Colette Kress takes charge of Nvidia: how its AI bet could change — Transcript
Full transcript
- 0:00This is Colette Kress, and she just took
- 0:02charge of NVIDIA, the company that makes
- 0:04the chips behind most of the world's
- 0:06artificial intelligence. And she doesn't
- 0:08come from designing chips.
- 0:11Until now, she had a different job, and
- 0:14NVIDIA's announcement puts it on the
- 0:15same line as her promotion.
- 0:17Colette Kress, Executive Vice President
- 0:20and Chief Financial Officer, appointed
- 0:22interim CEO.
- 0:24The Chief Financial Officer is the one
- 0:26who runs the books. She controls how
- 0:28much gets invested, what every dollar is
- 0:30spent on, and how much the company makes
- 0:33on each sale.
- 0:35And interim [music]
- 0:36is because Jensen Huang, the founder who
- 0:38ran NVIDIA for 33 years, left overnight,
- 0:42and the board is still looking for
- 0:43someone permanent. In the meantime,
- 0:45she's doing both jobs at once.
- 0:48NVIDIA used to be run by the engineer
- 0:50who founded it, and now it's run by the
- 0:53person who knows its numbers best.
- 0:55And that raises a huge question. Will
- 0:57she keep her foot on the gas with the
- 0:59same bet, or will she change what NVIDIA
- 1:02sells, who it sells to, and on what
- 1:04terms?
- 1:06That question breaks down into three.
- 1:08NVIDIA has already announced its next
- 1:10chips, and its customers are counting on
- 1:12them. Will it keep launching them at the
- 1:14same pace, or will a CFO prefer to slow
- 1:17down and deliver without mistakes?
- 1:20Right now, NVIDIA isn't even counting on
- 1:22selling its AI chips in China because
- 1:25that depends on what Washington allows.
- 1:27Can Kress open that door?
- 1:30And her job is interim because the board
- 1:32is looking for someone else.
- 1:34How much can someone who's just passing
- 1:35through really change?
- 1:38Those are the questions we'll answer
- 1:39throughout this video, and their answers
- 1:41shape what's at stake for NVIDIA's
- 1:43stock.
- 1:45And this affects you more than you might
- 1:46think. If you own a fund that tracks the
- 1:49US stock market, part of your money is
- 1:51in NVIDIA.
- 1:53But it's not just shareholders because
- 1:55its chips shape how much the AI you use
- 1:57every day costs and how fast it to you.
- 2:01The first clue comes on Tuesday,
- 2:03November 17th, when she reports earnings
- 2:05for the first time as the boss.
- 2:09To understand how she thinks, you first
- 2:11need to know what Nvidia sells.
- 2:14A regular computer does calculations one
- 2:16after another very fast. [music]
- 2:18Nvidia's chips do thousands at the same
- 2:20time, and that's exactly what an AI
- 2:23needs to learn and to answer you.
- 2:25Nvidia calls this accelerated computing.
- 2:30In 2024, back when she was only running
- 2:32the finances, she explained in an
- 2:34interview how she sees the future of
- 2:36this business.
- 2:38We are not going to get this all done in
- 2:39two years. These transitions are two
- 2:42decades, three decades of work to change
- 2:45to accelerating computing and to infuse
- 2:47AI in everything that we do.
- 2:50Do we know what everyday is going to be?
- 2:52Is it all going to grow at this same
- 2:54lovely speed? No. But we know it is
- 2:57going to be with us for decades.
- 3:00In other words, she's not promising that
- 3:02sales will keep skyrocketing every year.
- 3:05She's promising that this business is in
- 3:07it for the long haul with good years and
- 3:09weaker [music] years.
- 3:11And that comes from the person who now
- 3:13decides how much Nvidia invests and in
- 3:16what.
- 3:17In that same interview, she talked about
- 3:19who pays for all this. The companies
- 3:22that buy AI.
- 3:24Her answer has two parts. The first,
- 3:26making the investment now, knowing that
- 3:28it's not an instantaneous solution.
- 3:31And the second says where the money
- 3:33comes from.
- 3:34You will stop doing things that are not
- 3:36providing productivity.
- 3:38Picture an insurance company that spends
- 3:40a million dollars a year on people who
- 3:42review accident claims by hand.
