YT YouTube2Text

YouTube2Text → Transcripts → The Satisfying Downfall of Perplexity

The Satisfying Downfall of Perplexity — Transcript

by Siddharth Chopra · 4,761 words · 731 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Something is happening with Perplexity
0:02that seems almost impossible in the
0:04business world. Its market share is
0:06collapsing, yet its revenue is
0:08exploding here. And somehow, both of
0:10these things are happening at the same
0:12time. Let me explain. A few years ago,
0:14Perplexity was being called the future
0:16of search. The space Perplexity funding
0:31, company 18 billion, its latest signal
0:33focus in shifting beyond powerful
0:35models and toward user applications;
0:38some experts genuinely believed it
0:40could become the next Google killer,
0:42and this wasn't just empty hype. Within
0:46approximately four months of its launch
0:48, Perplexity had crossed 2 million
0:50users. By May 2025, Perplexity was
0:53processing approximately 780 million
0:55queries every month. Jeff Bezos had
0:59invested in it, and in fact, giants
1:01like Nvidia and SoftBank were also
1:02backing it, and its valuation had
1:04reportedly reached 20 billion. It's an
1:08AI search engine, starting to look to
1:10take on Google. 74 million Perplexity
1:25Opening search, now at one point,
1:27Perplexity was so confident that it
1:30made an offer of 34.5 billion to buy
1:32Google Chrome. Now just think about it.
1:36A 3-year-old startup is talking about
1:38buying the browser of the company it
1:40set out to disrupt. Insane, right? But
1:44then the entire game changes here.
1:46ChatGPT had introduced live search.
1:50Google had launched AI modes, and the
1:52same experience that made Perplexity
1:54special became available in almost
1:57every AI chatbot. To retain its users,
2:00Perplexity has tried everything here.
2:03It launched its own browser, Comet.
2:06Even in India, it partnered with Airtel
2:09and gave all Airtel users Perplexity
2:12Pro for free for one year. But you know
2:15what? Even after getting Perplexity Pro
2:17, users were not using it. So, none of
2:20these moves worked here. In fact, in
2:232025, Perplexity reportedly held
2:26approximately 6.6%of the chatbot market
2:29share. Today, this number is close to 2
2:32%. Which means that Perplexity has lost
2:3570%of its market share from its peak
2:37today. So, when Perplexity is losing
2:39its market share here. Then its
2:41business should also collapse here.
2:43Right? Well, the exact opposite has
2:45happened. In December 2024,
2:47Perplexity's reported annual recurring
2:50revenue was approximately 63 million.
2:53By March 2026, it had grown to over $
2:56450 million. But by April, it had
2:59reportedly reached $ 500 million. So,
3:01Perplexity is losing market share here.
3:04Yet, it is now making more money than
3:06ever before. But why? So in this video,
3:09we will decode how the company once
3:11called a "Google Killer" lost its
3:13biggest advantage. If users are
3:16increasingly choosing ChatGPT and
3:18Google, then who is actually paying
3:20Perplexity and exactly what are they
3:22paying for? And most importantly, what
3:26exactly did Perplexity do that is
3:28causing its revenue to keep growing,
3:30and how long will that strategy work?
3:52To understand Perplexity's situation,
3:54we have to go back to 2022. That was
3:57the time when ChatGPT launched, and it
4:00felt like magic. You could ask it
4:02anything, and instead of giving you 10
4:04links, it would give you a direct
4:06answer. But there was a very major
4:08problem here. ChatGPT did not have
4:10access to the internet. Its knowledge
4:13stopped at a particular date. So it
4:15could tell you the history of Apple.
4:18But it doesn't know what Apple
4:20announced today. Now, at the same time,
4:23Google had the exact opposite problem.
4:25It had the live internet. It had
4:28indexed billions of web pages and could
4:31find everything going on in the world.
