going to impose.
15
00:03:04.540 --> 00:03:19.170
I'm not sure what the fines would be if they don't meet it, but, uh, again, uh, the environmental, the environmental side of it is a critical part, uh, as we look forward to potentially other states such as Texas, uh, to look for templates to- What the fines would be...
16
00:03:19.180 --> 00:03:29.880
really focus on essentially the, uh, the issues of the town, the people that are near the data centers, what the environmental impact will be, water usage.
17
00:03:29.960 --> 00:03:40.540
All of these things that everyone's seen on the news are now explicitly outlined in the, the requirements that, uh, New York has set up during this moratorium. Right.
18
00:03:41.280 --> 00:03:58.880
And that could really have an impact on what other mor- moratoriums come out, because, say, like, you wouldn't think that Texas would be, uh, one of the states, but I've seen rumblings around Texas possibly bringing a, a, a ban on, on things as well.
19
00:03:59.000 --> 00:04:09.960
And, and what energy the, the data center uses is clearly a, a, a massive issue in this, in that situation. I mean, absolutely.
20
00:04:10.060 --> 00:04:22.079
Governor Abbott has expressed concerns about the data center growth in Texas, specifically Texas, for most people who aren't aware of it, they have their own grid, so they're not attached to the national grid.
21
00:04:22.180 --> 00:04:31.880
So they have a problem in that it's important that whatever data center projects that are ongoing and proceeding in Texas have their own, uh, source of energy.
22
00:04:32.000 --> 00:04:42.480
And if you look at, I think it was Amazon or whoever's building the world's largest natural gas energy plant there, you, you will have an environmental impact from that.
23
00:04:42.520 --> 00:04:50.340
But it, it is somewhat renewable from the sense of the, it's not being, it's not using up the energy from the grid.
24
00:04:50.420 --> 00:05:00.420
And for Texas, I think specifically, uh, they need to take that into account for all of the mega gigawatt data center projects that are happening over there in that state. Right.
25
00:05:00.430 --> 00:05:15.080
With the megawatt possible ban, bans or moratoriums, does that mean that there's a, a, a place for edge data centers to move in, or?
26
00:05:15.180 --> 00:05:25.060
Well, a- according to the document for New York at least- Right... anything that's a data center, which is a megaw- a megawatt and above would count.
27
00:05:25.780 --> 00:05:36.290
I think there may be some leeway if you're in the 5 or 10 megawatt versus a 20 and above that they highlight specifically as a, quote, "large data center." Yep.
28
00:05:36.360 --> 00:05:51.030
So I, I think that's all going to need to be reviewed by each and every state to figure out w- w- what is the environmental impact for a 5 to 10 mewa- megawatt data center versus the ones that are in the hundreds and, and the gigawatt level.
29
00:05:51.030 --> 00:06:01.300
Mm-hmm. And again, you're talking about hundreds of acres of land versus, uh, maybe 2,000, 5,000 square feet. Now, can you-
30
00:06:02.324 --> 00:06:18.724
Tell me, or maybe you can't, but you said you were working with a, uh, an edge data center, and they were trying to produce a, a data center in a box kind of situation. What, what can you tell me about that? Yes.
31
00:06:18.864 --> 00:06:29.984
So there are many startup companies out there, uh, one of which I'm am working with that- Right... have a different approach to what is going to actually be needed at the edge. Yeah.
32
00:06:30.174 --> 00:06:45.284
A- and these are data centers that would be sub 10 megawatts. Right, right. Uh, and essentially you're talking about a fully- Oh, man... uh, contained data center in a container, a 25 or 40-foot container.
33
00:06:46.004 --> 00:06:54.754
Uh, and instead of using air as the primary method for getting the heat out of the system- Yep...
34
00:06:55.174 --> 00:07:09.824
uh, these systems would be liquid-cooled inside, so they would have zero water impact on the environment for in any location that they would be placed, and the water situation is obviously a major issue in every state and every community that's out there.
35
00:07:10.304 --> 00:07:23.664
Yeah, yeah, for sure. Let's just, um, talk about the Elon Musk small cell plans for just a moment. I mean, do they even make sense particularly? Well, uh, I'm...
36
00:07:23.724 --> 00:07:37.324
You've obviously read some of my quotes from articles that have been out there, but I- Right... guess the issue is for his plan for doing small cells is integrating the small cell with the actual Starlink terminal.
37
00:07:38.024 --> 00:07:53.514
Right. So you would have a small cell transmitting 4G, 5G signal in addition to the transmission and receiving of the signals to the Starlink satellite network, and again, that would cover essentially a home- Right...
38
00:07:53.514 --> 00:08:07.944
uh, potentially, depending on the power levels of that small cell. The issue with that is if that is mounted on a rooftop for that existing Starlink user, how does that user use mobility services outside of their home?