- 3:44It decides to bring in an AI which costs
- 3:47$200,000 a year.
- 3:50In the first year, it doesn't save
- 3:51anything yet because people have to
- 3:53learn how to use it. So that year, it
- 3:55spends more than before.
- 3:57But in the second year, the AI handles
- 4:00half the reviews and the insurer only
- 4:02needs $500,000 for the ones still done
- 4:04by hand. Add the 200,000 for the AI, and
- 4:08it spends 700,000 in total, less than
- 4:11the million it started with.
- 4:13Its budget didn't go up. It moved money
- 4:16from manual reviews to the AI. And if
- 4:18the math hadn't worked out, it wouldn't
- 4:20pay for it again.
- 4:22And Nvidia depends on that math. Today,
- 4:25its biggest buyers are the big tech
- 4:27companies, which fill entire warehouses
- 4:30with its chips. But the way Kress sees
- 4:32it, the business only lasts for decades
- 4:34if thousands of ordinary companies, like
- 4:36that insurer, find that AI pays off for
- 4:39them.
- 4:40Will she look there for Nvidia's next
- 4:42wave of growth?
- 4:44As the boss, her first words came in the
- 4:47same announcement that named her.
- 4:49We will continue advancing the work that
- 4:51has made Nvidia a leader in accelerated
- 4:53computing and AI and execute on the
- 4:56opportunities ahead. Our focus remains
- 4:58on serving our customers, partners,
- 5:01developers, and shareholders.
- 5:03It's what anyone stepping in on an
- 5:05interim basis would say. Don't panic.
- 5:08Everything carries on.
- 5:10She doesn't announce a single change.
- 5:12But she doesn't say what those
- 5:13opportunities are or how she plans to
- 5:16seize them either.
- 5:18And the first test of that continuity is
- 5:20the chips Nvidia has already promised
- 5:22its customers. Will it keep launching a
- 5:24new one at full speed or will she focus
- 5:27on delivering on time what it's already
- 5:29sold?
- 5:31What Nvidia sells today isn't individual
- 5:33chips. It's machines like this one
- 5:35packed with its Blackwell chips, which
- 5:37big tech companies install in rows in
- 5:39their data centers, warehouses full of
- 5:42computers where AI runs.
- 5:45Until 2024, Nvidia released a new
- 5:48generation of chips every two years.
- 5:51That year, Jensen Huang announced that
- 5:53from then on it would release one every
- 5:55year.
- 5:56For its customers, it's a race. Each
- 5:59generation does more work than the last,
- 6:01and whoever doesn't keep up falls behind
- 6:03the competition. And for Nvidia, every
- 6:06new generation is another round of
- 6:08orders.
- 6:10But moving that fast comes at a price.
- 6:12In 2024, Blackwell came out with a
- 6:15design flaw. The chips worked, but each
- 6:18batch that was made produced very few
- 6:20good ones. Nvidia had to redo part of
- 6:22it, and deliveries to customers like
- 6:24Meta, Google, and Microsoft were delayed
- 6:27by months.
- 6:28Huang himself admitted it. It was 100%
- 6:32Nvidia's fault.
- 6:34Now it's the next generation's turn.
- 6:36Vera Rubin, which has already started
- 6:38being assembled and reaching its first
- 6:40customers. Amazon, for example, is
- 6:42already counting on it for its data
- 6:44centers. This is what Crest inherits,
- 6:47chips that are already in motion and
- 6:49that customers expect to receive on
- 6:51time.
- 6:52And in that 2024 interview, she had
- 6:54already explained how she allocates
- 6:56Nvidia's money,
- 6:58planning our capital to design more
- 7:00products and build more products with
- 7:02one condition. We do not waste a day. We
- 7:05don't waste a dime.
- 7:08She wants both to move fast and not
- 7:11waste money.
- 7:13And a delay like Blackwell's means
- 7:15losing both at once, days and dollars.
- 7:18Nothing she has said points to slowing
- 7:20down the pace of a new chip every year.
- 7:23What it does point to is watching
- 7:25closely that every chip arrives when it
- 7:27was promised.
