WEBVTT - Monologue: Into The Jensenverse

0:00:03.170 --> 0:00:08.100
<v Speaker 1>Hello and welcome to your latest Better Offline monologue, I'm

0:00:08.130 --> 0:00:18.880
<v Speaker 1>your host Ed Zitron. The Groundhog once again saw its

0:00:18.940 --> 0:00:22.360
<v Speaker 1>shadow with Nvidia reporting record earnings, continuing to inflate an

0:00:22.500 --> 0:00:26.160
<v Speaker 1>ever more dangerous AI bubble with $ 96 billion worth of

0:00:26.220 --> 0:00:30.120
<v Speaker 1>revenue and 16% of that coming from a single unnamed customer.

0:00:30.910 --> 0:00:33.790
<v Speaker 1>For the first half of the fiscal year 2027, 44%

0:00:33.790 --> 0:00:38.269
<v Speaker 1>of Nvidia's $ 177. 8 billion in revenue came from a mere

0:00:38.390 --> 0:00:41.150
<v Speaker 1>three customers. And as a note, this was all part

0:00:41.190 --> 0:00:43.830
<v Speaker 1>of its second quarter fiscal year 2027 earnings. You're going

0:00:43.850 --> 0:00:47.050
<v Speaker 1>to be confused, but Nvidia's fiscal years run from February

0:00:47.300 --> 0:00:49.519
<v Speaker 1>1st of the year before to January 31st of the

0:00:49.550 --> 0:00:52.900
<v Speaker 1>year of. So it's currently fiscal year 2027, which began

0:00:52.940 --> 0:00:56.210
<v Speaker 1>on February 1st, 2026. Annoying, but I didn't do it.

0:00:56.280 --> 0:00:58.560
<v Speaker 1>Don't blame me. Don't be mad at me, Farah. Nvidia

0:00:58.600 --> 0:01:02.560
<v Speaker 1>does business. Anyway, to make matters worse, 70% of NVIDIA's

0:01:02.600 --> 0:01:04.960
<v Speaker 1>current accounts receivable, so GID that's been shipped to a

0:01:05.000 --> 0:01:08.160
<v Speaker 1>customer that's yet to be paid for, is from five customers,

0:01:08.300 --> 0:01:12.610
<v Speaker 1>something that NVIDIA reported while adding that for certain investment-grade customers,

0:01:12.980 --> 0:01:15.410
<v Speaker 1>it was offering payment terms ranging from 90 days to

0:01:15.450 --> 0:01:20.330
<v Speaker 1>an entire year. If you're wondering what investment-grade could refer to,

0:01:20.630 --> 0:01:23.830
<v Speaker 1>both Corweave and Nebius, two unprofitable Neo clouds that exist

0:01:23.870 --> 0:01:26.510
<v Speaker 1>only to raise debt and buy GPUs, mostly to rent

0:01:26.530 --> 0:01:28.679
<v Speaker 1>them to the same three or four companies, have had

0:01:28.700 --> 0:01:32.460
<v Speaker 1>their debt rate as investment grade by ratings agencies because

0:01:32.480 --> 0:01:36.510
<v Speaker 1>they're collateralized by contracts by companies with good credit. What

0:01:36.550 --> 0:01:39.450
<v Speaker 1>is it with ratings agencies? It's just like every time

0:01:39.470 --> 0:01:42.430
<v Speaker 1>you put your hands in front of your face and

0:01:42.450 --> 0:01:45.150
<v Speaker 1>they'll be like, where'd they go? Anyway, all of this

0:01:45.209 --> 0:01:47.090
<v Speaker 1>is a very complex way of saying that the vast

0:01:47.150 --> 0:01:50.230
<v Speaker 1>majority of NVIDIA's revenue comes from a handful of companies,

0:01:50.390 --> 0:01:53.500
<v Speaker 1>each one including hyperscalers that make up more than 50%

0:01:53.500 --> 0:01:55.760
<v Speaker 1>of its revenue over the last few years. are now

0:01:55.820 --> 0:02:00.100
<v Speaker 1>reliant on debt thanks to skyrocketing costs of Nvidia's GPUs.

