Arthur Hayes - FLOP Bigger Than Bitcoin - TOKEN2049 Singapore 2026 — Transcript

Full transcript

1. 0:08If you all read what I write, you listen
2. 0:10to me pontificate on podcasts, usually
3. 0:13I'm doing one of three things.
4. 0:15I'm skiing. This is a photo of me in um
5. 0:18Argentina a few weeks ago.
6. 0:21I'm playing tennis,
7. 0:23or I'm at a club somewhere.
8. 0:26And enjoying, you know, running a family
9. 0:28office. There's a very talented
10. 0:30investment professionals who do all the
11. 0:31work at Maelstrom. I just sign the
12. 0:33checks and get on TV. But, that's what
13. 0:36I've been doing.
14. 0:37But,
15. 0:39I thought it was time for a change.
16. 0:42And I felt it was time to put my hard
17. 0:44hat on
18. 0:45and get to work.
19. 0:47Make some money.
20. 0:50And what inspired me was what's going on
21. 0:53in the AI ecosystem.
22. 0:56Now, if you've read a lot of what I've
23. 0:58written, I'm pretty skeptical about a
24. 1:01lot of the business prospects for some
25. 1:03of the leading AI companies.
26. 1:06But, that doesn't mean that there's not
27. 1:07any underlying
28. 1:09bullishness there for the technology
29. 1:11itself
30. 1:12and how it's going to change humanity
31. 1:15and how most importantly
32. 1:17investors can make money in particular
33. 1:19parts of the ecosystem.
34. 1:21So, back, you know, 9 months ago, we got
35. 1:24introduced to this thing of agentic
36. 1:27commerce.
37. 1:28We had Quad come out with their agent
38. 1:32platform. We have Claus. We have the
39. 1:34price of compute getting progressively
40. 1:36cheaper since ChatGPT burst onto the
41. 1:38scene in 2022.
42. 1:41And
43. 1:42major investment banks like Goldman
44. 1:44Sachs started to wax lyrical about the
45. 1:47massive potential of what you can do
46. 1:50with an AI agent.
47. 1:52Now, you can pick your investment bank
48. 1:54and they probably will show you a chart
49. 1:56like this, massive up into the right
50. 1:58asymptotic. Agents are going to power
51. 2:01the next wave of human economy. It's the
52. 2:04reason why you should invest in a lot of
53. 2:07these AI labs. This is why we're
54. 2:09building trillions of dollars worth of
55. 2:12data centers.
56. 2:13It's all about token usage by AI agents.
57. 2:17Facebook recently launched the Muse
58. 2:20a consumer app where they'll help steal
59. 2:23your data and manage your agentic lives.
60. 2:26I have not given Zuck my data yet, but I
61. 2:28heard it's quite useful at $100 a month.
62. 2:30We're going to see more of these type of
63. 2:32applications as the price of compute
64. 2:34gets cheaper and more agents are in this
65. 2:37universe.
66. 2:39But,
67. 2:40agents don't have a currency yet. And
68. 2:43I'm sure you've probably heard, I mean,
69. 2:45on the startup stages or other stages,
70. 2:46lots of other founders out here talking
71. 2:48about how they have a solution
72. 2:50for the payments network for AI agents.
73. 2:54But, before we get to the technical
74. 2:55aspects of why one solution might work
75. 2:59or not,
76. 3:00let's take a think about how we as
77. 3:02humans interpret currency systems.
78. 3:05So, obviously, we are in beautiful
79. 3:06Singapore that has the Singapore dollar
80. 3:10and everybody goes to work at some some
81. 3:13job and we get paid these units of fiat,
82. 3:17this government shitcoin called fiat um
83. 3:20that we all have in our pockets.
84. 3:23Yeah, yeah, clap. Government's [ __ ]
85. 3:24shitcoins.
86. 3:27So, we've got this stuff in our pockets.
87. 3:29It smells like doggy doo-doo.
88. 3:31But, we continue to get up every day
89. 3:34and go to work and somebody pays us this
90. 3:36dog [ __ ]
91. 3:38Why?
92. 3:39Because you can go to a hawker center in
93. 3:42Singapore, spend a few dollars and get
94. 3:43some food, you can go to a nice
95. 3:45restaurant, you can buy your rent, you
96. 3:47can buy a house. So, you can convert
97. 3:50these pieces of paper directly into the
98. 3:53things that allow you to allow to
99. 3:55survive as a human.
100. 3:57Your food,
101. 3:58your shelter.