- 7:28And for anyone building AI, that matters
- 7:31a lot because a faster chip that shows
- 7:33up late leaves their data center half
- 7:35finished and their new AI models
- 7:37waiting.
- 7:38Also, Crest has less say here than it
- 7:41might seem. The chips are designed by
- 7:43the same engineering team that worked
- 7:45with Huang, and the board has already
- 7:46said the product roadmap will continue
- 7:48without interruption.
- 7:51But chips are only part of what a
- 7:52company needs to use AI.
- 7:55Go back to that insurer. It buys the
- 7:57chips, and on their own, they do
- 7:58nothing. It needs the software that puts
- 8:01them to work, someone to update it when
- 8:03a new version comes out, and someone to
- 8:05secure it so nobody steals its
- 8:07customers' data.
- 8:09If it doesn't have engineers for that,
- 8:11it has to pay somebody.
- 8:13Kress said in that interview that most
- 8:15companies don't write their own software
- 8:17and that the somebody who takes care
- 8:19[music] of it is Nvidia.
- 8:20We have to keep it secure. We have to
- 8:23keep it continually upgraded.
- 8:25And Nvidia already has that storefront
- 8:28up and running. On its website, it
- 8:30offers more than 100 AI models ready to
- 8:32run on its chips, and many of them can
- 8:34be tried for free.
- 8:36But when a company uses them for real,
- 8:39it starts paying.
- 8:40On Nvidia's price list, a one-year
- 8:42subscription to its software with
- 8:44support included costs [music] $4,500
- 8:48per chip.
- 8:49If the insurer has 1,000 chips, that's
- 8:52$4.5 million a year, and it pays it
- 8:55again every year for as long as it uses
- 8:57them.
- 8:59And that's the difference a CFO cares
- 9:01about most. A chip is sold once. If the
- 9:04customer doesn't buy again the next
- 9:06year, that money doesn't come back.
- 9:09The subscription, on the other hand,
- 9:11comes in every year.
- 9:13If Kress put more weight on software,
- 9:15Nvidia would have steadier revenue even
- 9:17in a year when it sold fewer chips.
- 9:20She hasn't announced anything like this
- 9:21so far, but it's what she argued for
- 9:24when she was only running the finances.
- 9:27For companies, it cuts both ways. Paying
- 9:30Nvidia for the software saves them from
- 9:32building their own team of engineers,
- 9:34and for many, it's the fastest way to
- 9:36get started with AI.
- 9:38But the longer they use it, the more
- 9:40they depend on Nvidia.
- 9:42If one day they want to switch to
- 9:43another maker's chips, they have to
- 9:45rebuild everything they built on its
- 9:47software.
- 9:48But before it can charge for software,
- 9:50it has to sell the chips. And some of
- 9:53Nvidia's biggest buyers can't build
- 9:55their data centers on their own. Nvidia
- 9:57gives them a hand. [music]
- 9:59Will it keep doing that under Kress, or
- 10:01will a CFO think twice?
- 10:04An AI data center costs so much that
- 10:07even Nvidia's biggest customers look for
- 10:09someone to back them.
- 10:10One of them is OpenAI, the company
- 10:13behind ChatGPT.
- 10:15To build its facilities, it has to lease
- 10:17the land, secure the electricity, and
- 10:19put up the building. And whoever leases
- 10:21all that to it wants to be sure they'll
- 10:23get paid.
- 10:25And that's where Nvidia comes in. Its
- 10:27summer financial report puts it like
- 10:29this.
- 10:30In August 2026, we entered into
- 10:32guarantees capped at a total of $105
- 10:36billion.
- 10:37And it spells out when it would have to
- 10:39pay.
- 10:40Our payment obligations under the
- 10:41guarantees are triggered upon certain
- 10:43tenant defaults.
- 10:46It's like parents co-signing their kids'
- 10:48lease. If the kid pays every month, the
- 10:50parents don't put in a cent. If the kid
- 10:52stops paying, the landlord goes after
- 10:55them.
- 10:56Here, the kid is OpenAI. The landlord is
- 10:59the company building its data center,
- 11:01and the parents are Nvidia.
- 11:03So what does Nvidia get for taking on
- 11:06that risk?
