YouTube2Text

Is the AI Boom Over? Let's Talk About It. — Transcript

by AI News & Strategy Daily | Nate B Jones · 3,463 words · 545 segments · language en · Watch on YouTube

Full transcript

  1. 0:00The AI data center build-out had a
  2. 0:01terrible night on election night,
  3. 0:03November 3rd, 2026. AI is not popular.
  4. 0:07AI is something that most Americans are
  5. 0:10beginning to feel like
  6. 0:12is set against the American dream, is
  7. 0:14set against the idea of individuals
  8. 0:17achieving what they want to achieve in
  9. 0:18their lives.
  10. 0:20AI is being positioned as disempowering.
  11. 0:23In large measure,
  12. 0:24Silicon Valley has itself to blame on
  13. 0:26this. We have had every opportunity to
  14. 0:29message this differently,
  15. 0:31and we haven't.
  16. 0:33We haven't.
  17. 0:34People are out there mocking us and
  18. 0:36talking about
  19. 0:38how this is a victory for ordinary
  20. 0:42people versus people who want AI to make
  21. 0:45humans extinct. You're not winning if
  22. 0:48you're talking about making humans
  23. 0:49extinct.
  24. 0:50You're just not. Despite the vote,
  25. 0:52Americans use AI a lot. There is no sign
  26. 0:56of disinterest or lack of investment
  27. 0:58from American companies in AI. It's not
  28. 1:00that we're going to stop.
  29. 1:02It's that Americans seem to expect that
  30. 1:04that AI will come without data centers.
  31. 1:07They seem to expect that that AI will
  32. 1:09come [snorts] out of the air. Well, it
  33. 1:11doesn't.
  34. 1:12That AI is going to come from somewhere,
  35. 1:14and that's going to produce some
  36. 1:14outcomes that look really weird if
  37. 1:16you're only watching the politics. But
  38. 1:19this is not a video about data centers
  39. 1:21per se. This is a video about the second
  40. 1:23and third-order implications that happen
  41. 1:26when the underlying compute
  42. 1:28infrastructure that powers our new
  43. 1:31economy is constrained.
  44. 1:35I don't think people voting thought that
  45. 1:37one through, and that's fine,
  46. 1:39but I do think we have to think it
  47. 1:40through and understand what it means to
  48. 1:43make effective decisions in the next
  49. 1:44couple of years. A May survey from
  50. 1:46Edison Research at SSRS found that 65%
  51. 1:50of American adults had used an AI
  52. 1:52chatbot in the previous week, and that
  53. 1:53was 6 months ago, and that was up from
  54. 1:5652% in February. An October poll from
  55. 1:59ABC News, Post, and Ipsos found that 72
  56. 2:03% of Americans opposed a new data center
  57. 2:06in their community. So, widespread use
  58. 2:09of AI and widespread opposition to the
  59. 2:11infrastructure AI depends on can and do
  60. 2:14exist at the same time in the same
  61. 2:16population.
  62. 2:18The election just made the consequences
  63. 2:19of those two opposing choices very
  64. 2:22concrete. Communities across the country
  65. 2:24have voted to block or restrict data
  66. 2:26center development over and over and
  67. 2:28over again. In Edgewater, Florida,
  68. 2:30unofficial results show more than three
  69. 2:32quarters of voters back to prohibition
  70. 2:34on new data centers, and that is not
  71. 2:36unusual. Those are typical margins for
  72. 2:38what we saw across the country on
  73. 2:40November 3rd. The industry is going to
  74. 2:42have to deal with the implications of
  75. 2:44that.