4:34But the challenge here was that the
4:36user had to do most of the work. You
4:39searched for something. Google gave you
4:41a link. You opened the link. You
4:43avoided the ads. You rejected the
4:45cookie popups and combined all the
4:47information together. So there was a
4:50clear gap here. ChatGPT could explain
4:52things. But it didn't have internet
4:55access. Meanwhile, Google could search
4:58for everything via the internet, but it
4:59couldn't answer things in a
5:01conversational way. Now, Perplexity
5:04understood this exact gap and launched
5:06its product, which was quite genius but
5:09also very simple. Let me explain: in
5:13its early days, Perplexity combined
5:15OpenAI's GPT-3.5 model with Bing, which
5:18is Microsoft's search engine. So,
5:21whenever a user asks a query, for
5:23example, "What did Apple launch today?"
5:26Perplexity retrieves relevant web
5:28results from Bing. Then it passes those
5:31results into the GPT-3.5 context and
5:33generates a direct answer with proper
5:35links and citations. So here you get
5:38live searches, direct answers, and
5:40visible sources, all within one simple
5:43interface. And yes, that is exactly
5:45what Perplexity was. Now, this product
5:47explodes. Within just four months of
5:50launch, Perplexity reached 2 million
5:52monthly active users. But according to
5:54Aravind, they were already processing
5:563,000 queries on just the first day of
5:58launch. By May 2025, this number had
6:00reached approximately 780 million
6:02queries per month. And even within just
6:063 years, its valuation reached 20
6:08billion. Not million, but billion with
6:11a B. Now, a question might come to your
6:14mind here: if Perplexity could see this
6:16gap, were OpenAI and Google so stupid
6:18that they didn't see it? Well, they
6:21definitely knew about this gap.
6:23Understand what their condition was at
6:25that time. OpenAI and Google both had
6:28different priorities. For them,
6:30building something like Perplexity
6:32wasn't that important. See, after
6:34OpenAI launched GPT-3.5, other
6:36companies also started building
6:38foundation models. So OpenAI had to
6:41upgrade its model here. It launched GPT
6:44-4 and even started rolling out image
6:46and video models so that OpenAI could
6:48stay ahead of everyone else in this
6:51game. Now let's look at Google's
6:53situation. When ChatGPT was launched,
6:56Google announced a code red because its
6:58search business, from which it makes
7:01billions of dollars, was at risk. So
7:04its first priority was to build Large
7:06Language Models (LLMs) so that it could
7:09compete with ChatGPT here. And to be
7:12honest, Google's start was not good at
7:14all. It launched Bard, and it was
7:16terrible. It failed very badly. After
7:20this, it was even planning a comeback
7:22with Gemini. So AI search was an
7:24opportunity. But both companies were
7:26fighting much bigger battles at that
7:28time. So for them, a bet like
7:30Perplexity was very small. But then the
7:33situation changes. Now, understand this
7:40carefully. AI is evolving very rapidly.
7:43So, it is not just for companies, but
7:45for us too, it is essential to stay
7:47updated with this technology to remain
7:49future-ready. Now, it's not just
7:51important to learn tools, but it's also
7:53crucial to understand the core concepts
7:56of AI. But if you are also confused
7:58about where to learn this from, don't
8:00worry, I have a very amazing course for
8:02you. AI for Everyone on Coursera. This
8:05is a beginner-friendly course that
8:07anyone can take. In this, everything
8:09from basic AI concepts to real-world
8:11applications is explained in simple
8:13language. The instructor for this
8:16course is Andrew Ng, who has been a
8:17co-founder of Google Brain. And he is
8:19also the founder of Coursera and
8:21DeepLearning.AI. And if you
8:23specifically want to learn Generative
8:25AI, you can also check out his
8:27Generative AI for Everyone course.
8:29Coursera has sponsored this video, so
8:31there is a special offer for all of you
8:33. You don't need to buy every
8:35individual course. Coursera offers a
8:38Plus subscription. In which you will
8:40get annual access to all courses for
8:42only around ₹ 7,500 per year. Now, I
8:45have added the discount links for both
8:47these courses and the Plus subscription
8:49in the description below. So don't
8:50forget to check that out. Now let's get
8:52back to the topic. Perplexity's timing
8:55was extremely good. That is why it saw
8:58such massive growth. But this same
9:00growth also proved to be bad for it.
