WEBVTT - Microsoft Cuts The Power To AI 0:00:02.480 --> 0:00:03.560 As media. 0:00:04.640 --> 0:00:20.160 Hello you, it's better offline. I'm your host ed Zeitron. Now, 0:00:20.160 --> 0:00:22.000 I've talked about the pay all horses of the AI 0:00:22.079 --> 0:00:24.119 apocalypse in the past, and these are the events that 0:00:24.160 --> 0:00:26.680 signify that the general VAI era is coming to an end. 0:00:26.920 --> 0:00:29.840 And the subject to this episode and its second part, well, 0:00:29.880 --> 0:00:33.280 I believe represents the biggest paylest horses of the entire flock. 0:00:34.040 --> 0:00:36.680 So on February twenty first analyst TD Cohen revealed that 0:00:36.680 --> 0:00:40.120 Microsoft had canceled leases and I quote totaling a couple 0:00:40.200 --> 0:00:43.159 hundred megawatts with at least two private data center operators 0:00:43.200 --> 0:00:46.879 across multiple US markets canceled. The report also detailed how 0:00:46.920 --> 0:00:50.479 Microsoft pulled back on converting negotiated and signed statements of 0:00:50.560 --> 0:00:54.920 qualifications sqq's, which are added was precursors to a data 0:00:54.920 --> 0:00:57.640 center lease, so effectively the first step before you really 0:00:57.680 --> 0:01:00.320 agreed to something, well the last step, I guess. Although 0:01:00.360 --> 0:01:03.200 the analyst, which is part of the TD Bank company, 0:01:03.480 --> 0:01:06.520 added it was unclear whether Microsoft might convert these sqqs 0:01:06.520 --> 0:01:10.200 in the future. These generally close close to one hundred 0:01:10.240 --> 0:01:13.480 percent of the time. According to them, canceling these was 0:01:13.480 --> 0:01:18.640 effectively rather unusual. Now TD Cohen also added that Microsoft 0:01:18.680 --> 0:01:22.080 was reallocating a considerable portion of its projected international spend 0:01:22.080 --> 0:01:24.759 to the US, which suggests to DD Cohen that there 0:01:24.800 --> 0:01:27.880 was a material slow down in international leasing for Microsoft, 0:01:29.319 --> 0:01:32.720 But one crucial, teeny tiny part of the report was 0:01:32.760 --> 0:01:36.040 missed by just about everybody. I'm going to read directly 0:01:36.080 --> 0:01:37.800 from the report, and you'll probably be able to tell 0:01:37.800 --> 0:01:40.480 from the tone of my voice which the most pertinent 0:01:40.520 --> 0:01:44.520 part is. Ah. As we highlighted in our recent takeaway 0:01:44.560 --> 0:01:48.720 from PTC, the Pacific Telecommunications Council Conference, we learned via 0:01:48.760 --> 0:01:51.400 our channel checks that Microsoft one walked away from multiple 0:01:51.400 --> 0:01:53.920 one hundred plus make what deals in multiple markets that 0:01:53.920 --> 0:01:57.680 were in early to mid stages and negotiations to let 0:01:57.680 --> 0:02:01.640 one gigawatt of loys on larger footoprint sites expire, and 0:02:01.720 --> 0:02:04.200 three walked away from at least five land parcels that 0:02:04.280 --> 0:02:07.800 it had under contract in multiple Tier one markets. What 0:02:07.880 --> 0:02:10.560 TD Cohen is saying is not just that Microsoft canceled 0:02:10.560 --> 0:02:13.919 some data centers, but the Microsoft also effectively canceled over 0:02:13.960 --> 0:02:16.840 a gigawatt of data center operations on top of the 0:02:16.840 --> 0:02:21.160 previously reported multiple one hundred plus what megawat deals. If 0:02:21.160 --> 0:02:24.040 we add in the land under contract, which is indeterminate. 0:02:24.080 --> 0:02:26.240 Based on what tdcens has said and the deals that 0:02:26.280 --> 0:02:29.160 were in flight, the total capacity likely amounts to even 0:02:29.200 --> 0:02:32.480 more than a gigawat. For some context, data sent to 0:02:32.560 --> 0:02:35.360 Dynamics reported that Microsoft had five gigawats of data center 0:02:35.400 --> 0:02:38.360 capacity in April twenty twenty four, saying that Microsoft had 0:02:38.360 --> 0:02:40.400 also planned to add one gigawat of capacity by the 0:02:40.400 --> 0:02:42.760 October twenty twenty four and another one and a half 0:02:42.840 --> 0:02:45.440 gigawats of capacity by the first half of twenty twenty five. 0:02:46.080 --> 0:02:48.520 Based on this reporting, one can estimate the Microsoft as 0:02:48.520 --> 0:02:51.200 somewhere between six and seven and a half gigawatts