  76. 2:45But the customers who need the computing
  77. 2:47power for AI
  78. 2:49are still there, and their needs are
  79. 2:51growing. Google said on its latest
  80. 2:53earning call that usage through its
  81. 2:55model APIs is up 40% quarter over
  82. 2:58quarter and more than four times what it
  83. 3:00was a year earlier. That's not slowing
  84. 3:02down. It also said it had started using
  85. 3:04outside compute capacity because of
  86. 3:06supply constraints on Google Cloud. That
  87. 3:08is a super different signal from the one
  88. 3:10coming out of the ballot box, folks. If
  89. 3:12we put those two things together, people
  90. 3:14are using more AI. It's getting harder
  91. 3:16to build the infrastructure to serve
  92. 3:17them in the places company expected to
  93. 3:19build it, and that changes where the
  94. 3:21dollars are going to go. It changes what
  95. 3:23becomes valuable in the next couple
  96. 3:25years and which companies can make their
  97. 3:27plans work. It also changes the odds of
  98. 3:30an AI infrastructure bubble. And yes, it
  99. 3:32gives the people trying to put data
  100. 3:34centers in space another reason to keep
  101. 3:36on going. Now look at the companies that
  102. 3:38have already made commitments based on
  103. 3:40where they thought the build out was
  104. 3:42going. This is where the state of Texas
  105. 3:44in particular gets really interesting.
  106. 3:46There are specialist cloud companies
  107. 3:47called Neo Clouds that rent out the
  108. 3:49powerful chips that are used for AI, GB
  109. 3:51300s and others. Their business depends
  110. 3:53on getting those chips installed,
  111. 3:55powered, and earning revenue. If your
  112. 3:57plan assumed as a Neo Cloud that you
  113. 3:59could expand quickly on inexpensive low
  114. 4:02regulatory Texas sites, a longer or more
  115. 4:05restrictive approval process may not
  116. 4:07just change the economics of your plan,
  117. 4:09it may disrupt it enough that the
  118. 4:10business isn't viable.
  119. 4:12Texas was already restricting permits
  120. 4:14before the election, and the question is
  121. 4:16now what happens to the duration and
  122. 4:19conditions of those restrictions after
  123. 4:22electing a vocally concerned governor
  124. 4:25around data centers. And for work
  125. 4:27already underway, what matters is how
  126. 4:29long do approvals take? How long do
  127. 4:31power connections take? A building under
  128. 4:34construction and an operating facility,
  129. 4:36those are two different things, right?
  130. 4:37You have to actually get the power on.
  131. 4:39So imagine you've borrowed money and
  132. 4:40you've committed to equipment and you've
  133. 4:42signed a customer who expects capacity
  134. 4:44by a specific date. A delay pushes out
  135. 4:47or cancels that revenue. Your financial
  136. 4:48obligations, they keep arriving. The
  137. 4:50equipment is still depreciating. You may
  138. 4:53have to buy capacity from someone else
  139. 4:55to serve the customer you already
  140. 4:57signed. That might preserve the
  141. 4:58relationship, but it's going to eat your
  142. 4:59margins. The danger in this game is
  143. 5:02running out of money before you can
  144. 5:04serve demand the way you expected to.
  145. 5:06And there can still be plenty of
  146. 5:08customers who want what you are
  147. 5:10building, and you will still fail in
  148. 5:12that position because you can't afford
  149. 5:14to wait. And this is where consequences
  150. 5:17start looking almost backward. The
  151. 5:19company that fails doesn't necessarily
  152. 5:23take the project with it because AI
  153. 5:25demand keeps rising. Someone else is
  154. 5:27going to buy that project. So the buyer
  155. 5:29that has cash and existing operations
  156. 5:32can afford to wait longer and snap up
  157. 5:35those deals.