9:02Because now Perplexity had proved that
9:05AI search is a very big market. And as
9:08soon as OpenAI and Google came out of
9:10their survival mode, live search became
9:13a priority for them. OpenAI launches
9:16ChatGPT Search in October 2024.
9:19Meanwhile, Google introduces AI
9:22Overviews and also launches them
9:24broadly in May 2025. So the product
9:27that made Perplexity special is now
9:30slowly becoming a standard feature in
9:32every AI product, and with this,
9:34Perplexity's market share begins to
9:36collapse. Now, understand this: there
9:38are two things here in total.
9:40Foundation models and search engines.
9:42Now every company, OpenAI, Claude,
9:44Gemini, all of them had their own
9:45foundation models. Even Perplexity had
9:48created its own foundation model. And
9:51in fact, Perplexity had also added the
9:52top models from all other companies.
9:55Now, the important thing here is that
9:57everyone had a foundational model, and
9:58a separate race was going on for
10:00foundational models. But as for search
10:02engines, Google has been the king there
10:04for many years. Now after this, all
10:06companies are also launching deep
10:08research where the model can go through
10:10many websites. After that, it writes
10:12and provides a detailed answer there.
10:14Now, Google also plays a dirty tactic
10:17here. Google knew that it is the best
10:19search engine and other companies also
10:21use Google's search engine. So Google
10:24makes a small change inside its code.
10:27Before September 2025, Google used to
10:30support an optional parameter num = 100
10:33. In this, third-party tools could
10:36fetch 100 Google search results at once
10:38. But after September 2025, Google puts
10:41a cap at a maximum of 10. So now only
10:4410 results can be fetched at one time.
10:47If someone wants to fetch 100 results
10:49here, they will have to run the query
10:5110 times. Which will increase their
10:53costs here. Now, this won't make any
10:55difference to any normal user. But SEO
10:58tools, scrapers, and AI systems that
11:01gather results at scale will now have
11:03to pay more in steps, time, and even
11:06cost. Now, this was just a very small
11:08thing. Google's distribution is many
11:10times more powerful than this. Look at
11:13this. Microsoft spent billions of
11:15dollars to build Bing. It owns Windows.
11:19It owns the Edge browser. It had
11:21heavily invested in OpenAI. And
11:23integrated many generative features
11:25into its search engine. Even after
11:28doing so many things, Bing is not even
11:30close to Google Search today. Even
11:32Apple itself does not fight a search
11:34war with Google. In fact, it formed a
11:36partnership with Google. Many court
11:39documents reveal that Google paid Apple
11:4220 billion in 2022 alone. So that
11:45Google Search remains the default
11:47search engine on all Apple devices. Now
11:49just think about it. The world's most
11:52powerful company earns billions of
11:54dollars to keep Google in front of its
11:57users and not try to replace it, and
11:59that is because in search, the best
12:01product doesn't win, but the product
12:03that is already in front of users wins,
12:05and today Google is everywhere. Chrome
12:09is the world's largest browser today.
12:11Android is the biggest operating system
12:13. Google is the default search engine
12:16in Safari. In fact, "Google it" has
12:18even become a verb. So, for Perplexity
12:21to match this level of distribution was
12:23near impossible. But you know what?
12:26Perplexity has tried everything it
12:28possibly could here. For instance, in
12:31July 2025 in India, it partnered with
12:33Airtel to give all Airtel users one
12:36year of free Perplexity Pro. Which was
12:39worth more than ₹ 17,000. Now, this
12:42wasn't just a giveaway. This was a
12:43distribution strategy. India is one of
12:45the biggest internet markets in the
12:47world. By collaborating with Airtel,
12:50they would get access to millions of
12:53users overnight, giving Perplexity
12:55their attention here. Now, Google
12:58wouldn't let this strategy slide either
12:59. It collaborated with Jio in India and
13:03offered Google AI Pro for 18 months
13:05instead of 12, completely free. In
13:08which you get not just Gemini Pro, but
13:10also 5TB of storage and pro access to
13:13the entire Google ecosystem. After this
13:15, Perplexity tried an even bigger bet.
13:18It launched its very own AI browser.
