WEBVTT

NOTE Sentence-level transcript of https://www.youtube.com/watch?v=Wk2bUUGFC_o

NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/Wk2bUUGFC_o.html. A cue ends where the next begins, or 2 s after its last word.

s1
00:00:01.309 --> 00:00:03.309
[music]

s2
00:00:12.719 --> 00:00:15.280
So, this is a talk about compute markets.

s3
00:00:15.280 --> 00:00:19.920
Um, but it's kind of more of a talk about why to make a compute market.

s4
00:00:19.920 --> 00:00:22.240
Um, if you don't know me, I'm Evan.

s5
00:00:22.240 --> 00:00:24.880
Um, we run a company called San Francisco Compute.

s6
00:00:24.880 --> 00:00:29.359
Um, we make a physically settled order, physically settled forwards market.

s7
00:00:29.359 --> 00:00:37.600
Um, so if you were here for the last talk, last talk was basically why all the operators of GPU clusters um don't really work very well.

s8
00:00:37.600 --> 00:00:42.719
Um, when we say physically settled, we mean we're the operator.

s9
00:00:42.719 --> 00:00:49.840
We're both the operator of the hardware in the same way that a cloud is the operator of the hardware and then we're also the market at the same time.

s10
00:00:49.840 --> 00:00:51.840
It's kind of like a grid for compute.

s11
00:00:51.840 --> 00:01:05.119
So the general gist of this is that in order to run and operate a supercomputer um we think problems start right at the design of the supercomputer um right in the design of the bill of materials

s12
00:01:05.119 --> 00:01:18.240
um it's control of the backplane fabric as well it's also control of um bare metal orchestration it's also services on top VMs and so on um there's a lot of different parts of the stack that you have to do effectively

s13
00:01:18.240 --> 00:01:30.000
otherwise um the folks who have a data data center um and would like to make more money on that data center by becoming the GPU cloud um end up

s14
00:01:30.000 --> 00:01:31.680
failing for the customer.

s15
00:01:31.680 --> 00:01:33.360
So that's what we do.

s16
00:01:33.360 --> 00:01:35.840
We offer what's called supercomputing as a service.

s17
00:01:35.840 --> 00:01:37.840
We go to folks who want to become a cloud.

s18
00:01:37.840 --> 00:01:50.799
That's like folks who want to become what Nvidia would call an NCP um or venture firms that want to have clusters for their portfolio um and want to be able to reallocate that capacity when a company like Instinct is taking off.

s19
00:01:50.799 --> 00:01:59.439
um or AI companies who would like the residual value on the compute because they've realized that's really important to their financials um but don't want to operate a supercomputer.

s20
00:01:59.439 --> 00:02:08.239
So that's physical settlement and then on top of that we run a live order book for capacity that you can buy reserved either right now or in the future.

s21
00:02:08.239 --> 00:02:14.959
Um this um together is not at all what we thought we were going to be doing.

s22
00:02:14.959 --> 00:02:22.319
Originally, we were this company called Junlark um that was trying to make kind of like Sunno or UDIO, but prior to Sunno or UDIO.

s23
00:02:22.319 --> 00:02:33.120
Um at the time, um what you needed to do to realistically start a lab was you needed to raise a big deposit, a huge one, which was $10 million at the time.

s24
00:02:33.120 --> 00:02:35.440
Um we thought that was a crazy amount of money.

s25
00:02:35.440 --> 00:02:36.959
Um and so we thought, no, that's stupid.

s26
00:02:36.959 --> 00:02:40.959
We're going to raise 500k and obviously what we're going to do is we're going to rent from somebody else.

s27
00:02:40.959 --> 00:02:46.480
Um obviously today um it's not $10 million, it's a billion dollars.

s28
00:02:46.480 --> 00:02:48.336
And um

s29
00:02:48.336 --> 00:02:48.480
[snorts]

s30
00:02:48.480 --> 00:02:51.519
nonetheless, um we had raised our 500k.