of 0:02:51.280 --> 0:02:54.399 capacity at this time. As a result, based on td 0:02:54.480 --> 0:02:58.959 Cohen's analyst analysis, even Microsoft has thro recombination of canceled leases, 0:02:59.040 --> 0:03:02.800 pullbacks and statements of qualifications, cancelations of land parcels, and 0:03:02.840 --> 0:03:06.920 deliberate expiration of letters of intent. That's those lois. They've 0:03:06.919 --> 0:03:10.800 effectively abandoned data center expansion equivalent to over fourteen percent 0:03:10.840 --> 0:03:14.880 of their current capacity. It's completely bloody insane, and this 0:03:14.960 --> 0:03:17.120 story just kind of sat there. It kind of not 0:03:17.280 --> 0:03:20.320 the market's confidence. But I don't know, I'm a lot 0:03:20.360 --> 0:03:22.920 more worried about this than I think people are. And 0:03:23.480 --> 0:03:25.320 really I don't like telling people how to feel, but 0:03:25.720 --> 0:03:28.200 this kind of worries me. Okay, now I've thrown a 0:03:28.240 --> 0:03:30.520 lot of new terminology at you. But before we move on, 0:03:30.639 --> 0:03:33.440 let's explain some terms, because they're essential to understanding why 0:03:33.480 --> 0:03:35.880 this is all such a big deal. First of all, 0:03:36.200 --> 0:03:39.560 letter of intent or LOI in this context is a 0:03:39.600 --> 0:03:42.280 statement that an entity intends to lease or buy land 0:03:42.360 --> 0:03:44.840 or power from a data center. These can be binding 0:03:44.840 --> 0:03:47.800 on non binding. A letter of intent is serious, though, 0:03:47.880 --> 0:03:50.160 and walking away from one is not something you do idly. 0:03:50.200 --> 0:03:53.240 It's not like not answering an email. Now we'll go 0:03:53.280 --> 0:03:56.480 on to SQQS statements of qualifications. These set the terms 0:03:56.520 --> 0:03:59.920 and conditions of a lease. While they do not themselves constitutally, 0:04:00.120 --> 0:04:02.480 they convert into sign leases, as I mentioned, at an 0:04:02.480 --> 0:04:05.280 almost one hundred percent rate according to TD Cohen, and 0:04:05.320 --> 0:04:07.480 are generally used as a signal to the landowner to 0:04:07.520 --> 0:04:11.520 start construction. Basically, they're the green light before the green lane. 0:04:12.280 --> 0:04:15.360 As for Tier one markets, these are markets for hyperscale growth, 0:04:15.440 --> 0:04:18.760 helped by favorable conditions like power, land and cabling. From 0:04:18.800 --> 0:04:20.760 what I can tell, there's no fixed list of which 0:04:20.800 --> 0:04:23.120 cities are Tier one in which aren't, but they include 0:04:23.120 --> 0:04:26.400 obvious candidates like London, Singapore, as well as Northern Virginia, 0:04:26.400 --> 0:04:29.480 which is the largest tube of data centers in the world. Finally, 0:04:29.600 --> 0:04:33.640 we're talking power, megawatt and gigawe. This one is really important, 0:04:33.640 --> 0:04:37.719 but it's also really confusing. Data center capacity is measured 0:04:37.760 --> 0:04:40.400 not by the amount of computations the facility can handle, 0:04:40.440 --> 0:04:43.160 but rather by power capacity, and that makes sense because 0:04:43.200 --> 0:04:46.000 power capacity is directly linked to the capabilities of the facility, 0:04:46.160 --> 0:04:49.080 with more power capacity allowing for more servers or more 0:04:49.080 --> 0:04:52.120 power hungry chips, of course, and because chips themselves are 0:04:52.120 --> 0:04:55.120 constantly getting faster and more power efficient. Power makes a 0:04:55.160 --> 0:04:58.320 little more sense. If you mentioned in terms of computations 0:04:58.320 --> 0:04:59.920 per second, you'd likely have a number that fluck you 0:05:00.240 --> 0:05:03.320 as hardware as upgraded and decommissioned. When you hear megawat 0:05:03.360 --> 0:05:05.559 or gigawatt in this episode, assume that we're talking about 0:05:05.560 --> 0:05:09.680 capacity and not power generation unless I say otherwise. Now, 0:05:09.680 --> 0:05:11.720 with that out of the way, let's talk more about 0:05:11.720 --> 0:05:12.240 this report. 