  158. 5:36It may be able to serve those customers
  159. 5:38from another facility while the delayed
  160. 5:40project gets finished on its own
  161. 5:41timeline. It can acquire equipment and
  162. 5:43contracts or a partly completed site at
  163. 5:45a competitive price the original
  164. 5:48investors wouldn't have accepted when
  165. 5:49they thought everything was going
  166. 5:50according to plan. Meanwhile, this
  167. 5:52advantage extends anywhere where there
  168. 5:54are already working data centers. A
  169. 5:56company with a working data center
  170. 5:58already has capacity that newer
  171. 6:00competitors are fighting to bring
  172. 6:01online. As long as demand is holding up,
  173. 6:04fewer competing facilities means
  174. 6:06customers have fewer places to go, and
  175. 6:08that can support higher utilization,
  176. 6:10better pricing, lower depreciation
  177. 6:12rates, and stronger negotiating power
  178. 6:15for existing data centers. So the
  179. 6:16backlash against the AI build out may
  180. 6:19end up, ironically, strengthening big
  181. 6:21tech. The large operators that have
  182. 6:23projects exposed to delays
  183. 6:26can afford them. But the companies that
  184. 6:27have
  185. 6:29new investments that are startup like,
  186. 6:31that are neo clouds,
  187. 6:33they are going to have a harder time
  188. 6:35surviving the new construction
  189. 6:38schedules. In other words, the industry
  190. 6:40may take a very public political defeat
  191. 6:42and come out more concentrated around
  192. 6:44classic big footprint, big company kinds
  193. 6:47of places. So if you're a smaller AI
  194. 6:49company and you're trying to compete,
  195. 6:50you may end up renting more of your
  196. 6:52capacity from the big players that
  197. 6:55you're trying to compete with in other
  198. 6:56spheres. Your access to computing ends
  199. 7:00up becoming a bigger lever in whether
  200. 7:03your product is viable, and that lever
  201. 7:05is something that those big companies
  202. 7:06are sitting on even if you're competing
  203. 7:08with them. There's another consequence
  204. 7:10that points in a different direction,
  205. 7:11though. Slowing down construction may
  206. 7:14make the overall boom less likely to
  207. 7:16collapse from overbuilding. Derek
  208. 7:19Thompson made this argument after the
  209. 7:20election, and I think he was right. If
  210. 7:22demand is growing and politics slows the
  211. 7:24growth of supply in tokens, demand has
  212. 7:28more time to catch up, and that reduces
  213. 7:30the risk that a wave of new facilities
  214. 7:32arrives before there are enough paying
  215. 7:34customers to use them. Think about a
  216. 7:36group of developers all looking at the
  217. 7:38same growth forecast. Every single one
  218. 7:40of them decides we're going to go for
  219. 7:42it. Every individual project is going to
  220. 7:43look reasonable, but collectively they
  221. 7:46end up building out way more capacity
  222. 7:49than the market needs at the moment.
  223. 7:50That's how you get into a risky position
  224. 7:53where you have expensive facilities and
  225. 7:54too little demand. Now, to be very
  226. 7:56clear, we have not yet seen that, but
  227. 7:59that was a risk we were worried about
  228. 8:01from an AI perspective in 2027.
  229. 8:04And so right now, after November 3rd,
  230. 8:08we're looking at a world where customers
  231. 8:09have more time to grow into that
  232. 8:11capacity, existing facilities can stay
  233. 8:13busier. So moderating the US build out
  234. 8:16in 2027 could ironically reduce the
  235. 8:18chance of a supply glut over the
  236. 8:20following year or two, even as it
  237. 8:22creates very, very serious problems for
  238. 8:25individual developers, especially
  239. 8:26smaller players. So this can be a
  240. 8:28situation where a company running out of
  241. 8:31cash before the opening ends up going to
  242. 8:34the wall and having to liquidate on the
  243. 8:36cheap, but net net that effect across
  244. 8:40the economy ends up moderating the
  245. 8:42supply of tokens enough that we avoid a
  246. 8:45bubble situation entirely.
  247. 8:47Now, I'm not here to tell you what any
  248. 8:49given stock is worth or what any given
  249. 8:51Neo cloud is worth. I'm here to help you
  250. 8:54understand what the implications are of
  251. 8:57the political choices that Americans
  252. 8:59made on November 3rd, many of which I'm
  253. 9:01sure did not cross most voters' minds.
  254. 9:03You don't have to know what happens in
  255. 9:042028 to follow the implications here.
  256. 9:08All you have to do is look at it as a
  257. 9:10story of demand that has been blocked.
  258. 9:12Start with a company that needs more
  259. 9:14compute capacity next year. Literally
  260. 9:16every company I know. It might be
  261. 9:17training a model or running an AI
  262. 9:19service or serving customers who want
  263. 9:21their agents working all day.
  264. 9:23If their preferred project gets delayed,
  265. 9:25that need does not politely wait for the
  266. 9:28next legislative session. The company
  267. 9:30looks for another place to get work
  268. 9:32done.