13:21Comment. The logic behind launching
13:23this was very simple. If Perplexity
13:26controls the browser, users won't need
13:28to open Perplexity separately. It would
13:31sit right between the users and the
13:33internet. Exactly the same as it is
13:35with Google Chrome. Once you switch to
13:38Chrome, you'll use Google Search
13:39because that is the default search
13:41engine there. Now, Perplexity wasn't
13:44the only one that saw this opportunity.
13:46OpenAI at least launches its own
13:48browser. But fast-forwarding to today.
13:51Today, both of these browsers have
13:53failed miserably. Why they failed is a
13:55very interesting topic that would take
13:58too long to explain. So we will discuss
14:00that in another video. But the
14:03conclusion here is that when even
14:04Microsoft and Apple couldn't break
14:06Google's distribution, it was already
14:08impossible for Perplexity to do so. But
14:11Perplexity tried, and I really
14:13appreciate them for that. So now that
14:16all our basics are clear. Let's talk
14:18about the most important question. When
14:21Perplexity is losing market share in AI
14:23search, then how exactly is its revenue
14:26exploding? By late 2025, AI models had
14:33become incredibly intelligent. They
14:38could research any industry, analyze
14:40documents, write software, create
14:43business plans, and in fact, explain
14:45how to complete a task step-by-step;
14:48but there was one major problem here.
14:53AI can tell us how to do a task. But it
14:57cannot execute all these processes on
14:59its own. For example, you asked ChatGPT
15:03to plan your trip. Now, it will suggest
15:05destinations. It will compare hotels
15:07and create your complete itinerary. But
15:10here, you have to open all the websites
15:12yourself. Complete the bookings. Add
15:15events to the calendar and inform
15:17everyone going with you about it. You
15:20will have to do all these things
15:22yourself. So, intelligence existed at
15:24that time. But these large language
15:27models had no "hands" to turn that
15:29intelligence into real work. But then,
15:33in November 2025, a developer named
15:36Peter Steinberg launched ClaudeBot,
15:38which was later renamed to Open-Claude.
15:42Now, what actually was ClaudeBot and
15:44why was the hype so insane? To
15:46understand this, we won't use any
15:48definitions. Instead, I will show you a
15:51practical use case through which you
15:53will understand how powerful ClaudeBot
15:55really is. So, I have set up this
15:58Open-Claude here. This is the
15:59Open-Claude window. Now, I will explain
16:02the task a bit. I have created a folder
16:04named "Downloads" on my desktop. Now,
16:07there are many different types of files
16:09in here. There are music files, PDFs,
16:12and images. I mean, there is quite a
16:15diverse kind of data here. Now, I want
16:17all this data to be organized. So, I
16:19can tell it to do so. So, I have
16:25written a simple problem here. Organize
16:27all the files in the Downloads folder
16:29on the desktop by file type into new
16:31folders. Simple. Now, the demo I want
16:34to show you here is that this
16:36Open-Claude, that is, this large
16:37language model, or call it an AI model
16:39...It has access to my computer here.
16:42It can access all my files here and
16:43create new folders for them. And let's
16:46see how fast it is. Now it is running
16:48some commands here, through which it
16:49will get an idea of which files are
16:51inside the desktop's download folder.
16:53What are their names? What is their
16:55file type? And based on that, it will
16:56create a new folder here. So as it is
16:59writing here, it organized all 34 files
17:02. There were 11 PDFs, and this many
17:04images. It has saved all of these.
17:06Let's see. Now if you look, all the new
17:08folders have been created here. Let's
17:10open. All the PDFs have arrived here
17:12inside the PDF folder. All documents
17:14have arrived inside the documents
17:15folder. All images have arrived inside
17:17the images folder. So, in this way, it
17:19has saved all the data here. So this
17:21was just one example. Now I will show
17:23you another use case for this. For
17:24example, it now has access to my
17:26browser as well. So I can tell it,
17:28let's say, open YouTube. Now, usually,
17:31if you tell ChatGPT directly, it cannot
17:33open YouTube. But now it has access to
17:36my browser here. So now it can open
17:38YouTube in a new tab here. So now, as
17:40you can see here, it has opened YouTube
17:43here. Now, what is the actual use case
17:45for this? I'll tell you that too. Let's
17:47say I want to go to Thailand next week.