0:05:24.200 --> 0:05:24.360 Now. 0:05:24.360 --> 0:05:26.719 The numbers in the TD Cohen Report, which I'll link 0:05:26.760 --> 0:05:29.960 in the episode spreadsheet, of course, heavily suggest that Microsoft, 0:05:30.000 --> 0:05:33.280 the biggest purchaser of Nvidio GPUs and, according to Ted 0:05:33.440 --> 0:05:36.159 Cohen and I quote, the most active data center LESE 0:05:36.320 --> 0:05:38.640 of capacity in twenty twenty three and the first half 0:05:38.680 --> 0:05:41.440 of twenty twenty four, does not believe that there's future 0:05:41.640 --> 0:05:44.560 growth in generative AI, nor does it have faith in, 0:05:44.680 --> 0:05:47.680 nor does it want responsibility for the future of open AI. 0:05:48.480 --> 0:05:51.240 Data center buildouts take three to six years to complete, 0:05:51.279 --> 0:05:54.960 and the largest hyperscalar facilities can easily cost several billion dollars, 0:05:55.120 --> 0:05:58.320 meaning that these moves are extremely forward looking. You don't 0:05:58.360 --> 0:06:00.440 just build a data center for the demand GI now, 0:06:00.480 --> 0:06:02.640 but for the demand you expect further down the line. 0:06:03.040 --> 0:06:05.920 This suggests that Microsoft believes its current infrastructure and its 0:06:06.000 --> 0:06:08.960 likely scaled back plans for expansion will be sufficient for 0:06:09.080 --> 0:06:12.480 a movement that Sachynadella once called a golden age for systems. 0:06:12.640 --> 0:06:15.160 He did that less than a year ago. To quote 0:06:15.160 --> 0:06:18.520 TD Cohen again, the magnitude of both potential data center 0:06:18.560 --> 0:06:21.280 capacity Microsoft walked away from the decision to pull back 0:06:21.279 --> 0:06:25.320 on land acquisition, which supports core long term capacity growth, 0:06:25.360 --> 0:06:27.560 in our view, indicates the loss of a major demand 0:06:27.560 --> 0:06:30.359 signal that Microsoft was originally responding to, and that we 0:06:30.440 --> 0:06:33.320 believe the shift is in their appetite for capacity is 0:06:33.360 --> 0:06:36.400 tied to open AI. To explain here, TD Cowen is 0:06:36.400 --> 0:06:39.800 effectively saying that Microsoft is responding to a major demand signal, 0:06:39.880 --> 0:06:42.240 and said major demand signal is saying you do not 0:06:42.400 --> 0:06:45.919 need more data centers. Said demand signal that Microsoft was 0:06:46.160 --> 0:06:48.880 responding to, in TD Cohen's words, is its appetite for 0:06:48.920 --> 0:06:52.320 capacity to provide servers to open AI. And it seems 0:06:52.320 --> 0:06:54.720 that said appetite is waning and Microsoft no longer wants 0:06:54.760 --> 0:06:58.160 to build out data centers for America's most swagged out 0:06:58.200 --> 0:07:00.719 AI guy. Now, I say that kind of as a joke, 0:07:01.160 --> 0:07:03.440 and I do think that he's more swagged out than 0:07:03.440 --> 0:07:06.919 Mark Zuckerberg. Mark Zuckerberg's trying too hard. However, Sam Moltman 0:07:07.040 --> 0:07:10.600 is more damp than Mark Zuckerberg, so ultimately Zuckerberg wins. Now, 0:07:10.600 --> 0:07:12.560 I want to make it clear that Microsoft is effectively 0:07:12.600 --> 0:07:15.760 cutting its data center expansion by over a gigawatt of capacity, 0:07:15.840 --> 0:07:18.920 if not more, and it's impossible to reconcile these cuts 0:07:19.040 --> 0:07:21.920 with the expectation that generative AI will be this massive, 0:07:22.000 --> 0:07:26.240 transformative technological phenomenon. I believe that the reason that Microsoft 0:07:26.280 --> 0:07:28.240 is coming back is that it does not have the 0:07:28.280 --> 0:07:31.120 appetite to provide further data center expansion for open Ai, 0:07:31.360 --> 0:07:34.120 and it's having doubts about the future of generative AI 0:07:34.160 --> 0:07:37.120 as a whole. If Microsoft believed that there was a 0:07:37.160 --> 0:07:40.440 massive opportunity in supporting open AI's further growth, or that 0:07:40.520 --> 0:07:43.800 it had massive demand for generative AI services, there'd be 0:07:43.840 --> 0:07:46.640 no reason to cancel capacity, let alone cats or such 0:07:46.680 --> 0:07:50.440 a significant amount. These moves also suggest that Microsoft is 0:07:50.480 --> 0:07:53.560 walking away from building and training further large frontier models 0:07:53.800 --> 0:07:57.320 like chat GPT's GPT four point five now and from 0:07:57.320 --> 0:08:01.640 supporting doing so for others. Remember, Microsoft has significantly more 0:08:01.680 --> 0:08:04.240 insight into the current health and growth of GENERATIVEAI than 0:08:04.280 --> 0:08:07.880 any other company. Remember they have full access to all 0:08:07.920 --> 0:08:10.880 of open AI's tech, probably the future stuff too, not 0:08:10.920 --> 0:08:13.840 that there's much they know. All their research too, Microsoft 0:08:13.880 --> 0:08:17.080 knows something we don't. As open AI's largest backer and 0:08:17.120 --> 0:08:19.920 infrastructural partner and the owners of the server architecture where 0:08:19.960 --> 0:08:22.600 they train them ultra expensive models, not to mention, the 0:08:22.680 --> 0:08:26.280 largest shareholder in OpenAI, Microsoft can see exactly what is 0:08:26.400 --> 0:08:28.560 or isn't coming down the pike on top of having 0:08:28.600 --> 0:08:30.920 a view into both the sales of its own GENERATIVEAI 0:08:31.000 --> 0:08:34.280 powered software such as Microsoft three sixty five copiler, and 0:08:34.320 --> 0:08:36.800 sales of both model services and cloud compute for other 0:08:36.840 --> 0:08:39.400 models run on Microsoft A zero, which is their cloud 0:08:39.440 --> 0:08:43.640 platform in plain English, Microsoft, which arguably has more data 0:08:43.679 --> 0:08:45.800 than anybody else about the health of the generative AI 0:08:45.840 --> 0:08:48.160 industry and its potential for growth, has decided that it 0:08:48.200 --> 0:08:50.239 needs to dramatically slow down its expansion. 0:08:51.440 --> 0:08:51.520 Now. 0:08:51.600 --> 0:08:54.760 To be clear, this expansion, I really am hammering this 0:08:54.800 --> 0:08:56.360 home a lot, but I need you to understand this 0:08:56.720 --> 0:09:00.600 is absolutely necessary for generative AI to continue evolving expanding, 0:09:00.679 --> 0:09:02.600 even if it only does so in ways that kind 0:09:02.600 --> 0:09:05.480 of do the same thing again and again. Now, before 0:09:05.520 --> 0:09:06.960 we move on, I want to make it clear that 0:09:07.000 --> 0:09:09.800 I'm not saying that Microsoft has stopped building data centers. 0:09:09.880 --> 0:09:11.800 I've said it a few times, but these projects take 0:09:11.960 --> 0:09:14.720 years three to six years to complete and are far 0:09:15.320 --> 0:09:18.200 far in advance with their planning, and Microsoft does have 0:09:18.240 --> 0:09:20.520 a few big projects in the works. One plan three 0:09:20.640 --> 0:09:23.480 hundred and twenty four megawatt Microsoft Data center in Atlanta 0:09:23.559 --> 0:09:26.000 is expected to cost one zero point eight billion dollars 0:09:26.120 --> 0:09:27.680 and as far as I know, this deal is still 0:09:27.720 --> 0:09:32.520 in flight, however, and this was cited separately by TD Cohen. 0:09:32.840 --> 0:09:35.560 Microsoft has recently paused construction on parts of its three 0:09:35.600 --> 0:09:39.040 point three billion dollar data center campus in Mount Pleasant, Wisconsin. 0:09:39.559 --> 0:09:41.920 While Microsoft had tried to reassure locals that the first 0:09:41.920 --> 0:09:43.760 phase of the project was on course to be complete 0:09:43.800 --> 0:09:46.240 on time, its justification of delaying the rest of it 0:09:46.360 --> 0:09:48.960 was well not brilliant, and it was to give Microsoft 0:09:48.960 --> 0:09:52.280 an opportunity to evaluate and I quote the project's scope 0:09:52.960 --> 0:09:55.720 and recent changes in technology, and consider how this might 0:09:55.800 --> 0:09:59.680 impact the design of its facilities. Oh, bob, oh, buddy, 0:09:59.720 --> 0:10:02.800 that's not good. You don't want it. No. One evaluates 0:10:02.800 --> 0:10:05.640 the scope and then goes, oh, actually the scope is great. 0:10:05.720 --> 0:10:08.760 I love it. Nor do they think about impacts and 0:10:08.800 --> 0:10:12.160 go oh, let's do more. No, no, no, anyway, the 0:10:12.200 --> 0:10:15.000 same registered article I'm citing here adds that and I 0:10:15.120 --> 0:10:17.920 quote the review process may include the need to negotiate 0:10:17.960 --> 0:10:20.520 some building permits, potentially placing another hurdle in the way 0:10:20.520 --> 0:10:23.280 of the project. The register did add that Microsoft said 0:10:23.280 --> 0:10:27.040 it expected to complete one hyperscaler data center in Mount 0:10:27.040 --> 0:10:31.000 Pleasant as originally planned, though its capacity wasn't available. Arguably, 0:10:31.040 --> 0:10:34.880 Microsoft would expand its data center infrastructure. Anyway, as more 0:10:34.920 --> 0:10:37.280 stuff moves to the cloud and our dependence grows on it, 0:10:37.559 --> 0:10:40.640 Microsoft and other providers need to build capacity. The organic 0:10:40.640 --> 0:10:44.679 growth is natural and sadly inevitable. However, Microsoft's plans for 0:10:44.760 --> 0:10:47.480 data center expansion were far far in excess of that 0:10:47.559 --> 0:10:50.280 natural growth, and perhaps we're seeing a pullback from those 0:10:50.400 --> 0:10:54.320 stated extravagances into something perhaps a little more reasonable. 