s31
00:02:51.519 --> 00:02:55.360
We buy a big cluster and then we're underwater on our cluster every month.

s32
00:02:55.360 --> 00:02:58.640
Um and if we didn't subleasase our cluster, we would have gone bankrupt.

s33
00:02:58.640 --> 00:03:04.239
Um so if you've heard of GPU brokers, we were the first GPU broker because we were forced into being a GPU broker.

s34
00:03:04.239 --> 00:03:08.239
So the first year of the company was us just trying to sell our cluster, otherwise we would go bankrupt.

s35
00:03:08.239 --> 00:03:21.360
Um, so I would wander around with these like actual um pieces of paper at Nurups which both say H100 and A100 at the same time because that's how I made uh I made the uh poster so quickly.

s36
00:03:21.360 --> 00:03:28.480
Um, and so this talk is not really about um necessarily like how a compute market works.

s37
00:03:28.480 --> 00:03:32.319
It's how a compute market um gets to exist.

s38
00:03:32.319 --> 00:03:38.959
Uh we like to think about it like there's this pamphlet and on the pamphlet on the front side it's like man starting a Neolab sucks.

s39
00:03:38.959 --> 00:03:45.120
There's all these different like financial and economic problems and on the back of the p back of the pamphlet it says compute markets.

s40
00:03:45.120 --> 00:03:53.920
It's it's not super clear how one gets to the other until you open the pamphlet and then like oh my holy there's like so much stuff in this pamphlet and it keeps going and there's so many folds.

s41
00:03:53.920 --> 00:03:59.920
It's just a very difficult and challenging problem because you need to solve two like miracle problems to get working.

s42
00:03:59.920 --> 00:04:05.120
You have to both make a market which is uh a uniquely difficult problem because it has a cold start problem.

s43
00:04:05.120 --> 00:04:08.080
You also have to build a cloud and then you have to not become a cloud.

s44
00:04:08.080 --> 00:04:10.959
Um which is a whole separate challenge.

s45
00:04:10.959 --> 00:04:16.160
But these are the problems that the labs are facing and it's about to be the problems that everyone is about to facing.

s46
00:04:16.160 --> 00:04:18.560
So if you work at a startup these are your problems now.

s47
00:04:18.560 --> 00:04:21.280
Um right now everybody is sold out of compute.

s48
00:04:21.280 --> 00:04:26.240
Um this is um probably not going to change for a while.

s49
00:04:26.240 --> 00:04:30.720
Um, so that means that you're going to have to buy a new cluster, a net new one, and there's a lead time.

s50
00:04:30.720 --> 00:04:33.759
You have no way to pay for that cluster to come online now.

s51
00:04:33.759 --> 00:04:35.440
You're going to pay for it to come online later.

s52
00:04:35.440 --> 00:04:39.440
There's nothing in the market that can realistically get that cluster on online for you now.

s53
00:04:39.440 --> 00:04:43.600
Um, meanwhile, you're going to pay your entire bank balance as a deposit.

s54
00:04:43.600 --> 00:04:50.320
Um, the reason you're paying your entire bank balance as a deposit is because the lender has no idea how to render under underwrite AGI.

s55
00:04:50.320 --> 00:04:52.479
Um, how in the world are you supposed to do that?

s56
00:04:52.479 --> 00:05:00.080
You've got two person startups in San Francisco who or maybe New York um who say I'm going to build super intelligence.

s57
00:05:00.080 --> 00:05:03.120
No lender knows how to underwrite super intelligence.

s58
00:05:03.120 --> 00:05:05.840
Um so they have no idea what the asset is worth.

s59
00:05:05.840 --> 00:05:09.360
Um meanwhile your cluster is very likely to be delayed.

s60
00:05:09.360 --> 00:05:16.720
That's because the lender needed to underwrite against the credit risk of the customer and so they needed to know who is the customer first before you deploy the cluster.