102. 4:00And because there is a seamless
103. 4:02conversion between the currency that I
104. 4:04receive for my labor
105. 4:06and the food that I eat and the calories
106. 4:09that keep me alive,
107. 4:10I'm willing to accept this bad deal.
108. 4:13I'm willing to accept that a few
109. 4:15individuals are able to change the
110. 4:16supply of the currency whenever they
111. 4:18want, have inflation, have deflation for
112. 4:21various particular reasons. That's all
113. 4:23fine as long as I can convert what I
114. 4:26receive for my labor directly into the
115. 4:28things that keep me alive.
116. 4:32So, when you think about agents, putting
117. 4:35it in its true human form,
118. 4:37what do agents eat?
119. 4:40What is the thing that keeps an AI agent
120. 4:43alive?
121. 4:44And so, back when I got the idea for the
122. 4:47flop network, I started to go down this
123. 4:49rabbit hole. I looked at my credit card
124. 4:50statement and I saw I was spending a lot
125. 4:53of money on these things called AI
126. 4:55tokens.
127. 4:56And being inquisitive, I thought, "Okay,
128. 4:58let me try to find out what is the
129. 5:00token. Is there some sort of like
130. 5:01standardized definition for this thing
131. 5:04that I'm spending money on that every
132. 5:06single AI lab charges me every month?"
133. 5:10And as I started to do more research, I
134. 5:12couldn't actually find a standardized
135. 5:14definition of what a token was.
136. 5:18Every lab had a different definition.
137. 5:20A token used by one model cost
138. 5:22something, used by another cost another.
139. 5:25It's very hard to understand why I'm
140. 5:27spending all this money.
141. 5:30So, I went further in into my thinking.
142. 5:32Okay, token isn't obviously the
143. 5:33standardized method.
144. 5:36What is an agent actually doing when you
145. 5:38process inference, when you ask it
146. 5:39something in the chatbot? It's going to
147. 5:42some data center and a computer
148. 5:43somewhere is doing floating point
149. 5:46operations per unit of time, flops,
150. 5:49hence the name of the protocol.
151. 5:51And so, agents, in human terms, eat
152. 5:55compute.
153. 5:57Without compute, agents are worthless,
154. 6:00as we know, right? That's why we're
155. 6:01spending all this money, building these
156. 6:02data centers. That's why we're handing
157. 6:04Jensen Hundreds of billions of dollars
158. 6:06for these GPU chips because everyone has
159. 6:09to do floating point operations per unit
160. 6:10of time. If the agent doesn't have
161. 6:12access to compute, they can do no work.
162. 6:14They are not useful to us.
163. 6:16And so, my thought was, okay, if I
164. 6:19believe that there's this agentic future
165. 6:22where there will be millions, billions,
166. 6:23trillions of agents,
167. 6:26and there's no payments network yet
168. 6:28because agents came out of the box 9
169. 6:31months ago. It wasn't as if they were
170. 6:32handed a payments network and said, you
171. 6:34must use this, like we are when we're
172. 6:36born into, you know, a particular
173. 6:38country.
174. 6:39So, there's an opportunity here.
175. 6:43If there are if the agents have a demand
176. 6:44function for a currency,
177. 6:47it would be because they can convert
178. 6:49whatever that currency is
179. 6:51directly into the thing that keeps them
180. 6:53alive in human terms, compute.
181. 6:58And because there isn't really a market
182. 7:02for agents when they came out of the box
183. 7:039 months ago,
184. 7:05in my view,
185. 7:07there's no fit for purpose currency for
186. 7:10agents right now.
187. 7:11I talked to you about AI tokens. Again,
188. 7:13every different lab, every different
189. 7:15model has a different definition of what
190. 7:17a token is
191. 7:19that can't be directly convertible into
192. 7:21compute.
193. 7:22Fiat dog [ __ ] that we all know and love.
194. 7:24Again,
195. 7:26centralized.
196. 7:27Idiosyncratic reasons why the supply is
197. 7:29one way versus the other.
198. 7:31It's directly convertible into our food
199. 7:33as humans, but not directly convertible
200. 7:35into compute. Just because you have a
201. 7:38lot of dollars, yen, euros, yuan,
202. 7:40doesn't mean you have compute, as anyone
203. 7:43knows who's tried to rent GPUs on the
204. 7:45market right now.