  269. 9:33The first alternative might be signing
  270. 9:35up with a different cloud provider,
  271. 9:36working indirectly through a different
  272. 9:38community in the US. Suddenly a location
  273. 9:40that looked more expensive might be
  274. 9:42okay. Maybe it has power that can
  275. 9:44actually be delivered. Maybe the water
  276. 9:46is there and maybe the community is
  277. 9:48actually supportive and the developer
  278. 9:49has a good relationship with the
  279. 9:50community in the local power station.
  280. 9:52All of that is valuable in a way that
  281. 9:55cheap land by itself is not. If you can
  282. 9:57open on time, you're offering something
  283. 9:59different than anybody else right now.
  284. 10:01You're offering a customer of a data
  285. 10:02center the ability to deliver their
  286. 10:04product as promised and the cost of
  287. 10:06missing that date trickles all the way
  288. 10:09through the supply chain for AI. They
  289. 10:10think this changes the bargaining
  290. 10:12position that communities are in. They
  291. 10:14have something developers increasingly
  292. 10:16need, a credible place to build.
  293. 10:19That gives them room to negotiate over
  294. 10:21things like infrastructure costs, union
  295. 10:23jobs, local investment, what the
  296. 10:25developer has to deliver.
  297. 10:27A developer is going to accept more
  298. 10:29expensive terms because a dependable
  299. 10:31agreement is worth way more than more
  300. 10:34uncertain years trying to get this
  301. 10:36project off the ground. That makes the
  302. 10:38usual map of friendly and unfriendly
  303. 10:40states not super useful. You have to get
  304. 10:42down to specific sites, look at power
  305. 10:44agreements, look at local support. If
  306. 10:47every displaced project piles into the
  307. 10:49same few welcoming spots on the map,
  308. 10:52those places are going to face more
  309. 10:54pressure, too. So ultimately, the
  310. 10:57advantage is going to go to projects
  311. 10:59that are already on the ground in
  312. 11:00friendly states and communities that
  313. 11:03stayed friendly through November 3rd and
  314. 11:06that are going to make room for data
  315. 11:07centers locally. And some of that search
  316. 11:10because there's just only so much space
  317. 11:12in the US after November 3rd, that's
  318. 11:14going to go outside the US. There are
  319. 11:15countries actively trying to attract
  320. 11:18that investment. I know that Europe is
  321. 11:20campaigning for this with their
  322. 11:22gigafactory initiative. The UK has a new
  323. 11:24agreement setting out what it expects
  324. 11:26from data center developers. Net net, I
  325. 11:29hear much more interest outside the EU
  326. 11:32from folks developing data centers than
  327. 11:34inside the EU. And the reason for that
  328. 11:36is regulatory. Building in the EU
  329. 11:40is not perceived as being a whole lot
  330. 11:42faster than building in the US even
  331. 11:44after November 3rd. And so I think that
  332. 11:46developers are going to look other
  333. 11:48places. They're going to look at places
  334. 11:49that can set their own rules. I think
  335. 11:51[snorts] Canada is an interesting place.
  336. 11:54Mexico is an interesting place. They're
  337. 11:55both geographically proximate to the US.
  338. 11:58And I think anywhere else in the world
  339. 12:01that has stable land, stable power, and
  340. 12:04relatively friendly geopolitical
  341. 12:06relationship with the US stands to gain
  342. 12:09from the decisions voters made. So India
  343. 12:11comes to mind, Southeast Asia
  344. 12:12potentially, places in South America as
  345. 12:15well. And jobs will follow that
  346. 12:16investment pattern wherever it is, in
  347. 12:18the US or outside it. Someone has to
  348. 12:20construct the building, someone has to
  349. 12:22install the electrical equipment,
  350. 12:23someone has to operate facilities. Some
  351. 12:25of that employment, some of that
  352. 12:27spending is going to move abroad while
  353. 12:30the AI service gets sold back,
  354. 12:32effectively rented back to American
  355. 12:34customers. And that's an implication I
  356. 12:36think a lot of people miss. If you keep
  357. 12:38a data center out of the community, it
  358. 12:40doesn't keep the technology out. It just
  359. 12:42means you're renting the technology and
  360. 12:44you don't own it. My job is to analyze
  361. 12:46the second and third order effects that
  362. 12:47people may not have realized. I think
  363. 12:48the international shift is a classic
  364. 12:50example. I doubt anybody was thinking
  365. 12:52about where data centers go other than
  366. 12:54not in my neighborhood.