- 11:07In return, that data center will be
- 11:09filled almost exclusively with Nvidia
- 11:11equipment. In other words, it co-signs
- 11:14the lease to make sure those facilities
- 11:16fill up with its chips.
- 11:18If OpenAI pays, it's all sales. If one
- 11:21day it can't, Nvidia has to put up
- 11:23money, though only part of the rent, not
- 11:26the whole construction.
- 11:28Inside Nvidia, not everyone saw these
- 11:30deals the same way. According to
- 11:32Reuters, citing a person familiar with
- 11:34the matter, Cress and her staff clashed
- 11:37with Huang and the company's sales team
- 11:39over the massive and risky financing
- 11:41deals.
- 11:42In other words, the person now in charge
- 11:44was the one pushing back on these deals.
- 11:47She hasn't announced any changes so far,
- 11:49but she has two options.
- 11:52If she toughens them, for example, by
- 11:54guaranteeing less or asking for more in
- 11:56return, Nvidia risks less money if a
- 11:59customer fails.
- 12:01But some customers would have a harder
- 12:03time building their data centers. Those
- 12:05AI projects would move more slowly, and
- 12:07so would their chip orders.
- 12:10If she keeps them, the facilities go up
- 12:12sooner and fill up with Nvidia chips,
- 12:14but the risk stays on its books.
- 12:18There's another place where Nvidia could
- 12:19sell a huge number of chips, China.
- 12:23But there, Nvidia doesn't decide,
- 12:25Washington does.
- 12:27The US government worries that the most
- 12:28powerful chips could help China
- 12:30militarily, so it decides which ones can
- 12:33be sold there and to whom.
- 12:36In January, the Commerce Department
- 12:38opened the door a crack for one of
- 12:40Nvidia's chips. It will review each
- 12:42license case by case as long as certain
- 12:45security requirements are met.
- 12:48In other words, every sale needs a
- 12:50permit and the Chinese buyer has to
- 12:52prove who it's passing the chips onto.
- 12:55Even so, in August, Kress herself made
- 12:58it clear how much Nvidia trusts that
- 13:00door.
- 13:01There is no China data center compute
- 13:04revenue in our forward outlook.
- 13:06In other words, in the forecasts Nvidia
- 13:08makes for the coming months, whatever it
- 13:10sells in China in AI chips counts as
- 13:13zero.
- 13:14And that has a curious upside for
- 13:16shareholders. Since Nvidia is no longer
- 13:19counting on China, if the door stays
- 13:21shut, it loses nothing it was expecting.
- 13:24But if Kress manages to open it,
- 13:26everything it sells there is a bonus
- 13:28that isn't in its forecast today.
- 13:31The problem is the long term. While
- 13:33Nvidia isn't selling, Chinese customers
- 13:36buy chips from homegrown makers like
- 13:38Huawei, and every year that goes by,
- 13:40those makers get better and win more
- 13:42customers.
- 13:44So does Kress have anyone to negotiate
- 13:47with? Until now, the one who went to the
- 13:49White House to make Nvidia's case was
- 13:51Huang, and she has no known political
- 13:54connections of her own.
- 13:56What she does inherit is a tool. In
- 13:58August, Nvidia registered the Nvidia
- 14:01Corporation Employees Federal Political
- 14:03Action Committee in Washington, and the
- 14:05official record reads connected
- 14:07organization, Nvidia Corporation.
- 14:11A committee like this pools money from a
- 14:13company's employees to donate to
- 14:15politicians campaigns. Lots of big
- 14:17companies have one so that Washington
- 14:19listens when they need something.
- 14:22But it belongs to Nvidia, not to her.
- 14:24That kind of relationship with
- 14:26politicians is something Kress will have
- 14:28to earn on her own.
- 14:30And while China stays half open, other
- 14:32countries are knocking on Nvidia's door
- 14:34because they want their own AI.
- 14:37Could that be where it finds the growth
- 14:38China won't give it?
- 14:40Today, when a company uses AI, it
- 14:43usually rents computers over the
- 14:45internet from Amazon, Microsoft, or
- 14:47Google in facilities that may be in
- 14:49another country. That's what we call the
- 14:52cloud.
- 14:53And more and more governments don't want
- 14:55their AI and their citizens' data to
- 14:58depend on a handful of American
- 14:59companies. They want data centers on
- 15:01their own soil, run by companies from
- 15:04their own country.