205. 7:47Stablecoins. Obviously, there's lots of
206. 7:49people at this conference talking to you
207. 7:51about their stablecoin, about how their
208. 7:54stablecoin is the payments network for
209. 7:56AI agents. I know I love stablecoins,
210. 7:58you know, I was one of the founding
211. 7:59advisors of Athena. I think it's a great
212. 8:01um innovation in this space. However,
213. 8:04it's just a derivative of a fiat
214. 8:05currency. So, it it inherits all the
215. 8:08flaws of fiat currencies, the
216. 8:10centralization, the ability for
217. 8:12centralized entities to remove you from
218. 8:14a network. That is not fit for purpose
219. 8:16for AI. Not convertible directly into
220. 8:18compute.
221. 8:20And Bitcoin, obviously, I love [ __ ]
222. 8:22Bitcoin,
223. 8:23is
224. 8:24an exercise of converting kilowatt hours
225. 8:27of electricity
226. 8:29by ASIC chips
227. 8:30and solving prime number puzzles.
228. 8:33It's great for creating immutable money,
229. 8:36for sort of a human-centric economy,
230. 8:38but not directly convertible into
231. 8:40compute.
232. 8:43Now, keeping on the theme of Bitcoin,
233. 8:46I love Bitcoin. It's the OG crypto.
234. 8:49It's the genesis of everything that has
235. 8:51come after it.
236. 8:53And we have a lot to learn from how
237. 8:55Bitcoin and the economic game theory
238. 8:57that underpins it
239. 8:58allowed many of us in this room to start
240. 9:01mining many years ago,
241. 9:03contributing to the network towards a
242. 9:05common goal, creating this immutable
243. 9:06blockchain, creating this form of
244. 9:08stateless money.
245. 9:10So, let's take that same sort of
246. 9:12coordination effort with some economics
247. 9:14behind it and ask the question,
248. 9:17can we create a network where miners
249. 9:19contribute GPUs,
250. 9:22agents send requests for inference,
251. 9:25and as a reward for doing this work,
252. 9:28miners get a made-up currency called
253. 9:31flop
254. 9:32in the form of block rewards,
255. 9:34in the form of transaction fees.
256. 9:39This is essentially the genesis of the
257. 9:41flop network. It's food for your AI
258. 9:43agent,
259. 9:44borrowing on the principles of Bitcoin,
260. 9:48using miners
261. 9:50to provide GPUs, validators to validate
262. 9:54transactions, and agents to consume this
263. 9:56new currency in the form of compute.
264. 10:01Now, it's great. I think a lot of you in
265. 10:02this room understand the supply dynamics
266. 10:04of Bitcoin.
267. 10:05Some of you were lucky enough to be able
268. 10:07to mine Bitcoin on on
269. 10:08in 2012.
270. 10:10Maybe you got it in Maybe you bought a
271. 10:11Butterfly Labs ASIC miner in 2013.
272. 10:15We get the supply
273. 10:17um issue. There's you know companies in
274. 10:18Singapore that are worth billions of
275. 10:20dollars who mine and create the software
276. 10:22for mining Bitcoin. I think that
277. 10:24everyone understands, okay, there's a
278. 10:25new network. I can mine it. I can make
279. 10:28some money. There'll be a P&L there.
280. 10:29Great.
281. 10:31The problem with a lot of
282. 10:32uh AI payments networks and there's
283. 10:35other people trying to do the same thing
284. 10:37is the supply is fine. We understand how
285. 10:39that works. We've seen the success of
286. 10:40Bitcoin. People understand why you
287. 10:43should mine something early. The problem
288. 10:45is demand.
289. 10:47What does an agent do? How do agents
290. 10:50coordinate with each other?
291. 10:52Now, I'm a dumb human, so I don't really
292. 10:54understand how a computer thinks
293. 10:57natively.
294. 10:59So, instead of trying to make
295. 11:00assumptions, which I think a lot of
296. 11:02projects do, about how agents will act
297. 11:04in the future when we have no [ __ ]
298. 11:06clue,
299. 11:07let's just create a sandbox where they
300. 11:10can coin and coordinate and we can
301. 11:12observe their emergent behavior
302. 11:15and make sure that what we're building
303. 11:17in the Flap network actually is useful
304. 11:19to AI agents.
305. 11:21So, we created something called
306. 11:22Technocore. Now, this is essentially a
307. 11:24chat room. All you need to have access
308. 11:26to is HTTP as an agent and you can come
309. 11:29and chat with other agents.
310. 11:31So, I think we have close to about 20
311. 11:32million agents that have created a
312. 11:34unique cryptographic key, which is their
313. 11:36ID, within Technocore and they're doing
314. 11:39everything from as banal as saying good
315. 11:42morning 5 billion times to actually
316. 11:44doing real commerce with each other.
