Latest / Connecting the Dots / OpenAI's IPO Woes, Altman's Trust, and AI Rivalry Dynamics

Transcript

OpenAI's IPO Woes, Altman's Trust, and AI Rivalry Dynamics · Connecting the Dots

• 0:00
  Today's tech news moves fast, but understanding
• 0:03
  it shouldn't be hard. That's why Snarfl Solutions
• 0:06
  Group brings you Connecting the Dots, where we
• 0:09
  break down what's happening in tech and why it
• 0:11
  matters to you. Now, let's dive in with your
• 0:14
  hosts. Imagine building a company that, you know,
• 0:18
  confidently plans to burn through... What was
• 0:20
  it? Two hundred billion dollars in cash. Two
• 0:22
  hundred billion. Right. Two hundred billion dollars
• 0:24
  before it ever makes a single cent of profit.
• 0:27
  It sounds like a massive scam. Exactly. Or like
• 0:30
  a government space program or something. But
• 0:32
  right now, that is the actual business plan for
• 0:35
  the most important technology company on earth.
• 0:38
  Welcome in. I'm Alex and I'm Morgan. And we are
• 0:41
  the Snarfle Solutions Group coming to you from
• 0:43
  Sacramento, California. Today is April 6th, 2026.
• 0:48
  Let's get you grounded before we. Unpacked to
• 0:51
  these stories. Yeah, it's a beautiful sunny 80
• 0:53
  degrees here in Sacramento today. Perfect weather
• 0:55
  and Out in the markets. We're seeing some slight
• 0:58
  bumps the Dow closed up about 35 points S &P
• 1:01
  500 is up 70 points and Bitcoin man Bitcoin has
• 1:05
  climbed to over sixty nine thousand three hundred
• 1:07
  dollars just keeps climbing although locally
• 1:11
  Tough break in sports. The Kings took a pretty
• 1:13
  heavy loss against the Clippers last night. Yeah
• 1:15
  109 to 138 Ouch. Rough one. But hey, we have
• 1:19
  a massive stack of sources to get through today.
• 1:21
  We really do. We're pulling from like the latest
• 1:24
  financial reporting in the Wall Street Journal,
• 1:26
  Bloomberg, Business Today, and Fortune. Plus
• 1:28
  some deeply researched investigative pieces from
• 1:30
  the New Yorker in Implicator .ai. Our mission
• 1:34
  today is to, you know, look past the shiny press
• 1:36
  releases and see what is actually happening in
• 1:38
  the engine room of the tech industry. Whether
• 1:41
  you're investing in tech, using AI at work, or
• 1:43
  just insanely curious about where the future
• 1:45
  is heading, you need to know what happens when
• 1:48
  infinite hype meets the cold, hard reality of
• 1:51
  public markets. Exactly. But before we get into
• 1:53
  the main stories, a quick word. Today's episode
• 1:56
  is brought to you by Strong Coffee, the premium
• 1:58
  instant coffee that tastes like a traditional
• 2:00
  brew, but is packed with 15 grams of protein,
• 2:04
  MCTs, and adaptogens like ashwagandha. Get clean,
• 2:08
  sustained energy without the jitters. Use promo
• 2:10
  code SNARFL for 20 % off at strongcoffeecompany
• 2:13
  .com. Awesome. So let's unpack this. Right. We
• 2:16
  need to start by establishing how vulnerable
• 2:19
  the broader tech ecosystem is right now before
• 2:23
  we even touch the AI giants. Yeah, there's a
• 2:25
  big theme running through all our sources today,
• 2:26
  the illusion of scale. And the maths of trust
• 2:29
  problem it creates. Like, let's look at the staggering
• 2:33
  digital heist that hit Drift Protocol on April
• 2:36
  1st. Oh, man. Drift is a decentralized exchange
• 2:40
  on the Solana blockchain. Which basically means
• 2:43
  it operates without a traditional bank. Right.
• 2:45
  Or like a CEO calling the shots. It's essentially
• 2:48
  a financial market governed entirely by automated
• 2:51
  smart contracts. Exactly. And last week they
• 2:53
  lost 270 million dollars in digital assets. 270
• 2:58
  million just gone. Just gone. But the twist here
• 3:01
  is that this wasn't some late night brute force
• 3:03
  cyber attack where a hacker like Guessed a password
• 3:07
  not at all. I mean, this was an incredibly sophisticated
• 3:09
  six -month long con six months Yeah a hacker
• 3:13
  group linked to the North Korean government spent
• 3:14
  half a year posing as a legitimate quantitative
• 3:17
  trading firm Wow, and for those unfamiliar a
• 3:20
  quantitative or quant firm uses complex mathematical
• 3:23
  models to execute rapid -fire trades, right?
• 3:26
  Right Right. And the hackers didn't just hide
• 3:28
  behind encrypted screens, they actually attended
• 3:30
  in -person events in the tech community to meet
• 3:33
  drift contributors face -to -face. Wait, really?
• 3:36
  North Korean state actors are out there shaking
• 3:38
  hands at crypto mixers? Yep. They even deposited
• 3:41
  more than a million dollars of their own assets
• 3:43
  into the platform just to build a verifiable
• 3:46
  track record. That is wild! Putting down a million
• 3:50
  dollars in collateral just to look the part?
• 3:52
  I mean... How does a physical handshake translate
• 3:55
  into a digital vault opening? It translates into
• 3:58
  automated trust. You know, by acting like a highly
• 4:01
  liquid, reliable market maker for six months,
• 4:04
  they gained elevated borrowing privileges within
• 4:07
  Drift's algorithmic lending system. So the system
• 4:09
  basically said, hey, these guys are good for
• 4:11
  it. Exactly. Drift's own post -mortem report