39
00:08:08.864 --> 00:08:18.564
Right. And would there need to be another terminal mounted to their car somehow in order- Yep... to continue to provide that signal as they leave their home?
40
00:08:19.244 --> 00:08:29.304
Or, you know, it's not really a mobile service if they can't leave their house. Yeah. Yeah. It's like, uh, we're indoors with Starlink.
41
00:08:29.824 --> 00:08:35.484
[laughs] E- exactly, and again, line of sight is required for satellite communication services. Yep, yep.
42
00:08:35.544 --> 00:08:45.684
So even for the directed device, if you're underground, in a parking lot, in a shopping mall, or any enclosed area, it's not going to work. Well, that's...
43
00:08:45.764 --> 00:09:00.744
I mean, we're clearly gonna be covering that for probably the next year in permutations of, of what actually happens. But, uh, from the, from what I can tell and from what I've researched and all of that,
44
00:09:02.384 --> 00:09:13.904
it's gonna be very difficult to make this, this situation work. Right, a- and I think if, if you talk to any mobile operator in the world- Right... in any country in the world,
45
00:09:15.124 --> 00:09:28.824
small cells are the last thing on the to-do list- Right... always and forever. So to start with that topology doesn't make logical sense, at least for me, as a wireless guy. Right.
46
00:09:28.944 --> 00:09:43.574
I, I think it'd be easier for SpaceX to buy out and renegotiate all of the Dish Wireless leases on all the towers in the US that the FCC set aside a huge fund for, since, uh, they don't wanna pay up, and- Yeah...
47
00:09:43.604 --> 00:09:50.364
they've reneged on all of the leases. So those towers are already out there. The locations are already out there.
48
00:09:51.084 --> 00:10:03.214
It would be easy to repopulate those towers with macro cell equipment from a vendor or from whoever, or Open RAN, whatever it- Right... needs to be, a- and then you would have a footprint.
49
00:10:03.284 --> 00:10:07.664
But that alone does not solve the problem that SpaceX has with their strategy.
50
00:10:07.704 --> 00:10:23.074
They need roaming, because no operator when they're starting off, just like 1&1 in Germany, uh, when they were at Greenfield, you need a national roaming agreement with somebody because your subscribers expect that they can go anywhere and everywhere with their phone.
51
00:10:24.124 --> 00:10:39.844
Right. Right. Including indoors. Indeed. Indeed. Let's also talk about another controversial topic, Nokia's AI RAN and the, and the kind of launch that they're planning fall this year into 2027.
52
00:10:40.764 --> 00:10:52.224
Now, you've talked about NVIDIA as their crucial supplier for their GPUs, clearly. Um, will they be able to supply enough for, for Nokia?
53
00:10:53.334 --> 00:11:08.044
Nokia right now on the baseband side for their silicon partner, uh, is using or is partnered with Marvell. Right. And then for the radios, they are using both Marvell as well as Broadcom. Right.
54
00:11:08.084 --> 00:11:15.684
For, this is for the ReefShark platform of ASICs. Yeah, I mean, basically that, that isn't the...
55
00:11:15.724 --> 00:11:29.984
What, um, the ReefShark stuff that, uh, Nokia has now, right, is not the NVIDIA GPU that it's planning to use for its AI RAN stuff, or not solely that, right? Correct.
56
00:11:30.404 --> 00:11:44.914
And that is, again, an alternative to the existing ReefShark products that Nokia has- Right... and, uh, is meant to be as an upgrade path for the future for mobile operators to slide out one of their cards.
57
00:11:44.944 --> 00:11:56.444
And, you know, I figured show and tell is really good. I actually have one of the Nokia cards that would go- Oh, wow... into the AirScale, and basically s- it's several pounds. And
58
00:11:57.484 --> 00:12:10.794
essentially the card that NVIDIA is going to make, the AVIG for GPU, gonna look something like this. Okay. So again, it's just one card that has the GPU in it.
59
00:12:11.984 --> 00:12:22.044
And the operator will then have to make a choice as to how many of these cards they want to purchase to put into every one of their Nokia Airscale base stations. Right...
60
00:12:22.084 --> 00:12:36.984
depending on their requirements and their capacity needs and what they expect from this AI or this GPU chip to do that's different from the existing chips that Marvell is already supplying. Right.
61
00:12:39.104 --> 00:12:51.264
And I imagine that's probably gonna be a- an extra expense for operators as well because- Well, ob- obviously these cards, they have to pay for them. Right And the question is, is what are they getting?
62
00:12:52.473 --> 00:13:04.094
I would assume the performance has to be at least 25% better than the current cards they have to justify the cost of wanting to swap them out and, and do the upgrade. So- Right.
63
00:13:04.164 --> 00:13:14.224
I mean, Nokia had said 100 times better spectrum efficiency at one point, right? Yes, but I don't understand what kind of physics they're using- [laughs] Okay...