0:02:00.840 --> 0:02:04.640
<v Speaker 1>While Nvidia is unquestionably profitable and thriving revenue-wise, it's only

0:02:04.720 --> 0:02:06.880
<v Speaker 1>doing so because of a few companies who are willing

0:02:06.960 --> 0:02:10.150
<v Speaker 1>to be fin-dommed by Jensen Huang, and it's all beginning

0:02:10.190 --> 0:02:12.410
<v Speaker 1>to look like the AI bubble only continues to inflate

0:02:12.450 --> 0:02:14.410
<v Speaker 1>as long as Nvidia can keep pumping it full of

0:02:14.470 --> 0:02:17.450
<v Speaker 1>money and hype. You see, in the run-up to earnings,

0:02:17.530 --> 0:02:21.389
<v Speaker 1>Nvidia sunk $ 6 billion into Poolside, a flailing open-source LLM

0:02:21.430 --> 0:02:23.660
<v Speaker 1>company that failed to raise billions of dollars earlier in

0:02:23.680 --> 0:02:26.520
<v Speaker 1>the year, Hiring away most of its staff in what

0:02:26.560 --> 0:02:29.550
<v Speaker 1>its founders said was not an acqui-hire. Other than the

0:02:29.590 --> 0:02:32.650
<v Speaker 1>fact that everyone's leaving, and other than the founders. Anyway,

0:02:32.770 --> 0:02:35.530
<v Speaker 1>NVIDIA is also rumored to be sinking billions into pathetic

0:02:35.610 --> 0:02:39.150
<v Speaker 1>AI search company Perplexity, a $ 30 billion valuation, which is

0:02:39.210 --> 0:02:42.300
<v Speaker 1>insane and it's not worth that. Likely because Perplexity is

0:02:42.360 --> 0:02:44.040
<v Speaker 1>one of the few companies in the world that buys

0:02:44.260 --> 0:02:47.340
<v Speaker 1>any significant amount of AI compute. Likely a few hundred

0:02:47.360 --> 0:02:49.980
<v Speaker 1>million dollars a year, but I'm just guessing. It's also

0:02:50.040 --> 0:02:54.639
<v Speaker 1>investing in Stargate, Abilene, Lansium at a multi-billion dollar valuation,

0:02:54.700 --> 0:02:57.079
<v Speaker 1>as well as considering putting money into AI data center

0:02:57.100 --> 0:03:01.010
<v Speaker 1>power company Cloverleaf, and backstopping over $ 100 billion of a

0:03:01.050 --> 0:03:04.929
<v Speaker 1>SoftBank-backed data center for OpenAI to rent out in Ohio,

0:03:05.030 --> 0:03:07.590
<v Speaker 1>though that backstop only kicks in if the thing actually

0:03:07.630 --> 0:03:11.880
<v Speaker 1>gets built. Shit, shit, sorry, forgot one. NVIDIA is also

0:03:11.980 --> 0:03:15.160
<v Speaker 1>buying AI model hosting platform Hugging Face for a little

0:03:15.200 --> 0:03:19.669
<v Speaker 1>under $ 13 billion. For some context, Hugging Face had $ 150

0:03:19.310 --> 0:03:21.690
<v Speaker 1>million in annualized run rate, which works out to a

0:03:21.750 --> 0:03:26.990
<v Speaker 1>pathetic $ 12. 5 million a month, which is insanely small considering

0:03:27.030 --> 0:03:30.260
<v Speaker 1>its notoriety. It's almost as if every AI company just

0:03:30.300 --> 0:03:40.020
<v Speaker 1>kind of stinks. As a reminder, NVIDIA has invested in Anthropic, OpenAI, CoreWeave, Nebius, Iren, Nscale, Intel, SpaceX,

0:03:40.060 --> 0:03:43.190
<v Speaker 1>and both Wrecker and Weka, which are somehow two different companies.