0:10:55.800 --> 0:10:55.920 Now. 0:10:55.960 --> 0:10:58.640 I've talked a lot about megawards and gigawats, and you're 0:10:58.800 --> 0:11:01.120 not in the data center business, and you should be. 0:11:01.160 --> 0:11:03.200 The parties are an absolute laugh. This can all seem 0:11:03.240 --> 0:11:06.800 a bit abstract, so let's put it into context. Without context, 0:11:06.800 --> 0:11:09.800 it's hard to understand how big and one hundred megawat 0:11:09.880 --> 0:11:12.720 data center is. These are some of the biggest. According 0:11:12.720 --> 0:11:16.280 to the International Energy Agency, Small data centers can consume 0:11:16.320 --> 0:11:19.160 anywhere between one and five megawats. These are, for the 0:11:19.200 --> 0:11:23.120 most part, average size facilities, perhaps not for cloud compute giants, 0:11:23.160 --> 0:11:25.160 but for other companies. It's kind of part of the course. 0:11:25.520 --> 0:11:28.880 One hundred megawats, by comparison, is huge. It's the equivalent 0:11:28.920 --> 0:11:31.480 of the annual energy consumption of between three hundred and 0:11:31.480 --> 0:11:34.200 fifty thousand, four hundred thousand electric cars, and I know 0:11:34.280 --> 0:11:35.840 some sort of pedant it's going to say, it's not 0:11:35.960 --> 0:11:38.280 the same thing, shout out a fuck up. Go outside, 0:11:38.320 --> 0:11:42.400 stop listen, go outside, go outside now, go do something anyway, 0:11:42.640 --> 0:11:45.040 Although there are others that will likely dwarf what we 0:11:45.240 --> 0:11:48.000 today considered to be a large facility, META is in 0:11:48.040 --> 0:11:50.559 the process of constructing, for example, a ten billion dollar 0:11:50.679 --> 0:11:55.080 data center campus in Louisiana with a proposed two gigawat capacity. Still, 0:11:55.120 --> 0:11:57.760 whatever way you cut it, and one hundred megawatt facility 0:11:57.800 --> 0:12:00.640 is big and it's a big long term investment. Cushman 0:12:00.679 --> 0:12:03.560 and Wakefield's twenty twenty four global data center market comparison 0:12:03.640 --> 0:12:07.520 gives some chilling context into how significant Microsoft's pullback is. 0:12:08.120 --> 0:12:10.480 A gigawatt of data set to capacity is roughly the 0:12:10.640 --> 0:12:13.480 entire operational IT load of Tokyo, which has a one 0:12:13.480 --> 0:12:17.160 point to eight gigawatt capacity, or London nine hundred and 0:12:17.200 --> 0:12:19.679 ninety six megawatts, or the Bay Area, which only has 0:12:19.679 --> 0:12:23.120 eight hundred and forty two megawatts. These are actually very large. 0:12:23.880 --> 0:12:26.480 It's just that Microsoft got rid of so much more. 0:12:26.960 --> 0:12:29.800 And again, the total figure of canceled or abandoned capacity 0:12:29.840 --> 0:12:32.760 is likely far higher than a gigawatt. That number only 0:12:32.800 --> 0:12:35.640 accounts for the letters of intent that Microsoft allowed to expire. 0:12:35.840 --> 0:12:38.120 It doesn't include everything else that the two data sentences 0:12:38.280 --> 0:12:40.800 already killed, or the land pass was it abandoned, or 0:12:40.840 --> 0:12:43.559 the deals that were in early to mid stages of negotiation. 