17:51So I want to, now I want it to find the
17:55cheapest flight for me here. Find me
17:59the cheapest flight. So I have given it
18:02the task here that I want to go to
18:04Thailand on September 1st. I want the
18:06cheapest flight. Now I have told it one
18:07thing, that I am currently in Delhi.
18:09But I can also travel to nearby
18:11airports. If the flight is cheaper from
18:14there, and I am telling it to use
18:15Chrome. So based on this, what will
18:18happen is that it will already know it
18:20has to use Chrome. Now it might show me
18:22some things on the front end. Some
18:23things it might not show me. That is
18:24why I am entering it here. Now
18:26understand the task here, that it has
18:27to find a flight, but not just from
18:28Delhi. It has to go to different
18:30airports and compare these things. So
18:32now it will start this task. Oh, so
18:34there is some error. This is the same
18:35thing I was explaining, that whenever
18:36this is open-flow, there are many
18:37errors in it. So you might have to fix
18:39those things. But let's see if it can,
18:42okay. So it's still working. Let's see.
18:45If there is an error, it can fix it on
18:47the fly itself if I have given it full
18:49access. So, while it is executing this
18:52task. I will show you another demo here
18:54, which is from the phone. You can
18:56connect Open Cloud directly from your
18:59phone too. I have a 'Down One' folder
19:00inside my desktop as well. So I can
19:03write here on the phone, "List all
19:06files inside Down One." Now, as you can
19:15see, it has accessed all the folders
19:17here. So, through this, I just want to
19:19show you one thing: you can access your
19:21entire laptop from your phone. It just
19:23needs to be turned on. Besides this,
19:25now I am anywhere in the world. If my
19:26laptop is on and I have my phone with
19:28me. I can access my entire laptop right
19:30here through my phone. And I can
19:32execute all tasks through it here. Now,
19:35one use case for this is that I might
19:37need a PDF. So I will just tell it that
19:39I need this specific PDF. Just give me
19:40the file here. And it will directly
19:42send this file to me inside the phone
19:44here. So now, let's come back to this
19:46travel task. So now you will see here
19:48that the task is running. It has also
19:51opened a website for me. I think this
19:54is probably the cheapest option it has
19:56found for me. Let's see. Okay, yeah, so
19:58it has given the result here. Open the
20:00best result in Chrome, cheapest found.
20:02It has told me the price, that I am
20:04getting it at this price here. Which is
20:06available for 14,000 rupees. So now you
20:09might say, okay, this was a small task
20:10we could have done here. Now I am
20:12telling you its actual use cases here,
20:13what you can do. Right? For example, I
20:16want to shoot a podcast and I want a
20:18list of all podcast studios near my
20:21location. How can I contact them? What
20:24are their images? And I want that
20:26instead of me going through those 50
20:28studios one by one. It should create an
20:31Excel sheet for me with everyone's
20:32phone number, where available. What is
20:35their distance? What are the reviews?
20:37And based on that, it gives me the top
20:38five, and I can directly call those top
20:40five to get the details. So, in this
20:42way, whenever your computer and browser
20:44are connected here, you can unlock
20:46infinite capabilities through this. It
20:49was the first glimpse of your personal
20:51AI employee, and its hype was
20:52absolutely insane. Just after the
20:54launch of Open Interpreter, the demand
20:56for Apple Mac Minis started to rise.
20:58People started buying it more and more.