s61
00:05:16.720 --> 00:05:21.039
So right when the customer signs is the riskiest point in the entire operation.

s62
00:05:21.039 --> 00:05:25.360
um you wanted it to be the case that you built the cluster first and then you sell it.

s63
00:05:25.360 --> 00:05:29.759
This also helps reduce all sorts of different types of problems um where clusters break.

s64
00:05:29.759 --> 00:05:35.680
Meanwhile, most of the people who are setting up clusters are people who have the actual asset that makes the thing cheap.

s65
00:05:35.680 --> 00:05:38.560
Like you've got the data center space already.

s66
00:05:38.560 --> 00:05:41.680
What that doesn't get is supercomputing expertise.

s67
00:05:41.680 --> 00:05:45.120
um as we saw in the last talk if you've um watched that one.

s68
00:05:45.120 --> 00:05:52.720
Um meanwhile, you're going to run into the exact same problem that Junark did, which is you're going to either way overby or you're going to weigh underby.

s69
00:05:52.720 --> 00:05:56.320
Um that's because AI doesn't have margins that look like SAS companies.

s70
00:05:56.320 --> 00:05:58.560
And so you don't have very much margin for error.

s71
00:05:58.560 --> 00:06:07.120
And so the chances that you either underby and somebody else wins because they did it correctly or you over buy and then you just become stability AI or you blow up.

s72
00:06:07.120 --> 00:06:11.280
you um end up like character AI um are really high.

s73
00:06:11.280 --> 00:06:12.560
So these are all the same problem.

s74
00:06:12.560 --> 00:06:14.160
They're all just different types of risk.

s75
00:06:14.160 --> 00:06:16.720
They're technical risk, their financial risk, their credit risk.

s76
00:06:16.720 --> 00:06:19.039
Um they're all one big bundled risk problem.

s77
00:06:19.039 --> 00:06:24.960
Uh which is the problem that we end up solving and it's also the problem that we think compute markets broadly the entire industry solves.

s78
00:06:24.960 --> 00:06:32.960
Um so we'll walk through each one of these and walk how that gets to a compute market because if you don't understand these, it's kind of like unclear why you would want a compute market.

s79
00:06:32.960 --> 00:06:36.319
Um, and it almost seem like this problem that somebody thinks, "Oh, wow.

s80
00:06:36.319 --> 00:06:38.560
Compute markets are cool, so I'm going to go start one."

s81
00:06:38.560 --> 00:06:44.160
We think they should be derived from the actual problems that people face, which is first, everyone is sold out.

s82
00:06:44.160 --> 00:06:45.199
Why are they sold out?

s83
00:06:45.199 --> 00:06:49.520
Because the lender couldn't finance the cluster unless you had a long-term offtake agreement.

s84
00:06:49.520 --> 00:06:54.960
In CPUs, uh, this wasn't the case because CPU buyers are SAS companies.

s85
00:06:54.960 --> 00:06:56.560
SAS companies have high margins.

s86
00:06:56.560 --> 00:07:08.400
And because they have high margins, um you can build a product that rewards you for having um a great um that rewards you with high margins um for your um your CPUs.

s87
00:07:08.400 --> 00:07:11.039
Um think about it like this way.

s88
00:07:11.039 --> 00:07:15.759
If you're a SAS company and your underlying cloud doubles their prices, you don't go bankrupt.

s89
00:07:15.759 --> 00:07:18.720
You hate your cloud, you're mad, but you don't go bankrupt.

s90
00:07:18.720 --> 00:07:24.720
If you're an AI company and your underlying cloud doubles their prices, you go bankrupt unless you go to some other underlying cloud.

s91
00:07:24.720 --> 00:07:27.440
And what that means is that you're actually price sensitive.

s92
00:07:27.440 --> 00:07:32.400
And because you're actually price sensitive, um, and there's nothing keeping you at any particular cloud.