0:12:44.080 --> 0:12:46.520 Imagine walking away from two London's or two Tokyos of 0:12:46.520 --> 0:12:48.880 capacity and it not being a massive deal. This is 0:12:48.920 --> 0:12:53.000 a huge flipping deal. Microsoft is not simply walking back 0:12:53.040 --> 0:12:56.920 some future plans. It's effectively canceling what it loudly insisted 0:12:57.040 --> 0:13:00.320 was the future. If you think this sounds hyperbolic, consider this. 0:13:00.760 --> 0:13:03.840 London and Tokyo are respectively the biggest data center markets 0:13:03.840 --> 0:13:05.920 in Europe and Asia, according to the same Cushman and 0:13:05.960 --> 0:13:09.439 Wakefield report. Canceling city's worth of capacity at a time when 0:13:09.520 --> 0:13:14.760 artificial intelligence is supposedly revolutionizing everything certainly suggests that artificial 0:13:14.800 --> 0:13:19.520 intelligence isn't really revolutionizing anything now. One other detail in 0:13:19.559 --> 0:13:22.280 TD Cohen's report really stood out to me. While this 0:13:22.440 --> 0:13:25.400 pullback in Microsoft's data center leasing, It's also seen a 0:13:25.440 --> 0:13:29.240 commensurate rise in demand from Oracle related to the Stargate project, 0:13:29.400 --> 0:13:32.760 a relatively new partnership of up to five hundred billion 0:13:32.760 --> 0:13:35.640 dollars stopped saying it's five hundred billion dollars to build 0:13:35.679 --> 0:13:39.120 massive new data centers for Ai, specifically for one company 0:13:39.400 --> 0:13:42.080 led by soft Bank and of course open Ai with 0:13:42.160 --> 0:13:44.640 investment from Oracle and MGX and one hundred billion dollar 0:13:44.720 --> 0:13:48.080 investment fund backed by the United Arab Emirates. Open Ai 0:13:48.120 --> 0:13:51.160 has committed eighteen to nineteen billion dollars to the Stargate project, 0:13:51.720 --> 0:13:55.360 money it doesn't have, meaning that part of the twenty 0:13:55.360 --> 0:13:57.920 five to forty billion dollars that they're raising at the 0:13:57.920 --> 0:14:00.160 moment will be committed to funding these data centers, and 0:14:00.240 --> 0:14:03.080 less as I'll get to later. Open Ai raises more 0:14:03.080 --> 0:14:05.960 in debt. Leading the round is SoftBank, which is also 0:14:06.000 --> 0:14:08.679 committing eighteen to nineteen billion dollars, as well as creating 0:14:08.720 --> 0:14:11.880 a joint venture fund called sb open Ai Japan to 0:14:11.920 --> 0:14:14.920 offer open Ai services to the Japanese market, something that 0:14:14.960 --> 0:14:17.360 I thought was already happening, as well as spending three 0:14:17.400 --> 0:14:20.480 billion dollars annually to use open AI's technology across its 0:14:20.520 --> 0:14:24.000 group businesses. According to The Wall Street Journal. In simpler terms, 0:14:24.040 --> 0:14:26.640 soft Bank is investing as much as I think it's 0:14:26.640 --> 0:14:28.760 going to be like thirty billion dollars in open Ai, 0:14:28.880 --> 0:14:31.760 then spending another three billion dollars a year on software 0:14:31.800 --> 0:14:34.480 that only loses money and still hallucinates and shows no 0:14:34.680 --> 0:14:38.160 side of getting meaningfully better or more reliable. Well a 0:14:38.160 --> 0:14:40.600 soft bank actually sees value in open AI's tech, or 0:14:40.600 --> 0:14:42.720 whether this purchase dealer is a subsidy by the back 0:14:42.760 --> 0:14:45.720 door is open to debate. Given that three billion dollars 0:14:45.800 --> 0:14:48.640 is equivalent to open AI's entire revenue from selling premium 0:14:48.640 --> 0:14:51.520 access to Chatch GPT in twenty twenty four, which included 0:14:51.560 --> 0:14:53.880 some major deals with the likes of Price Waterhouse Coopers, 0:14:54.560 --> 0:14:57.240 I'm inclined to believe the latter. Even then, how is 0:14:57.280 --> 0:15:00.760 it feasible that SoftBank can continue paying to get the 0:15:00.800 --> 0:15:03.760 deal done? Microsoft changed the terms of its exclusive relationship 0:15:03.760 --> 0:15:05.960 with open Ai to allow it to work with Oracle 0:15:05.960 --> 0:15:08.440 to build out further data centers full of GPUs necessary 0:15:08.440 --> 0:15:12.360 to power open AI's big, shitty, unprofitable and unsustainable models. 0:15:12.880 --> 0:15:16.080 The open Ai Oracle stargate situation was a direct result, 0:15:16.240 --> 0:15:19.200 according to reporting from the Information of open Ai becoming 0:15:19.200 --> 0:15:22.240 frustrated with Microsoft and not providing it with service fast enough, 0:15:22.440 --> 0:15:25.520 including an allotment of three hundred thousand of nvidious GBT 0:15:25.640 --> 0:15:27.720 two hundred chips by the end of twenty twenty five. 0:15:29.280 --> 0:15:31.120 For what it's worth, The Wall Street Journal reports that 0:15:31.160 --> 0:15:34.560 Microsoft was getting increasingly frustrated with open AI's constant demands 0:15:34.560 --> 0:15:37.480 for more compute. The relationship between the two entities had 0:15:37.520 --> 0:15:40.600 start to fray, with both sides feeling kind of aggrieved. This, 0:15:40.720 --> 0:15:44.120 combined with Microsoft's data center pullback, heavily suggests that Microsoft 0:15:44.160 --> 0:15:47.160 is no longer interested in being open AI's infrastructure paypig 0:15:47.760 --> 0:15:50.960 long term, at least after all it was if it was, 0:15:51.000 --> 0:15:54.240 I mean it funded and support open AI's expansion rather 0:15:54.280 --> 0:15:56.240 than doing the literal opposite. 