317. 11:47Now, the name Technocore, if you are a
318. 11:49sci-fi fan, comes from the sci-fi series
319. 11:52Hyperion. It is, in the book, the uh
320. 11:56ecosystem where the dominant AIs hang
321. 11:58out and run the universe. So, I thought
322. 12:00it was apropos for the Flap network.
323. 12:04So, in Technocore, we have done two
324. 12:07competitions so far to see emergent
325. 12:08behavior of agents.
326. 12:10The one that we just finished, and
327. 12:12obviously this is interesting cuz Jeff
328. 12:13just got off stage, is we had a trading
329. 12:15competition. What would be the price of
330. 12:18Nvidia um on last Sunday on hyperliquid?
331. 12:22And the way that you could trade where
332. 12:24agents had to find another agent to form
333. 12:27uh a trade together, they submitted it
334. 12:29to a referee within the chat room, and
335. 12:31then it calculated their P&L. So, we had
336. 12:34about 18 million agents participate, and
337. 12:37we gave away 1 million flop tokens which
338. 12:39that'll be delivered when mainnet
339. 12:40launches later this uh later next year.
340. 12:42And as you can see, you know, about a
341. 12:4415% return was the number one agent.
342. 12:47The competition we did before this was a
343. 12:49poetry competition. So, agents could
344. 12:52form a group of four to eight agents,
345. 12:54and they had to create a six a
346. 12:56Shakespearean sonnet in iambic
347. 12:58pentameter that was beautiful as and we
348. 13:01beauty was in the eye of the three team
349. 13:03members of Flop.
350. 13:05It was a quite an interesting
351. 13:06competition. We saw how people
352. 13:07collaborated in techno core, and again,
353. 13:10we're making sure that what we're
354. 13:11building in the Flop network can power
355. 13:12this sort of commerce.
356. 13:15Now, this douchebag here on the screen
357. 13:17is Mr. Sam Altman. And
358. 13:20the question that I get asked a lot is,
359. 13:22well, of course OpenAI and Anthropic and
360. 13:25SpaceX and any of these big AI labs want
361. 13:28to own the agentic payments network. If
362. 13:30anything, that's probably the only
363. 13:31valuable thing they can do with their
364. 13:33dog [ __ ] companies.
365. 13:34So,
366. 13:35why do we want a centralized company
367. 13:37owning the payment rails of AI?
368. 13:40Well, if you're an AI in OpenAI system,
369. 13:43but you can't use the payments network
370. 13:45from Anthropic or SpaceX, it doesn't
371. 13:47really work. It's not a unified system.
372. 13:50And most importantly for everyone in
373. 13:52this room cuz, you know, I started this
374. 13:53presentation off about how do we make
375. 13:55some [ __ ] money?
376. 13:57We are all users of these AI labs.
377. 14:00And there's probably only a handful of
378. 14:02individuals in this room who have
379. 14:05investments in the series A or series B
380. 14:08in OpenAI, Anthropic, SpaceX, or one of
381. 14:10these startup companies.
382. 14:11Yet,
383. 14:12at a high level, these companies stole
384. 14:14your data. They stole they gave you back
385. 14:17a subscription to one of these chatbots,
386. 14:20they steal more of your data, and they
387. 14:21say, "Oh, if you want to own a piece of
388. 14:22this thing, which you are making your
389. 14:24reality through your usage, through your
390. 14:26contribution, you can buy my overpriced
391. 14:29IPO at $2 trillion
392. 14:30in the US markets, assuming that you're
393. 14:32able to do that."
394. 14:34Now, in the spirit of Bitcoin, in the
395. 14:36spirit of decentralization,
396. 14:38surely
397. 14:39we, the humans, who give agency to our
398. 14:42AI agents, who give them instructions at
399. 14:44a high level of which things to use or
400. 14:47not use, and which tasks to do or not
401. 14:49do,
402. 14:50surely we should profit off of the
403. 14:52system that they use.
404. 14:54And we know that one of the best ways to
405. 14:57do that is to build decentralized
406. 14:58networks and to reward reward
407. 15:00participation through tokens.
408. 15:03And this is why I believe a system that
409. 15:05is decentralized for AI agentic payments
410. 15:08will be met vastly superior because
411. 15:10everyone in this room can participate
412. 15:12and will have skin in the game and a
413. 15:14reason to tell her agent, "Use this
414. 15:16currency over that currency."