s93
00:07:32.400 --> 00:07:37.840
So you can't sell on demand because your customers will immediately leave to whoever is cheaper.

s94
00:07:37.840 --> 00:07:47.120
And so um, that means that in order to finance GPU clusters, you have to have long-term take-pay contracts, hopefully with someone who has investment grade credit.

s95
00:07:47.120 --> 00:07:58.879
However, if you don't have liquidity on those, if you're selling reserved and your customer can't get out, then that also means that some other customer who wants the capacity right now because they're like instinct and their product is taking off can't pay for anything.

s96
00:07:58.879 --> 00:08:03.840
They have to finance a new cluster which has a lead time because the physical world just exists and it's going to take a while.

s97
00:08:03.840 --> 00:08:06.160
Sometimes you even have to build the damn data center.

s98
00:08:06.160 --> 00:08:13.840
Um, and then because you're financing a new cluster, you're going to pay a deposit because almost nobody in AI has any reasonable credit.

s99
00:08:13.840 --> 00:08:17.199
And because this is a brand new technology, all these companies are new.

s100
00:08:17.199 --> 00:08:23.039
Meanwhile, even as you um start to scale up that ladder, you're still going to run into credit issues all the way through.

s101
00:08:23.039 --> 00:08:25.039
It's the number one bottleneck of AI at the moment.

s102
00:08:25.039 --> 00:08:25.680
It's not power.

s103
00:08:25.680 --> 00:08:31.120
You hit credit way before you hit AI or way before you hit um um uh power.

s104
00:08:31.120 --> 00:08:39.200
So, um what that means is every inference company, um uh every Neol, they're bottlenecked by how much cash do they have in the bank account.

s105
00:08:39.200 --> 00:08:47.920
And so they go back to the venture capitalist who says, "Oh, I'm going to fund at the model layer, the application layer, the inference layer because I don't want to fund all of that capex stuff."

s106
00:08:47.920 --> 00:09:04.720
Um, and so the venture capitalist then funds all the capex stuff by giving the um the new neolab or new inference company um the check that they then send to the cloud um which is the deposit for the cluster and that's how you underwrite the customer.

s107
00:09:04.720 --> 00:09:15.200
So um the clusters then are delayed because um the c the cloud couldn't actually deploy the cluster until they had the customer in hand.

s108
00:09:15.200 --> 00:09:17.200
And so that was the riskiest point.

s109
00:09:17.200 --> 00:09:27.839
So therefore um what should have happened was the cloud had bought the cluster ahead of time and then waited until it was landed in the data center and then sold it.

s110
00:09:27.839 --> 00:09:31.040
And that means you have way less likelihood of delays.

s111
00:09:31.040 --> 00:09:37.360
Meanwhile, the cluster is very likely to break when you get the cluster as the customer because you don't have liquidity.

s112
00:09:37.360 --> 00:09:50.640
And so, if the failure rate is higher than what you expect and you don't have the ability to transition a node in, then you don't have anywhere you can pull nodes from, which means that if you need new nodes to make up for the failed ones, you have to RMA them, which means you have to get a

s113
00:09:50.640 --> 00:09:51.440
new node.

s114
00:09:51.440 --> 00:09:57.360
And then that gets us back to several steps before where you have uh the lead time problem again.

s115
00:09:57.360 --> 00:10:00.720
And so, everything breaks for everybody all the time.

s116
00:10:00.720 --> 00:10:08.800
Once you have gotten through all of that as a new AI lab, um your problems are you you you know you nobody was sold like everyone was sold out.

s117
00:10:08.800 --> 00:10:11.680
Um you then had to wait a lead time, you paid this big deposit.

s118
00:10:11.680 --> 00:10:15.440
Um the thing was delayed and then it broke and finally you got through all of that.

s119
00:10:15.440 --> 00:10:19.120
Um and the likelihood that you predict your demand correctly is very low.