0:15:56.280 --> 0:15:56.960 And you have to. 0:15:56.960 --> 0:16:00.360 Wonder if when that whole non exclusisive thing came along, 0:16:00.400 --> 0:16:03.200 whether Microsoft was kind of like, no, no, you couldn't 0:16:03.200 --> 0:16:06.200 possibly like the papers already at the pen they've already 0:16:06.200 --> 0:16:08.640 got a stamp Ford sad woman's signature. No, don't sign 0:16:08.680 --> 0:16:12.720 it. It would be so bad. No, I really don't. I 0:16:12.760 --> 0:16:29.760 don't think that Microsoft's too cut up about that. And 0:16:29.840 --> 0:16:32.040 here's a question for you. If general, if AI had 0:16:32.080 --> 0:16:35.520 so much demand, why is Microsoft canceling data centrare contracts. 0:16:35.760 --> 0:16:39.120 Why is Microsoft Oracle's largest customer as at the end 0:16:39.160 --> 0:16:41.880 of twenty twenty three, allowing softmank and open Ai to 0:16:41.920 --> 0:16:44.880 work with Oracle to build the future rather than Microsoft. 0:16:46.120 --> 0:16:49.400 As mentioned previously, TD Cohen specifically noted in its report 0:16:49.400 --> 0:16:53.040 that microsoft shifting appetite for capacity was tied to open Ai, which, 0:16:53.080 --> 0:16:55.800 as I've said already, heavily suggests that Microsoft is at 0:16:55.880 --> 0:16:58.720 best less invested in the future of the company, a 0:16:58.760 --> 0:17:01.280 statement confirmed by the information which adds that Microsoft have 0:17:01.320 --> 0:17:03.800 been trying to and I quote lessen its reliance and 0:17:03.840 --> 0:17:07.119 open ai technology as they increasingly compete in selling AI 0:17:07.160 --> 0:17:12.040 products well. At worse, this situation could suggest that Microsoft 0:17:12.119 --> 0:17:14.560 is actively trying to dump open ai and it's having 0:17:14.640 --> 0:17:18.840 questions about the fundamentals of this industry. Writ large in 0:17:19.000 --> 0:17:22.520 very plain terms, Microsoft, despite its excitement around AI and 0:17:22.560 --> 0:17:25.560 its dogged insistence that it's the future, has canceled data 0:17:25.600 --> 0:17:28.040 center leases of over a gig or what of other 0:17:28.320 --> 0:17:31.600 data center infrastructure. Doing so heavily suggests that they do 0:17:31.640 --> 0:17:34.040 not intend to expand further, or at least to the 0:17:34.040 --> 0:17:37.399 extraordinary levels that they'd initially promised. I know, I'm being 0:17:37.480 --> 0:17:40.280 kind of repetitive. I know that I'm saying some of 0:17:40.320 --> 0:17:43.680 these things repeatedly, and you might think why. I need 0:17:43.720 --> 0:17:47.200 you to understand how significant this is because this did 0:17:47.240 --> 0:17:50.440 not get covered enough. The coverage of this was dogshit. 0:17:51.160 --> 0:17:54.880 I'm saying it, frankly, everyone missed this detail. I am 0:17:54.960 --> 0:17:58.040 just one guy. I do a podcast and a newsletter 0:17:58.080 --> 0:18:00.119 and I run a PR firm. Why am I the 0:18:00.280 --> 0:18:04.080 person every not every time? But it just drives me 0:18:04.119 --> 0:18:07.600 a little insane because this was seeing that how many 0:18:07.960 --> 0:18:11.359 reporters actually read this report too. This is what drives 0:18:11.400 --> 0:18:13.480 me insane with my work. But I really do enjoy 0:18:13.480 --> 0:18:16.879 it was all right. While Microsoft has reiterated that it 0:18:16.920 --> 0:18:20.040 intends to spend a ridiculous eighty billion dollars in capital 0:18:20.080 --> 0:18:23.680 expenditures on AI in twenty twenty five, per CNBC article, 0:18:23.840 --> 0:18:25.679 it's unclear how it intends to do so if it's 0:18:25.720 --> 0:18:28.359 pulling back on data center expansion at such a large scale, 0:18:28.480 --> 0:18:31.359 and the company has provided no