21:00But the question now is, why would the
21:02launch of a software have anything to
21:04do with hardware sales? See, the reason
21:06for this was quite simple. Whenever an
21:09agent performs a task on your computer,
21:11you cannot work on that system normally
21:13. And the Mac Mini proved to be a game
21:16changer here because it was a small,
21:18energy-efficient, and powerful machine,
21:21which is why developers started buying
21:23dedicated Mac Minis to run Open
21:25Interpreter continuously. Now,
21:27Perplexity observed this rise of Open
21:29Interpreter and identified two major
21:31gaps here. Number one, it was too
21:34technical. To properly set up Open
21:36Interpreter, you had to be a software
21:39engineer. Because you needed to know
21:41how to use the terminal. You needed to
21:43know how to install repositories from
21:44GitHub. You have to configure API keys
21:47here. You have to connect different
21:49tools here. You have to manage
21:51permissions. And if even one error
21:53occurred in any of these steps, you had
21:55to troubleshoot those errors and solve
21:58them yourself. Which is essentially a
22:00software engineer's job. So, for normal
22:03users, let alone using Open Interpreter
22:05, even installing it was a very
22:07difficult task. Number two, it was
22:10extremely powerful. To use Open
22:13Interpreter, you have to give it
22:14extensive access to your computer.
22:17Files, folders, and even everything
22:19from connected tools. If you connect
22:21Gmail, WhatsApp, messages, or any other
22:23tool, then Open Interpreter gets access
22:26to that as well. Naturally, most users
22:28do not want to give these AI models
22:30access to their personal and sensitive
22:32information. Because this is not just
22:34read-only information. Open Interpreter
22:37can also execute those things. So, only
22:40if you are an experienced developer
22:41should you use Open Interpreter here.
22:43Otherwise, I would suggest that you
22:45avoid it. In fact, whenever I use Open
22:48Interpreter, I make sure to provide it
22:50access to a closed environment, like a
22:53specific folder or a tab, so it doesn't
22:55hamper any of my other work. Now, Open
22:58Interpreter simultaneously proved two
23:00important things here. Number one,
23:02people desperately want AI to have
23:04computer access and execute actual work
23:07on their behalf. Number two, but most
23:10users don't want to install, maintain,
23:12and secure the system themselves. And
23:16Perplexity attacks this exact gap here.
23:19On February 25, 2026, it launched
23:21Perplexity Computer, and its initial
23:23proposition was quite simple. Users
23:26don't need to purchase a dedicated Mac
23:28mini. They didn't have to install any
23:30complex open-source system. They didn't
23:33have to manually choose a correct AI
23:34model for every step. And the best part
23:37is that Perplexity runs all its work in
23:39a managed, isolated cloud environment.
23:42So, you literally just have to open
23:44their website. No complex setup, no
23:47installation, and tasks are executed in
23:49a cloud environment instead of your
23:50personal computer. So, your personal
23:53data doesn't go to these AI models. Now
23:56, let's understand a bit about how
23:57Perplexity Computer actually works.
24:00Perplexity breaks a large task into
24:02multiple small tasks and assigns a
24:04suitable model to each one. One model
24:06might be conducting research. One is
24:08analyzing data, another is creating
24:11images, and one is assembling the final
24:13presentation. For example, I need some
24:16sponsors for this YouTube channel. So,
24:18I need to reach out to some brands here
24:21. What could I have done in the old
24:23Perplexity? I could have asked it to
24:25suggest some brands that could sponsor
24:27a tech and business channel, and it
24:29would give me some brand names. But now
24:31, watch what all I can execute here.
24:34Find creators who work in the same
24:36domain. Check their YouTube channels
24:39and identify which brands have
24:41sponsored them. After that, identify
24:44the correct person who handles
24:46sponsorships within that company. After
24:49that, for whatever brands we've found,
24:51we need to track at least five of their
24:52competitors here; in fact, we need
24:54their LinkedIn profiles and contact
24:56details too, and all this information
24:57needs to be stored in an Excel sheet.
25:01In the old Perplexity, we only got
25:03answers, but now I get a complete sheet
25:05here that includes not just brand names
25:07, but the actual people, their LinkedIn
25:09, and even their email IDs, through
25:11which I can easily reach out to them.
25:15And in fact, if I have connected my
25:17LinkedIn, Gmail, and even WhatsApp here
25:19, it will send a personalized message
25:21to all of them. Insane, right? But now
25:24the question is, why is this Perplexity
25:26Computer having such a massive impact
25:28on Perplexity's business? Well,
25:31Perplexity Computer changes the entire
25:34equation here. Now, there are two
25:36important things here. Number one: from
25:39flat pricing to usage-based pricing.