s120
00:10:19.120 --> 00:10:23.440
And so you're either going to be way overprovisioned or way underprovisioned and you have no way out.

s121
00:10:23.440 --> 00:10:27.680
Um there's no way to sell your capacity to somebody else and there's no way to buy more capacity if you need it.

s122
00:10:27.680 --> 00:10:33.920
So, computer markets um to solve these problems, you need to solve a bunch of different problems all at once.

s123
00:10:33.920 --> 00:10:40.880
Um you need to underwrite the value of the GPU cluster rather than underwriting the value or underwriting the credit risk of the customer.

s124
00:10:40.880 --> 00:10:42.560
Um this is how it could work.

s125
00:10:42.560 --> 00:10:45.120
Um it's not guaranteed to work like this.

s126
00:10:45.120 --> 00:10:49.040
In other industries, what happened is everything just consolidated.

s127
00:10:49.040 --> 00:10:53.440
You got one big creditworthy customer that does the whole damn industry.

s128
00:10:53.440 --> 00:10:57.040
That's what the path of AI is currently on track to do.

s129
00:10:57.040 --> 00:11:03.279
It's instead of going to be um a liquid market where anybody can participate, it's going to look like TSMC.

s130
00:11:03.279 --> 00:11:07.440
You're going to get one buyer um or a few buyers that are the clouds.

s131
00:11:07.440 --> 00:11:10.480
You're going to get a few model labs that are the AI labs.

s132
00:11:10.480 --> 00:11:12.399
And then you won't get other startups.

s133
00:11:12.399 --> 00:11:14.880
Um this is what we think is going to happen.

s134
00:11:14.880 --> 00:11:19.360
And this is why 3 years ago when we were doing our AI lab, we had a chance.

s135
00:11:19.360 --> 00:11:27.360
We could have said, "No, we're not gonna do our new um we're not gonna like, you know, stick it out for several years just trying to keep our company alive.

s136
00:11:27.360 --> 00:11:29.760
Um we're just going to go do a new AI lab."

s137
00:11:29.760 --> 00:11:37.839
But we saw this problem and we said, "Oh if we don't do anything about this, then we will just hit this 3 years from now."

s138
00:11:37.839 --> 00:11:42.640
And here we are 3 years from now hitting this problem and where everyone's going to hit this problem.

s139
00:11:42.640 --> 00:11:53.440
So um as if you watched the last talk um it's very hard to do solve this problem unless you are operating both the market and the clusters at the same time.

s140
00:11:53.440 --> 00:11:54.720
Um so this is Dylan Page.

s141
00:11:54.720 --> 00:11:57.360
Um Dylan is a co-chair of the CNCF infrastructure group.

s142
00:11:57.360 --> 00:12:01.519
That's the organization that makes Kubernetes formerly worked at um Lambda doing infrastructure there.

s143
00:12:01.519 --> 00:12:11.600
Um and the reason and he joined San Francisco compute because um GPU compute isn't sufficiently standardized reliable for the market operator to be divorced from the physical infrastructure.

s144
00:12:11.600 --> 00:12:14.639
It doesn't work if you aren't the one who's buying the thing.

s145
00:12:14.639 --> 00:12:17.519
Sorry, it doesn't work if the person you're buying from isn't operating it.

s146
00:12:17.519 --> 00:12:26.800
To give an example of why this doesn't work is I'm sure some of you are flying in from overseas or sorry um out of town um and you have a hotel room.

s147
00:12:26.800 --> 00:12:31.360
Um wouldn't it be great if you could, you know, sell that hotel room to somebody else.

s148
00:12:31.360 --> 00:12:34.399
I go to you and I want to buy that hotel room.

s149
00:12:34.399 --> 00:12:37.360
Um and I say, "Can I get room service?

s150
00:12:37.360 --> 00:12:38.959
Can I get my room cleaned?