64
00:13:14.264 --> 00:13:27.704
as opposed to the physics everyone else is, is using out there. And, uh, a lot of it could be lab simulations. Right And the real question is what happens when you're out in the field?
65
00:13:27.764 --> 00:13:32.214
And I think that's where we'll see some of the data coming out of T-Mobile's- Mm...
66
00:13:32.264 --> 00:13:45.344
field trials by the end of this year, and that will really tell us or not tell us exactly what the performance really looks like on the base stations when they're out in the field with actual traffic- Right... on them.
67
00:13:45.374 --> 00:13:57.634
And then what is the power consumption? Obviously they don't have a B100 or B200 GPU or a more advanced GPU on these cards because the power envelope would not be able to handle that.
68
00:13:57.724 --> 00:14:10.004
Right So they have a much stripped down version of that GPU. But I would argue it's a processor, and ASIC always wins out over a processor, and that was the problem with Open RAN,
69
00:14:10.984 --> 00:14:28.074
in that you were trying to catch up with Open RAN off-the-shelf processors that could never keep up or, uh, pass an Ericsson or a Huawei or a ZTE or- Right. Right. Right... someone else's ASIC. Yeah. So it'll work.
70
00:14:28.484 --> 00:14:42.344
Will it be the most efficient is still the crucial question that needs to be answered. Right. And are operators gonna spend to roll this out in, like, nowhere Wyoming?
71
00:14:43.304 --> 00:14:57.384
I think the question really is for the operator, what kind of AI do they really need- Right... in their RAN? The for, the in, the on, whatever flavor, uh, of that AI is really going to be crucial for them.
72
00:14:57.564 --> 00:15:06.644
I mean, today, if you look, there's already AI being used for energy efficiencies, for putting the radios to sleep- Right... all of these other things.
73
00:15:06.724 --> 00:15:19.774
So there is another layer of AI that could be utilized to improve that radio link. Uh, and I guess according to Ericsson, you can buy their software subscription, which is another way to say- Mm...
74
00:15:19.774 --> 00:15:25.804
that you will have to pay more money- Yep... I'm sure on both the radio and on the base band because they have to work together.
75
00:15:26.544 --> 00:15:34.024
So the cost of that, uh, is again going to be a- another expense for the operator for every radio for every site.
76
00:15:34.124 --> 00:15:45.054
So if you look at a Verizon or, or AT&T with their 95,000 or 100,000 sites and the number of radios, that's going to get expensive. And then the question- Right...
77
00:15:45.204 --> 00:16:02.084
is, is how much money is it really saving or what is it actually doing in your network to justify the cost? Right. I mean, I definitely see operators will use AI RAN, but I see the...
78
00:16:02.104 --> 00:16:18.144
Probably they're gonna roll it out in cities and, and like Beijing, Shanghai, New York, London, Paris. I don't think they're gonna roll it out in the countryside. I just don't, I just don't see that. Think again.
79
00:16:18.224 --> 00:16:23.864
What, what is the reasoning and the purpose that they want to roll out the technology? Is it...
80
00:16:24.474 --> 00:16:50.094
There already is energy efficiencies and savings that are implemented in today's radios for 5G that have some level of machine learning or initial AI in them, so that's already being rolled out and available to any operator today, and the operators are already taking advantage of those sleep modes and everything else regarding energy, 'cause that was low-hanging fruit- Sure...
81
00:16:50.104 --> 00:17:00.614
for the OEMs to input into the radios and into the base stations. So the question is, is to get that next level of functionality and efficiency, where do they want to deploy that?
82
00:17:00.664 --> 00:17:12.384
And like you said, I would agree that they're more worried about, uh, the efficiencies in a metro market, right? Sure. Like New York City or Boston or San Francisco or LA, uh, or Dallas.
83
00:17:12.764 --> 00:17:19.084
Those are the cities that have massive MIMO primarily- Yeah... which is where you want, you're going to want to deploy this.
84
00:17:19.164 --> 00:17:33.064
But if we look at a European market, the European market is vastly different than the United States or North America- Yeah... in that massive MIMO only constitutes roughly 20 to 25% of the total network over there.
85
00:17:33.204 --> 00:17:47.534
So the rest of it is just macro radios where you may not necessarily need to deploy, uh, unless you're doing it for energy, in which case you care about every site because it all adds up for you as operational expenses as an operator.
86
00:17:48.434 --> 00:18:03.744
Right. Right. That brings the AI RAN stuff into perspective at least a little bit. We should, uh, wrap it up here, I think, for the time being. Thank you so much for your time, Earl. Thank you for having me, Dan.
87
00:18:03.824 --> 00:18:16.934
That was Earl Lum from EJAL Wireless. This is nextTM 10x Rocks. We come to you every Tuesday with a podcast. Tune in next time. We'll be here. [upbeat rock music]