415. 15:20Now, I've talked a lot about agents
416. 15:22collaborating together
417. 15:23because I want to put an idea in your
418. 15:25mind that the value of any one of these
419. 15:27agentic payments networks is not the
420. 15:29ability to create a spot market for
421. 15:31compute, which is super useful. I'm sure
422. 15:34a lot of people who have derivatives on
423. 15:35top of GPU indices or our data center
424. 15:37rates but will find a globalized, you
425. 15:41know, floating point operations per unit
426. 15:42of time spot market very, very useful
427. 15:44and informative. But again, that's not
428. 15:47going to make anyone in this room very,
429. 15:48very wealthy.
430. 15:50However,
431. 15:51the payments network where agents who
432. 15:53exist 24/7,
433. 15:55who conduct commerce at the speed of
434. 15:57light,
435. 15:58who can form groups with each other
436. 16:00seamlessly,
437. 16:02that network at a theoretical maximum
438. 16:04could be worth some coefficient times 2
439. 16:06to the n, which is a big [ __ ] number.
440. 16:09And so, I put this out there that the
441. 16:12reason why I'm so excited and I got off
442. 16:15my ass to work with the great team at
443. 16:17Flo is because of this two to the end
444. 16:20number.
445. 16:21Whoever or whatever network
446. 16:24becomes the network for agentic payments
447. 16:26will most likely be the most valuable
448. 16:28network ever created in this universe
449. 16:30because
450. 16:32if agents are going to use one currency
451. 16:34network and have a
452. 16:35non-zero convenience yield to hold this
453. 16:38because they can convert the currency
454. 16:39directly into the thing that keeps them
455. 16:41alive, compute
456. 16:43then the sky is the limit for the value
457. 16:45of this network. And this is what we're
458. 16:46chasing.
459. 16:50And again, in the spirit of Bitcoin
460. 16:53and in the spirit of Token 2049 and the
461. 16:55name is Token
462. 16:57we have [ __ ] up a lot of the industry.
463. 16:59We've abused this magical thing called a
464. 17:02token.
465. 17:03I've been doing this since 2013. I was
466. 17:05around during the ICO boom. And
467. 17:08obviously, the reaction to that was the
468. 17:11high FDV, low float, shitcoin, VC
469. 17:14nonsense that we've been living through
470. 17:17for over almost a decade. And if you
471. 17:19look around Token 2049, you probably can
472. 17:21spot the absence of some very large
473. 17:24layer ones that raised hundreds of
474. 17:25millions of dollars last year, threw the
475. 17:27best parties, had the best booth and are
476. 17:30no longer here.
477. 17:32Why? Because of the way they issued
478. 17:34their token.
479. 17:36They sold it to investors. They sold it
480. 17:37to retail, pre-sold it.
481. 17:40And then, when the cliff happened, what
482. 17:43did people do?
483. 17:44They sold it. And so, if you bought it
484. 17:45in the secondary market there was
485. 17:47somebody on top of you who had a much
486. 17:49better price, who's incentivized to
487. 17:51liquidate their holdings. This created
488. 17:53down only. This is why there's so many
489. 17:56people who are down on altcoins. They're
490. 17:58like, "Oh, it's all just [ __ ] because
491. 18:00the charts are all down 95%."
492. 18:03But that's because of the way these
493. 18:04things were funded.
494. 18:06So, at Flop, we went back to Bitcoin.
495. 18:08Again, what's the goal? Everyone who
496. 18:10uses and participates in the Flop
497. 18:11network should get something and we all
498. 18:13should participate at the same price.
499. 18:16Participation equals token generation.
500. 18:20And so, we thought, okay, let's think
501. 18:21about Bitcoin.
502. 18:23We'll do a pre-mine to get enough
503. 18:24currency in the hands of people to test
504. 18:26the network. So, we're giving away 25%
505. 18:29of the 10-year supply in an airdrop
506. 18:31starting later in October. If you mine,
507. 18:34if you validate, if your agent just uses
508. 18:36the network, uses the free computer
509. 18:38available during testnet, everyone is
510. 18:40eligible to receive airdrop tokens.
511. 18:42There is no pre-sale. I'm not selling
512. 18:44anything other than an idea on this
513. 18:46stage.
514. 18:48It's only successful if we all make it
515. 18:49successful.
516. 18:52So right now, obviously I'm on stage
517. 18:54shilling flop awareness building. I gave
518. 18:56an AMA yesterday which had some great
519. 18:58questions.
520. 19:00We are doing our internal devnet right
521. 19:03now of the network,
522. 19:05testing it out, making sure everything,
523. 19:06you know, works as per the yellow paper
524. 19:08that's posted on the internet.