s151
00:12:38.959 --> 00:12:40.480
Um I've got a problem with my room.

s152
00:12:40.480 --> 00:12:41.600
who do I call?

s153
00:12:41.600 --> 00:12:46.560
Um, if you don't operate the hotel, you can't fix any of those problems.

s154
00:12:46.560 --> 00:12:48.880
That's why brokering clusters doesn't work.

s155
00:12:48.880 --> 00:12:51.279
You have to operate the cluster like the hotel.

s156
00:12:51.279 --> 00:12:53.040
You have to run the whole damn thing.

s157
00:12:53.040 --> 00:12:55.040
Otherwise, you can't actually make a market.

s158
00:12:55.040 --> 00:12:55.600
It doesn't work.

s159
00:12:55.600 --> 00:12:57.279
You can't do physical settlement.

s160
00:12:57.279 --> 00:13:00.880
Um, so why does nobody do physical settlement?

s161
00:13:00.880 --> 00:13:04.959
Um, the problem is there's a very strong incentive to never do physical settlement.

s162
00:13:04.959 --> 00:13:17.120
Um, that's because if you build out physical settlement for compute, what that literally means is you're building out the entire operating capacity of a GPU cloud and if you are building out the entire operating capacity of a GPU cloud,

s163
00:13:17.120 --> 00:13:23.200
you should just be a GPU cloud because it looks to be the case that you will make an ungodly amount of money.

s164
00:13:23.200 --> 00:13:27.040
Um, you will sign billion dollar deal after billion dollar deal.

s165
00:13:27.040 --> 00:13:29.200
And so, it.

s166
00:13:29.200 --> 00:13:31.680
Let's not make a GPU or let's not make a compute market.

s167
00:13:31.680 --> 00:13:38.639
let's instead just become a GPU cloud, which is what basically everybody does because the numbers are astronomical and it's a very reasonable thing to do.

s168
00:13:38.639 --> 00:13:48.240
Um, the reason why we ended up building physical settlement was because we knew that if we didn't start this, um, we would then not be able to start an AI lab, which was the real damn thing that we wanted to do in the first place.

s169
00:13:48.240 --> 00:13:49.680
We wanted to make a lab.

s170
00:13:49.680 --> 00:13:52.880
We didn't get to make a lab unless this exists.

s171
00:13:52.880 --> 00:13:59.199
So, when we started, it was 10 million um, as the deposit that you had to pay um, to start a new lab.

s172
00:13:59.199 --> 00:14:02.079
This is like how you get over the credit risk problem.

s173
00:14:02.079 --> 00:14:04.959
Um then the cluster sizes got bigger.

s174
00:14:04.959 --> 00:14:07.839
The sort of minimum ante into the space got bigger.

s175
00:14:07.839 --> 00:14:09.120
Then it was 100 million.

s176
00:14:09.120 --> 00:14:10.320
Now it's a billion.

s177
00:14:10.320 --> 00:14:12.160
We don't think it can continue.

s178
00:14:12.160 --> 00:14:20.560
We don't think there is a world in which you can get a $10 billion round right out the bat before you raise uh before you have revenue or traction or your product or anything.

s179
00:14:20.560 --> 00:14:28.240
Um and what that means is that if you don't solve this problem, we are currently on track to not have any startups in AI.

s180
00:14:28.240 --> 00:14:35.839
Um, now one of the things that you could think could happen um is maybe there can be new inference companies and new model labs.

s181
00:14:35.839 --> 00:14:41.040
And the way that they solve the credit problem is that instead they just ship the risk to the application layer.

s182
00:14:41.040 --> 00:14:47.360
They say look I will um I I need to get over my credit um problem.

s183
00:14:47.360 --> 00:14:53.839
So what I'm going to do is I'm going to only sign customers who are investment grade um credit, you know, Fortune 500.