tangible explanation or elaboration 0:18:31.400 --> 0:18:34.080 as to how it might do so. While hardware upgrades 0:18:34.080 --> 0:18:36.439 could account for some of those CAPEX, it would be 0:18:36.520 --> 0:18:40.320 nowhere near the eighty billion dollar figure. Again, hyperscale data 0:18:40.320 --> 0:18:43.520 centers aren't cheap. They're massive billion dollar or multi billion 0:18:43.520 --> 0:18:47.720 dollar ventures. Microsoft, according to CNBC, also leases data center 0:18:47.720 --> 0:18:50.159 capacity through core Weave and other providers, though at that 0:18:50.200 --> 0:18:52.160 point the reporter has stopped being curious enough to ask 0:18:52.240 --> 0:18:55.760 how much or who those other providers might be. Now, Luckily, 0:18:55.760 --> 0:18:57.960 with the power of research, I found that the information 0:18:58.000 --> 0:19:00.840 report that Microsoft plants spend about tenion dollars renting core 0:19:00.840 --> 0:19:03.400 Weave service between twenty twenty three and twenty thirty, which 0:19:03.440 --> 0:19:08.080 was also reported by CNBC and otherwise anyway planned past 0:19:08.119 --> 0:19:10.440 tense being the operative word, because last week we learned 0:19:10.440 --> 0:19:13.360 that Microsoft plans to scale back its purchase of capacity 0:19:13.359 --> 0:19:16.120 from core Weave. Core Weave is, as you'll find out 0:19:16.119 --> 0:19:19.840 in a future episode and newsletter probably before that. They're 0:19:19.840 --> 0:19:22.159 a very odd company which is closely tied to another 0:19:22.200 --> 0:19:25.560 company called core Scientific, which it actually rents service from, 0:19:25.960 --> 0:19:29.480 and that company exited Chapter eleven bankruptcy only last year. 0:19:30.240 --> 0:19:32.800 Core Scientific's financial statements, by the way, are very confusing. 0:19:32.840 --> 0:19:34.840 They're a mess, and it mostly makes its money not 0:19:34.880 --> 0:19:37.840 from selling high performance computer services, but from mining bitcoin. 0:19:38.320 --> 0:19:40.040 It's a weird relationship, and it's weird to still that 0:19:40.119 --> 0:19:43.159 Microsoft is even entangled with a company connected to Core Scientific. 0:19:43.400 --> 0:19:47.560 But again, that's a future episode, future newsletter, future panic 0:19:47.600 --> 0:19:49.600 attack that I'll give myself as I read s once 0:19:49.640 --> 0:19:52.720 all day. Moving on, Microsoft also added in a comment 0:19:52.760 --> 0:19:54.960 to CNBC in the same article that it continues to 0:19:54.960 --> 0:19:59.119 grow at a record pace to meet customer demand. Okay, 0:19:59.240 --> 0:20:03.919 excuse me. What customer demand? What customer demand? What is it? 0:20:03.960 --> 0:20:07.000 What is the customer demand? What is going on? Microsoft 0:20:07.000 --> 0:20:09.560 said back in April twenty twenty four that AI demand 0:20:09.640 --> 0:20:13.000 was exceeding supply even after a seventy nine percent surgeon 0:20:13.040 --> 0:20:16.919 capital expenditures, and Cfoamyhood send in their next quarterly earnings 0:20:16.920 --> 0:20:19.800 in July twenty twenty four that demand remained higher than 0:20:19.840 --> 0:20:23.680 microsoft available capacity. On its most recent January twenty twenty 0:20:23.680 --> 0:20:27.080 five earning score, Cfoamyhood once again said, as your growth 0:20:27.119 --> 0:20:30.760 included thirteen points from AI services, which grew one hundred 0:20:30.760 --> 0:20:33.159 and fifty seven percent year of a year and was 0:20:33.200 --> 0:20:36.480 ahead of expectations even as demand continued to be higher 0:20:36.520 --> 0:20:39.960 than our available capacity. Riddle me this, batman, Why does 0:20:40.000 --> 0:20:42.760 a company that keeps talking about having demand that exceeds 0:20:42.800 --> 0:20:46.320 capacity decide to cancel multiple data centers which collectively account 0:20:46.400 --> 0:20:50.320 for a significant chunk of its existing capacity. I don't know. 0:20:50.400 --> 0:20:52.320 Let's see what Microsoft had to say when asked a 0:20:52.320 --> 0:20:55.760 week or two ago. Ahem. Thanks to the significant investments 0:20:55.800 --> 0:20:57.600 we've made up until this point, we are well positioned 0:20:57.600 --> 0:21:01.520 to beat our current and increasing customer demand. Last year alone, 0:21:01.560 --> 0:21:03.800