0:03:43.730 --> 0:03:47.510
<v Speaker 1>It has backstopped Corweave's leases, signed over $ 30 billion in

0:03:47.570 --> 0:03:51.050
<v Speaker 1>multi-year agreements to rent back its GPUs, and created bizarre

0:03:51.110 --> 0:03:54.190
<v Speaker 1>revenue share deals where it agrees to backstop data centers

0:03:54.460 --> 0:03:57.540
<v Speaker 1>and receive a cut of the revenue above a certain threshold,

0:03:57.800 --> 0:04:00.680
<v Speaker 1>which is obviously dependent on the bloody thing being built

0:04:00.740 --> 0:04:05.100
<v Speaker 1>and also being paid for. And on its latest earnings call,

0:04:05.460 --> 0:04:08.380
<v Speaker 1>CFO Colette Kress said that the company did not see

0:04:08.420 --> 0:04:11.540
<v Speaker 1>this as circular financing, despite there being so many different

0:04:11.600 --> 0:04:13.900
<v Speaker 1>deals where the money moves in a perfect circle, the

0:04:13.940 --> 0:04:17.560
<v Speaker 1>mainer James Keenan is mumbling some shit about Judith and Magdalena.

0:04:18.180 --> 0:04:21.260
<v Speaker 1>The AI bubble is becoming increasingly about whether Jensen Huang

0:04:21.300 --> 0:04:23.620
<v Speaker 1>and NVIDIA can keep slapping duct tape on the sides

0:04:23.640 --> 0:04:26.320
<v Speaker 1>of data center and AI model companies, and in turn

0:04:26.370 --> 0:04:29.150
<v Speaker 1>whether hyperscalers can keep raising enough debt to keep paying

0:04:29.190 --> 0:04:32.290
<v Speaker 1>for their GPUs and more. To quote Lindsay Tyler and

0:04:32.330 --> 0:04:35.789
<v Speaker 1>Nishant Sadyam of Morgan Stanley, NVIDIA is now a balance

0:04:35.810 --> 0:04:40.140
<v Speaker 1>sheet as a service company with over... $ 366 billion in commitments,

0:04:40.200 --> 0:04:43.480
<v Speaker 1>including $ 25 billion of data center leases yet to commence,

0:04:43.520 --> 0:04:46.300
<v Speaker 1>which is strange considering NVIDIA is the company that sells

0:04:46.339 --> 0:04:49.880
<v Speaker 1>the data centers, the GPUs. Why does NVIDIA need to

0:04:50.140 --> 0:04:52.340
<v Speaker 1>rent back the GPUs? I thought the demand for AI

0:04:52.380 --> 0:04:55.680
<v Speaker 1>was so high, Jensen. And if you weren't already a

0:04:55.720 --> 0:04:59.400
<v Speaker 1>little bit concerned, Huang also claimed on an analyst call

0:04:59.420 --> 0:05:02.580
<v Speaker 1>that the world has already achieved artificial general intelligence, which

0:05:02.620 --> 0:05:05.950
<v Speaker 1>he defines as the ultimate form of the technology where

0:05:05.970 --> 0:05:08.560
<v Speaker 1>the machines can think and act for themselves. This has,

0:05:08.600 --> 0:05:10.820
<v Speaker 1>of course, not actually happened, will not actually happen, and

0:05:11.100 --> 0:05:13.440
<v Speaker 1>should genuinely get it lambasted in the press, but because

0:05:13.520 --> 0:05:18.890
<v Speaker 1>number has gone up, everybody is happy. Yay! Yay! Good job, Jensen!