25:42Earlier, in Perplexity search, a user
25:43could ask questions. Perplexity would
25:46search the internet. It would call an
25:48AI model and generate an answer. So
25:50users got a few requests for free.
25:53Meanwhile, in the $ 20 Pro plan, you
25:55got approximately 500 requests per day,
25:58and there was also a $ 200 higher plan
26:01that gave you even higher usage limits.
26:04Now just think, how many people
26:06actually need 500 queries a day? Very,
26:09very few. Right? In fact, I myself used
26:11to run approximately 100 to 200 queries
26:13. Even though my job is to stay updated
26:16with the news. And constantly do
26:18research. So for most users, the $ 20
26:20per month plan was more than enough.
26:23But Perplexity Computer works like a
26:26digital laborer. You can continuously
26:28assign it tasks. And here, the pricing
26:31is usage-based. The more complex your
26:33task, the more credits it will consume.
26:36So you haven't just taken the basic $
26:3920 or $ 200 plan. But if your usage
26:42exceeds the included limit, you will
26:44purchase additional credits here.
26:47Because by spending a few dollars, you
26:49can easily save 5 to 10 hours of your
26:52time. So even though there are fewer
26:55casual users here, there are many power
26:57users who want to save their time, and
26:59that is why they keep purchasing more
27:01computers and credits, which is causing
27:04Perplexity's revenue to keep growing.
27:07Number two: From consumers to
27:09enterprises. Perplexity has started
27:12integrating the computer with tools
27:14that businesses already use. It has
27:17been connected with Microsoft Teams,
27:19Word, Excel, PowerPoint, Outlook,
27:20Snowflake, Databricks, and even the
27:22internal knowledge systems of those
27:24companies. Along with this, Perplexity
27:27has also added specialized financial
27:28data from providers like Morningstar,
27:30PitchBook, and DealRoom. Now,
27:32understand the difference between
27:34consumers and enterprises here. A
27:36normal consumer might just purchase a $
27:3820 subscription. But a company
27:41purchases seats for hundreds of
27:42employees. On top of that, they pay
27:45separately for additional usage,
27:47security controls, specialized data,
27:49and custom integrations. For example,
27:52imagine a company purchases 100
27:54enterprise seats at approximately $ 40
27:56per user per month. That single company
28:00alone can generate approximately $
28:0248,000 per year for Perplexity. And we
28:05haven't even included the additional
28:07usage charges in this. To generate the
28:10same subscription revenue, Perplexity
28:12would need approximately 200 individual
28:14Pro subscribers who have taken a yearly
28:16plan. While free users don't generate
28:20direct subscription revenue at all. And
28:22this is the reason why, even though
28:24Perplexity's consumer market share is
28:26falling, its enterprise market share is
28:28growing, and that is why its revenue is
28:30moving in the opposite direction. Now,
28:33Perplexity is certainly a first mover
28:35in the enterprise AI computer race. But
28:38its advantage here is not permanent.
28:40Large players like OpenAI, Anthropic,
28:42xAI, and Google have also entered this
28:45race. For example, xAI recently
28:47launched Grokbot, which works much like
28:50the Perplexity computer. At the same
28:52time, Claude and OpenAI's current
28:54approach is a bit different. Claude
28:56launched "Computer Use," and OpenAI has
28:58also launched ChatGPT Work. Now, both
29:01of these are definitely different from
29:03the Perplexity computer because tasks
29:05are executed on your computer, not on
29:07the cloud's isolated computer. So you
29:09can call it the elder brother of the
29:10Perplexity computer. There is simply no
29:12need for installation or setup here.
29:14And here, you get the trust of
29:16Anthropic and OpenAI. So let's see
29:18whose approach makes sense in the long
29:21term. And yes, don't forget to check
29:23out Coursera. The link is in the
29:24description below. See you in another
29:26video. Bye-bye.

About this transcript

This page contains the full transcript of The Satisfying Downfall of Perplexity by Siddharth Chopra, generated from the public captions YouTube serves with the video. The transcript has 4,761 words across 731 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 translate the transcript into 100+ languages.

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, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.