  367. 12:56But those of us who are involved in AI
  368. 12:58do have to think about that. Token
  369. 13:00demand keeps growing. If we're figuring
  370. 13:03out where to move data centers, we have
  371. 13:05to start to do that now, assuming that
  372. 13:08we have an uncertain political
  373. 13:09environment for at least the next two
  374. 13:11years in the US and that we have to
  375. 13:12figure out state by state where we can
  376. 13:15site projects. And if we can't site them
  377. 13:17at all in the states, where can we go
  378. 13:19that isn't the US that's still friendly?
  379. 13:21This brings me to perhaps the most
  380. 13:23speculative bet of 2026, which is the
  381. 13:28increased odds that the orbital data
  382. 13:30center project is going to get real.
  383. 13:32Now, we already know that XAI has said
  384. 13:34publicly they're going after this. Elon
  385. 13:36has talked about this. Jensen has talked
  386. 13:38about this. Google already has a project
  387. 13:40Suncatcher prototype in orbit, which not
  388. 13:43a lot of people know about.
  389. 13:45So, there's movement here, but there is
  390. 13:47a huge difference between a prototype in
  391. 13:50orbit and whether you can overcome the
  392. 13:52hardware conditions around radiation,
  393. 13:55around thermal, around depreciation,
  394. 13:57around getting and maintenance of your
  395. 13:58data center to actually get orbital
  396. 14:00compute live. All of that being said,
  397. 14:03that is a series of technical
  398. 14:05challenges.
  399. 14:07And if you run those down, you are in a
  400. 14:09much looser regulatory environment in
  401. 14:12orbit, and that is going to be something
  402. 14:14that the players who can reach orbit,
  403. 14:16namely Google and xAI, are very very
  404. 14:19aware of, and that they are going to be
  405. 14:21incentivized to lean into after this
  406. 14:24election. So ironically,
  407. 14:26this election has, in my mind, increased
  408. 14:30the odds that we will attack those
  409. 14:32technical problems with more capital and
  410. 14:34that we will find out very quickly
  411. 14:37whether orbital compute works at scale
  412. 14:40and that we will push on the space tech
  413. 14:42tree hard enough to get there faster.
  414. 14:45That doesn't mean that we're going to
  415. 14:46have orbital compute in two years, but
  416. 14:49it means that if I were a betting
  417. 14:50person, my bet would go from 10 years to
  418. 14:53maybe six to eight because this election
  419. 14:57creates enough uncertainty around
  420. 14:59compute in the US that you'd rather have
  421. 15:01an option in orbit. We'll see. It's not
  422. 15:04a done deal,
  423. 15:05but it's something that I think is
  424. 15:06higher probability now. So if you want
  425. 15:08to understand what the election means
  426. 15:10for AI, watch what companies do with the
  427. 15:12next project they were planning. Are
  428. 15:14companies waiting? Are they moving? If
  429. 15:16they're moving, where to? Are they
  430. 15:17paying more for a location they can
  431. 15:19deliver on? Are they buying capacity
  432. 15:21from someone else? Are they investing
  433. 15:23and producing that capacity somewhere
  434. 15:25that doesn't have the same constraints?
  435. 15:27The way companies make those decisions
  436. 15:29around token supply over the next couple
  437. 15:31of years will tell you a whole lot more
  438. 15:34about how committed that company is to
  439. 15:36AI than whatever they put in a press
  440. 15:38release.
  441. 15:40I'm always a fan of reading where people
  442. 15:42are allocating capital on the balance
  443. 15:43sheet because I think that tells the
  444. 15:44truth.
  445. 15:45And for your own work, be careful about
  446. 15:48using construction headlines as a
  447. 15:50forecast for adoption.
  448. 15:52Your employer is going to keep buying AI
  449. 15:55services and they will figure out how to
  450. 15:57get AI from somewhere.
  451. 15:59The price may go up a little bit,
  452. 16:01available suppliers may change, but
  453. 16:03fundamentally the cat is out of the bag
  454. 16:05on AI and demand is going to keep
  455. 16:07growing. And so
  456. 16:10the challenge that I have for all of us
  457. 16:13is in this world
  458. 16:15where we very clearly have a lot of
  459. 16:18political constraints.