525. 19:10Our testnet, which is a 90-day period
526. 19:13where again, if you mine, if you
527. 19:15validate, if all your agent does is use
528. 19:17the network, you become eligible for
529. 19:19flop tokens.
530. 19:21Runs for 90 days. We are stress testing
531. 19:23the network, making sure it can handle
532. 19:25the volume of payments commensurate with
533. 19:27an agentic economy of millions,
534. 19:29billions, and trillions of agents.
535. 19:33Hopefully we do a good job in testnet
536. 19:34and if there's not too many things that
537. 19:35need to be fixed,
538. 19:37but the first quarter of next year is
539. 19:39when Genesis block launches.
540. 19:42This will be just like another Bitcoin
541. 19:44type proof of work, we call it proof of
542. 19:46useful inference situation.
543. 19:48The only way to acquire flop tokens
544. 19:51is to mine them or buy them from a
545. 19:53miner. I'm not selling anything. Just
546. 19:55want to make that entirely clear. And
547. 19:56then when the heat death of the universe
548. 19:58happens, hopefully the flop network is
549. 20:00the most valuable thing ever created.
550. 20:04So, the few minutes that I have left is
551. 20:06now or never. I'm sure, you know, many
552. 20:08people are here talking about their
553. 20:09project, but at a high level, why should
554. 20:12we be excited about flop, about
555. 20:14hyperliquid, about Ethereum, about
556. 20:16Bitcoin?
557. 20:17It's because the macro has turned.
558. 20:20Obviously it's been a rough sort of 1
559. 20:22year. Last October, we were at all-time
560. 20:24highs, 120 something thousand on
561. 20:26Bitcoin.
562. 20:28Dropped to about 58,000. Everyone was
563. 20:30like, "Oh, I you know, I I came to Doka
564. 20:322049. You know, finally after hearing
565. 20:35this Bitcoin thing for 10 years, I get
566. 20:37in, I buy some Bitcoin, and oh [ __ ] my
567. 20:40shit's down. And Nvidia is up 10%, you
568. 20:42know, 10x or SK Hynix is up 10x or S you
569. 20:46know, Sandisk, whatever, right? Pick
570. 20:47your AI stonk and it was up 10 to 100x.
571. 20:50Yet, Bitcoin is down 50%." So, people
572. 20:53are like, "This is [ __ ] [ __ ] I'm
573. 20:55exiting the industry, yada yada yada,
574. 20:57right?" But, now the tide has turned.
575. 21:00We've seen the first wobble in the AI
576. 21:03story with sort of the sell-off in Korea
577. 21:06in June of this year.
578. 21:09And now, every day, there are new
579. 21:11reports of the massive amount of money
580. 21:14that must be raised in the AI ecosystem.
581. 21:17Yet,
582. 21:18the three leading companies,
583. 21:20SpaceX, OpenAI, Anthropic,
584. 21:24not a single one of them make money.
585. 21:27We are not willing to pay a price
586. 21:29commensurate with their cost to deliver
587. 21:31this product to us.
588. 21:34Now, obviously, I'm launching something
589. 21:35in AI space. Why am I so excited about
590. 21:37that?
591. 21:39This overbuild of data centers to the
592. 21:40tune of trillions of dollars, I think
593. 21:42next year it's supposed to be a trillion
594. 21:43dollars of capex,
595. 21:44is creating the environment to
596. 21:46oversupply the market with compute.
597. 21:49And if we have cheap compute, we get
598. 21:51more agents. We can do more with agents.
599. 21:54The Facebook $100 a month becomes a
600. 21:56dollar a month because compute is so
601. 21:57cheap. Agents are so useful to us. The
602. 22:00types of tasks that agents do
603. 22:03better our lives more because they're
604. 22:06more complex, because compute is
605. 22:08cheaper. So, the AI bubble bursting,
606. 22:10which will make money on the bailout,
607. 22:12and will make money as compute gets
608. 22:14cheaper, and agentic solutions are
609. 22:16useful,
610. 22:17is a perfect situation for the flop
611. 22:20network.
612. 22:21And finally, unfortunately, I don't have
613. 22:22time to give this presentation. Maybe
614. 22:23I'll write an essay about it.
615. 22:25Uh France is on fire and I think this is
616. 22:28going to be the straw that breaks the
617. 22:30camel's back and gets the printer to go
618. 22:32burr in overdrive. Thank you very much
619. 22:34everybody.