0:05:19.170 --> 0:05:24.020
<v Speaker 1>AGI's here! Yay! Fucking kill me. Yet somehow the dumbest

0:05:24.060 --> 0:05:26.240
<v Speaker 1>thing that was said on the earnings call came from

0:05:26.320 --> 0:05:29.140
<v Speaker 1>Nvidia's CFO, who said that Nvidia will grow its revenue

0:05:29.190 --> 0:05:34.109
<v Speaker 1>by 70% in fiscal year 2028, which begins February 1st, 2027.

0:05:34.110 --> 0:05:36.810
<v Speaker 1>I know it's annoying. And based on consensus estimates for

0:05:36.830 --> 0:05:40.430
<v Speaker 1>the current fiscal year being at $ 396 billion, this means

0:05:40.470 --> 0:05:42.609
<v Speaker 1>that Nvidia is expecting to make over $ 674 billion next year.

0:05:45.290 --> 0:05:49.680
<v Speaker 1>It's not impossible, but to do this, Nvidia will need

0:05:49.710 --> 0:05:53.960
<v Speaker 1>its current customers, predominantly Amazon, Google, Microsoft, Meta, SpaceX, and Oracle,

0:05:54.220 --> 0:05:56.660
<v Speaker 1>to likely double their current orders for GPUs, as well

0:05:56.700 --> 0:05:59.520
<v Speaker 1>as neo-clouds like CoreWeave to raise even more debt than

0:05:59.560 --> 0:06:02.659
<v Speaker 1>they're currently raising. And to be abundantly clear, CoreWeave is

0:06:02.680 --> 0:06:04.910
<v Speaker 1>already having to pay over 9% interest on its debt

0:06:05.150 --> 0:06:08.790
<v Speaker 1>just to get investors through the door. Again, not impossible,

0:06:09.350 --> 0:06:12.720
<v Speaker 1>but it's going to require... So much more debt. I mean,

0:06:12.740 --> 0:06:14.710
<v Speaker 1>this must be over a trillion at this point in

0:06:14.730 --> 0:06:18.770
<v Speaker 1>the next year. It's completely fucking insane. And much like

0:06:19.110 --> 0:06:22.469
<v Speaker 1>OpenAI and Anthropic must continue growing their revenues, customer base

0:06:22.490 --> 0:06:26.680
<v Speaker 1>and funding rounds to meet over $ 1. 1 trillion in compute agreements,

0:06:27.040 --> 0:06:30.080
<v Speaker 1>NVIDIA must, through science or magic, find ways for its

0:06:30.140 --> 0:06:31.940
<v Speaker 1>customers to be able to buy more and more and

0:06:31.980 --> 0:06:35.760
<v Speaker 1>more GPUs every single quarter from here into eternity. As

0:06:35.800 --> 0:06:37.840
<v Speaker 1>I've said again and again, the AI boom is only

0:06:37.880 --> 0:06:40.810
<v Speaker 1>possible as long as debt can sustain it. SoftBank has

0:06:40.850 --> 0:06:42.919
<v Speaker 1>had to take out tens of billions of dollars of

0:06:42.980 --> 0:06:45.020
<v Speaker 1>debt to fund OpenAI and now is trying to sell

0:06:45.020 --> 0:06:49.420
<v Speaker 1>$ 20 billion of bonds to refinance that debt. CoreWeave can

0:06:49.480 --> 0:06:51.760
<v Speaker 1>only keep building data centers as long as the banks

0:06:51.800 --> 0:06:55.180
<v Speaker 1>give it money. Oracle, Google, Amazon, Meta, and SpaceX cannot

0:06:55.240 --> 0:06:57.800
<v Speaker 1>afford to buy GPUs out of cash flow and thus

0:06:57.839 --> 0:07:01.080
<v Speaker 1>are dependent on debt to continue buying those GPUs. Even

0:07:01.140 --> 0:07:03.670
<v Speaker 1>NVIDIA itself has had to take out over $ 25 billion

0:07:03.710 --> 0:07:07.030
<v Speaker 1>in bonds. Maybe it's $ 20 billion. Regardless, still a lot