  460. 16:20How do we understand not just the second
  461. 16:22and third order effects I've discussed
  462. 16:24here, but understand that this rebuke of
  463. 16:27data centers came from a place of
  464. 16:31concern about the future of AI in the
  465. 16:34American Republic.
  466. 16:36The United States has been a place where
  467. 16:38individual freedom is celebrated, where
  468. 16:40the American dream is celebrated.
  469. 16:42If AI feels disempowering, if AI feels
  470. 16:46like it is against the American dream,
  471. 16:49we're not going to win.
  472. 16:51It doesn't matter. I had a note after
  473. 16:53the midterms from a friend of mine who's
  474. 16:55in tech
  475. 16:57and
  476. 16:58he feels left out of the AI race. And
  477. 17:01get this, he's at an AI company and he
  478. 17:04doesn't feel caught up. He worries about
  479. 17:06his job.
  480. 17:07If that's the world that we're living
  481. 17:09in,
  482. 17:10if that's the world we're building, if
  483. 17:11that's the world we're talking about,
  484. 17:13we're just doing it wrong. We're just
  485. 17:15doing it wrong. We need to think about
  486. 17:17our messaging to
  487. 17:19voters, to communities across the United
  488. 17:22States
  489. 17:24so that we are clear about the benefits,
  490. 17:26why AI matters, why AI adds value to
  491. 17:29your day, and doesn't just automate away
  492. 17:31your job, why AI is not just something
  493. 17:34you worry about over the Thanksgiving
  494. 17:36dinner table as an extinction risk,
  495. 17:37which by the way, we're going to have to
  496. 17:40talk about at Thanksgiving now. Like
  497. 17:41that is enough in the public discourse.
  498. 17:43So we're going to get asked about that
  499. 17:45all at the same time that people are
  500. 17:48using Muse more, people are using
  501. 17:49ChatGPT more, people are using AI at
  502. 17:51work more. That's the world we're living
  503. 17:53in. How do we take that demand and make
  504. 17:56it into demand that people are hopeful
  505. 17:58about using, bring in American optimism?
  506. 18:02That's the larger question over the next
  507. 18:04two years. That's the question we can
  508. 18:05shape with storytelling. That's the
  509. 18:07question we can shape with how we talk
  510. 18:09about empowerment and tooling and
  511. 18:12training and preparing the American
  512. 18:14workforce for what's next. And it's a
  513. 18:15global challenge too. It's not like many
  514. 18:18in Europe are optimistic on AI. If you
  515. 18:20look at public surveys in Europe,
  516. 18:22they're not very optimistic either.
  517. 18:24We need to think about this globally as
  518. 18:26an opportunity or we're going to end up
  519. 18:29fragmenting out what AI could be around
  520. 18:33a bunch of local challenges to
  521. 18:37particular AI infrastructure sites. That
  522. 18:39would be a tragedy. We would be missing
  523. 18:41the potential of what AI could deliver.
  524. 18:43And that's what a lot of my channel has
  525. 18:45been about, the tools I release, Open
  526. 18:47Brain, Open Engine, talking about some
  527. 18:50of the work I've done on making my talks
  528. 18:53and thinking available in the MCP. It's
  529. 18:55all about basically making it easy for
  530. 18:58you to get perspective, empower
  531. 19:01yourself, and get where you want to go
  532. 19:03and set your goals up. I believe
  533. 19:04strongly in
  534. 19:06people being able in this moment
  535. 19:08especially to use AI to get where they
  536. 19:11want to go.
  537. 19:13And I've seen it over and over again
  538. 19:14that people have been able to figure
  539. 19:15that out, figure out how to get started,
  540. 19:18and then they're off to the races.
  541. 19:20And you can do that too. I'll see you
  542. 19:22next time. You tell me
  543. 19:23where you think AI is going after
  544. 19:26November 3rd, 2026 in the comments.
  545. 19:28Cheers.

About this transcript

This page contains the full transcript of Is the AI Boom Over? Let's Talk About It. by AI News & Strategy Daily | Nate B Jones, generated from the public captions YouTube serves with the video. The transcript has 3,463 words across 545 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.