s184
00:14:53.839 --> 00:14:58.480
Um, I'm going to only sign customers on five-year contracts for tokens.

s185
00:14:58.480 --> 00:15:03.839
I'm going to only sign customers for tokens that require a big deposit upfront.

s186
00:15:03.839 --> 00:15:07.760
Um, that is the end of everything.

s187
00:15:07.760 --> 00:15:18.320
Um, if you have inference layer getting bottlenecked by credit, what happens is you can't get tokens unless you have credit or you have a big deposit.

s188
00:15:18.320 --> 00:15:22.959
And the problem of that is that tokens are used by the whole damn economy.

s189
00:15:22.959 --> 00:15:26.160
There is no rule that says that startups have to exist.

s190
00:15:26.160 --> 00:15:35.920
Um if every startup is suddenly bottlenecked by credit um in order to access intelligence, what happens is existing firms just ship everything.

s191
00:15:35.920 --> 00:15:48.240
They themselves get to make the product um because they've got credit and then they make the revenue from that product and then it makes their credit better and that's a flywheel that makes consolidation happen throughout the entire economy.

s192
00:15:48.240 --> 00:15:51.360
If you don't fix this bottleneck, there aren't any startups.

s193
00:15:51.360 --> 00:16:03.360
This is why we started doing this and because at one point um uh or what we think is going to happen is you're going to end up in like 3 to 5 years you're going to look out and say I want to start a new startup

s194
00:16:03.360 --> 00:16:10.959
and to get there you need to buy so many tokens in order to compete with the existing partner or like existing players.

s195
00:16:10.959 --> 00:16:14.480
So you need to buy a hundred million dollars of tokens.

s196
00:16:14.480 --> 00:16:20.880
Um we think this is what you are actually going to have to do in order to start a new company because other people are spending hundreds of millions of dollars of tokens.

s197
00:16:20.880 --> 00:16:23.519
If you just draw the line out, this is what happens.

s198
00:16:23.519 --> 00:16:29.839
So, you need a big giant deposit because you're two people who have no idea what they're doing and so nobody trusts you.

s199
00:16:29.839 --> 00:16:33.440
So, you're going to go and raise $100 million from your venture capitalist.

s200
00:16:33.440 --> 00:16:37.759
Um, meanwhile, nobody wants to work with you because you're a twoerson startup.

s201
00:16:37.759 --> 00:16:46.079
Then, you're going to be asked to pay for um that hundred or you're going to be asked to pay for a 5-year contract on that 100 million um if you're not paying it all as a deposit.

s202
00:16:46.079 --> 00:16:47.839
And then you just don't get new model labs.

s203
00:16:47.839 --> 00:16:49.040
You don't get new startups.

s204
00:16:49.040 --> 00:16:51.519
um you get a consolidation in the industry.

s205
00:16:51.519 --> 00:17:09.360
So we at one point um uh were this little lab um and what we did is we lived in San Francisco um and we knew all the other people who were also just two little folks um who had started a startup and wanted to be able to participate in the most important technology of our time.

s206
00:17:09.360 --> 00:17:13.520
Um so I want these people to exist.

s207
00:17:13.520 --> 00:17:16.240
I want this person to exist and I want this person to exist.

s208
00:17:16.240 --> 00:17:20.000
I want this startup to exist and this startup to exist and this startup to exist.

s209
00:17:20.000 --> 00:17:24.720
This one and this one and this one and this one and this one and this one and this one and this one.

s210
00:17:24.720 --> 00:17:36.799
If you get these companies going, maybe they'll become big folks that build your cancer cures or they build something that nobody thought of um or something that wasn't just sort of AI slopped out.

s211
00:17:36.799 --> 00:17:39.280
This is kind of the problem that compute markets solve.

s212
00:17:39.280 --> 00:17:43.732
And if you don't solve it, we don't think there's going to be any startups.

s213
00:17:43.732 --> 00:17:45.732
[applause]