- 15:06Kress laid it out in August when she
- 15:08presented earnings.
- 15:10A country or region can allocate land
- 15:12and power directly to a regional cloud
- 15:14partner in ways it never would to a
- 15:16foreign hyperscaler.
- 15:19Hyperscaler is what the industry calls
- 15:21giants like Amazon or Microsoft. What
- 15:24she's saying is that a government will
- 15:26give a local company the land and
- 15:28electricity a data center needs,
- 15:30something it wouldn't give an American
- 15:32one.
- 15:33And who does that local company buy its
- 15:35chips from? Nvidia.
- 15:38So every country that joins in is a new
- 15:40buyer, and Nvidia depends less on the
- 15:42big tech companies that are its biggest
- 15:44buyers today.
- 15:46South Korea is one of those countries.
- 15:49There, Nvidia has teamed up with the
- 15:51country's big phone carrier, SK Telecom,
- 15:54and the Korean company Rebellions as
- 15:56part of Korea's plan to have its own AI.
- 16:00Will this be Kress's big bet? She hasn't
- 16:03said anything new since taking over, but
- 16:05it's exactly the path she herself laid
- 16:07out in August.
- 16:09And it changes who gets to build AI, not
- 16:12just a few American companies anymore,
- 16:14but also a company in India or Malaysia
- 16:17backed by its government.
- 16:20But there's a limit, and Nvidia admits
- 16:22it in its own filings.
- 16:24Finally, our business depends on our
- 16:26ability to receive consistent and
- 16:28reliable supply from our overseas
- 16:30partners, especially in Taiwan and South
- 16:33Korea.
- 16:35Nvidia designs the chips, but it doesn't
- 16:37make them. They're made in factories in
- 16:39Taiwan, and key parts like memory come
- 16:41from South Korea.
- 16:43So no matter how many countries want
- 16:45chips, Nvidia can only sell as many as
- 16:48those factories can make. And if
- 16:50something cuts off that supply, like a
- 16:52conflict in the region, sales slow down
- 16:54for everyone, countries and big tech
- 16:57alike.
- 16:58With so many fronts open, we're back to
- 17:00the question from the start.
- 17:02How much can someone who's just passing
- 17:04through really change?
- 17:07Being interim doesn't mean she can't
- 17:09call the shots.
- 17:11Nvidia's bylaws, the internal rules that
- 17:13say who's in charge of what, allow her
- 17:16to hold two jobs at once to begin with.
- 17:19Any one person may hold any number of
- 17:21offices.
- 17:22And on the CEO's powers, the text ends
- 17:24like this.
- 17:26Shall also perform such other duties and
- 17:28have such other powers as the board of
- 17:30directors shall designate from time to
- 17:32time.
- 17:34In other words, how much Cress can
- 17:35decide is up to the board, the group
- 17:37that represents the shareholders, which
- 17:39just named a chair for the first time in
- 17:41its history.
- 17:43And Nvidia hasn't disclosed what
- 17:45specific powers it has given her.
- 17:48On top of that, there's a rule that
- 17:49weighs on every deal she inherits.
- 17:52Unless authorized or ratified by the
- 17:54board of directors or within the agency
- 17:57power of an officer, no officer, agent,
- 18:00or employee shall have any power or
- 18:02authority to bind the corporation by any
- 18:04contract or engagement.
- 18:07So Nvidia's big commitments, like the
- 18:09guarantee for OpenAI, are signed with
- 18:12the board's approval, and Cress needs it
- 18:14to change them too.
- 18:16And a signed contract doesn't get torn
- 18:18up just because the boss changes. The
- 18:20other side has to agree as well.
- 18:23Plus, the board is searching for her
- 18:24replacement inside or outside the
- 18:27company. If it's not her, any big change
- 18:30she makes may last only as long as she
- 18:32stays in the job.
- 18:34That's why in these months the easiest
- 18:36thing for her is to adjust what she
- 18:37knows best, the money and the terms of
- 18:40new deals rather than the whole
- 18:41direction.
- 18:43But even just by adjusting the money,
- 18:45she can change a lot about what Nvidia
- 18:47earns, which direction suits it best,
- 18:50and what's at stake for its stock with
- 18:52each one.