0:07:07.070 --> 0:07:11.920
<v Speaker 1>of money. And that supposed $ 500 billion fund, I'm doing

0:07:11.980 --> 0:07:13.640
<v Speaker 1>air quotes, you can't see it because this is an

0:07:13.700 --> 0:07:16.840
<v Speaker 1>audio podcast, per Bloomberg, is nothing of the sort. I'll

0:07:16.860 --> 0:07:18.800
<v Speaker 1>link to the article in the notes, but that announcement

0:07:18.840 --> 0:07:20.880
<v Speaker 1>was literally just NVIDIA saying that a group of asset

0:07:20.900 --> 0:07:23.840
<v Speaker 1>managers would invest half a trillion dollars in AI data centers,

0:07:23.860 --> 0:07:26.960
<v Speaker 1>and that number, and I quote, had no obvious provenance.

0:07:28.300 --> 0:07:31.960
<v Speaker 1>Hyperscalers are only spending this money because Anthropic and OpenAI

0:07:32.230 --> 0:07:35.910
<v Speaker 1>have committed to spend over $ 400 billion renting GPUs in

0:07:35.930 --> 0:07:37.830
<v Speaker 1>the next three and a half years, according to analysts.

0:07:38.110 --> 0:07:41.110
<v Speaker 1>And without that spend, 30% or more of their cloud

0:07:41.150 --> 0:07:45.740
<v Speaker 1>revenues will evaporate. For example, per UBS, Anthropic and OpenAI

0:07:45.780 --> 0:07:50.520
<v Speaker 1>will make up over 48% of Google Cloud's revenues in 2027,

0:07:50.520 --> 0:07:52.760
<v Speaker 1>which will only be possible if they can both afford

0:07:52.780 --> 0:07:54.980
<v Speaker 1>to spend the money and the data center capacity is

0:07:55.020 --> 0:07:58.510
<v Speaker 1>actually available. And data centers are taking forever. Another problem.

0:08:00.170 --> 0:08:03.570
<v Speaker 1>NVIDIA is investing in every schmuck with a penchant, that's

0:08:03.590 --> 0:08:06.310
<v Speaker 1>definitely how you say it, for AI models because any

0:08:06.390 --> 0:08:09.570
<v Speaker 1>AI company becoming insolvent or doing a down round would

0:08:09.590 --> 0:08:12.390
<v Speaker 1>break the illusion that AI is the next industrial revolution

0:08:12.470 --> 0:08:14.890
<v Speaker 1>rather than a series of different companies handing money to

0:08:14.930 --> 0:08:17.490
<v Speaker 1>each other in the hopes that a business model arrives.

0:08:18.290 --> 0:08:20.380
<v Speaker 1>The reason nobody is freaking out is that nothing bad

0:08:20.390 --> 0:08:23.180
<v Speaker 1>has happened yet. NVIDIA's big customers can still raise money

0:08:23.200 --> 0:08:26.250
<v Speaker 1>to pay it. and its horrible Neo-Cloud progeny can continue

0:08:26.270 --> 0:08:28.610
<v Speaker 1>to loot the debt markets from gullible investors who have

0:08:28.630 --> 0:08:30.880
<v Speaker 1>been lied to by analysts and the media about AI's

0:08:30.940 --> 0:08:35.000
<v Speaker 1>promise and stability. Everything comes down to whether near-infinite resources

0:08:35.040 --> 0:08:38.400
<v Speaker 1>are available for AI in perpetuity, and when those resources

0:08:38.460 --> 0:08:41.600
<v Speaker 1>slow or stop, so too will the music. I'll be

0:08:41.620 --> 0:08:43.560
<v Speaker 1>back next week with my boys Caleb and Arif to

0:08:43.600 --> 0:08:46.520
<v Speaker 1>talk about horrible AI slop and whatever else crosses our

0:08:46.580 --> 0:08:49.400
<v Speaker 1>wretched little minds. Thank you for listening. I love you all.