- 18:54To know what's at stake for the stock,
- 18:55you first need to understand what you're
- 18:57paying for when you buy a share.
- 19:00Imagine a bakery that makes $10,000 a
- 19:02year, and every year it makes 30% more.
- 19:06Someone is willing to pay you $300,000
- 19:09for it because in a few years it'll be
- 19:11making triple.
- 19:12But one day it becomes clear that from
- 19:14now on it will only grow 5% a year. It
- 19:17still makes $10,000, but nobody will
- 19:20give you 300,000 anymore. At most, what
- 19:23people pay for an ordinary bakery, say
- 19:25150,000.
- 19:28It hasn't lost a single customer, and
- 19:30yet it's worth half.
- 19:33It's the same with Nvidia. Its stock
- 19:35price already assumes it will keep
- 19:37growing very fast. So what moves its
- 19:39stock isn't just how much it earns, but
- 19:42whether the path Crest chooses makes
- 19:44people believe it'll keep growing like
- 19:45that.
- 19:47The first path is to stay the course and
- 19:49deliver. Get Vera Rubin out on time and
- 19:52deliver what was promised. That's what
- 19:54the market already expects, so there are
- 19:56no surprises, good or bad.
- 19:59But she herself said the business won't
- 20:01always grow at this speed, and the day
- 20:03that shows the stock will behave like
- 20:05the bakery.
- 20:07The second is to charge more for
- 20:09software and sell to the countries that
- 20:11want their own AI. If it works, Nvidia
- 20:14earns steadier money and depends less on
- 20:16a few tech giants, and that makes people
- 20:18trust that the growth will last.
- 20:20The risk is that it takes years to show
- 20:23up, and the market doesn't always have
- 20:24that kind of patience.
- 20:27The third is to get tougher with big
- 20:29customers and guarantee less. And Nvidia
- 20:32already admits in its filings what could
- 20:33happen. Customers may postpone
- 20:36purchasing new architectures, and one of
- 20:38the reasons it gives is constraints on
- 20:40capital to have sufficient funding to
- 20:42purchase our products.
- 20:44In other words, if a customer is short
- 20:46on money, it doesn't buy the new
- 20:48generation of chips. If Kress stops
- 20:50giving them a hand, Nvidia sells less in
- 20:53the short term.
- 20:54But if the AI frenzy cools off and a
- 20:56customer can't pay, Nvidia will have
- 20:59risked much less. It earns less today in
- 21:01exchange for fewer scares tomorrow.
- 21:05And hanging over all three paths is
- 21:07confidence. One analyst summed it up to
- 21:09Reuters like this. Putting Kress in the
- 21:12interim role makes sense to keep the
- 21:13markets calm, but I don't think she is
- 21:15the long-term visionary that Nvidia
- 21:17needs to continue its current
- 21:19trajectory.
- 21:20That's an opinion, not a fact, but it
- 21:23explains what's at stake for the stock.
- 21:25As long as Kress is interim, the market
- 21:27will also be watching who comes next.
- 21:31The first test already has a date.
- 21:33Nvidia will host its earnings call on
- 21:35Tuesday, November 17th, and ahead of it,
- 21:38it will provide written commentary on
- 21:40its third quarter results from Colette
- 21:42Kress.
- 21:44That's where we'll see which way she
- 21:45leans.
- 21:47If she confirms Vera Rubin is on
- 21:49schedule and little else, she's taking
- 21:51the first path. If she talks more about
- 21:53software and countries, she's leaning
- 21:55toward the second. And if she changes
- 21:57the terms of the guarantees for
- 21:59customers like OpenAI, she's taking the
- 22:01third.
- 22:03Nvidia has a new boss, but so far, not a
- 22:06new direction.
- 22:07That direction won't come from a press
- 22:09release, but from the numbers, what it
- 22:11spends on, who it sells to, and how much
- 22:14risk is taken by the person who knows
- 22:16its numbers best.
About this transcript
This page contains the full transcript of Colette Kress takes charge of Nvidia: how its AI bet could change by Finance for everyone, generated from the public captions YouTube serves with the video. The transcript has 3,683 words across 606 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.