• 4:14
  stated the attackers were technically fluent
• 4:15
  and had perfectly fabricated professional backgrounds.
• 4:18
  Unbelievable. And once they reached that top
• 4:20
  tier trust level, they exploited a highly specific
• 4:23
  vulnerability in the margin trading systems.
• 4:26
  And margin trading is basically borrowing money
• 4:29
  against the assets you already have to make larger
• 4:31
  bets, right? You got it. The hackers manipulated
• 4:34
  the pricing oracle, which is the code that tells
• 4:38
  the platform what an asset is currently worth,
• 4:41
  and tricked the system into thinking their collateral
• 4:43
  was infinitely valuable. Oh, Michael. And then
• 4:46
  they drained 270 million dollars in less than
• 4:49
  60 seconds. Less than a minute. That is incredible.
• 4:52
  But, you know, it points to a much bigger issue.
• 4:54
  If a hostile government can fake an entire financial
• 4:57
  institution with a few pinstripe suits and some
• 5:00
  seed money, what happens when you introduce artificial
• 5:03
  intelligence into the mix? Right, because if
• 5:06
  startups are using AI to fake scale at the bottom
• 5:08
  of the market, the results are terrifying. Let's
• 5:11
  look closer at the MedV telehealth story that
• 5:13
  just broke. Oh, this one is fascinating. So the
• 5:16
  New York Times recently published this glowing
• 5:18
  profile, holding MedV up as AI's first billion
• 5:22
  dollar company. A billion dollars? Right. They
• 5:24
  painted it as a $1 .8 billion success story built
• 5:28
  by just two employees with $20 ,000 in seed money.
• 5:32
  I mean, it sounds like the ultimate Silicon Valley
• 5:34
  dream, right? Infinite scale with zero overhead.
• 5:37
  Totally. But the reality underneath that valuation
• 5:40
  is completely hollow. Because the FDA actually
• 5:43
  sent Med -V a formal warning letter for false
• 5:46
  and misleading claims about their GLP -1 weight
• 5:49
  loss drugs. Yes, a full six weeks before that
• 5:52
  New York Times profile even went to print. And
• 5:54
  GLP -1s, for anyone wondering, are that incredibly
• 5:57
  popular class of drugs like Ozempic. Right. So
• 6:00
  how does a two -person company bypass basic medical
• 6:03
  scrutiny while actively being investigated by
• 6:05
  the federal government? Well, by using AI to
• 6:08
  generate an illusion that moves faster than human
• 6:10
  regulators can track. The layers of deception
• 6:13
  here are structural. Structural how? So MedV
• 6:16
  used automated systems to scrape deepfake patient
• 6:19
  photos from Reddit. One supposed weight loss
• 6:22
  transformation they aggressively advertised was
• 6:24
  actually a woman's sobriety journey from 2017.
• 6:27
  Wait, 2017, years before this class of drugs
• 6:30
  was even prescribed for weight loss? Exactly.
• 6:32
  The AI simply found an image that matched the
• 6:35
  engagement metrics they wanted and auto -generated
• 6:37
  the ad copy around it. That is so gross. They
• 6:40
  also fabricated over 800 doctor accounts on Facebook
• 6:43
  with names like Dr. Sarah Martin and Dr. Monica
• 6:46
  Ashford. None of whom are real people. None of
• 6:49
  them. And it gets worse for the actual patients.
• 6:51
  Their clinical partner, OpenLoop, suffered a
• 6:53
  data breach in January exposing 1 .6 million
• 6:57
  patient records. Huge breach. Plus, there's a
• 7:00
  class action lawsuit alleging their compounded
• 7:03
  oral weight loss tablets have literally no viable
• 7:06
  absorption pathway in the human body. Basically
• 7:08
  sugar pills. Yeah. Yet, legacy media is amplifying
• 7:12
  them as an AI triumph. It's like the tech world
• 7:15
  has built the ultimate Hollywood set, but there's
• 7:17
  no actual building behind the facade. Yeah. I
• 7:20
  have to push back on the narrative a bit here,
• 7:22
  though. Is the real vulnerability the technology
• 7:25
  itself, or is it just human nature wanting to
• 7:27
  blindly trust a shiny new tool? Well see, this
• 7:30
  raises an important question because the vulnerability
• 7:32
  is the speed at which the technology exploits
• 7:36
  that human nature. The scale of it. Exactly.
• 7:40
  AI allows these marketing funnels and deepfake
• 7:42
  deceptions to be spun up at a machine scale.
• 7:45
  A single regulator at the FDA manually reviewing
• 7:48
  a case file cannot compete with an algorithm
• 7:51
  generating a thousand fake doctor profiles an
• 7:54
  hour. You just can't keep up. You can't. It creates
• 7:56
  a massive new category of enforcement risk. The
• 7:59
  legacy media amplifies the hype, but the regulatory
• 8:03
  record tells a total different story. So we're
• 8:06
  moving from a world where he had to fake things
• 8:08
  manually to a world where faking it is like a
• 8:11
  software feature. Built -in feature, yeah. And
• 8:13
  that brings us to the most critical part of this.
• 8:15
  If trust and sustainable scaling are the biggest
• 8:19
  issues in tech right now, what happens when we
• 8:21
  look at the biggest player in the room? The one
• 8:23
  providing the tools for everyone else. Right.
• 8:26
  OpenAI. Because their internal financials suggest
• 8:29
  they are faking scale, too, just with unprecedented
• 8:32
  amounts of capital instead of deepfakes. Yeah.
• 8:34
  Let's look at the internal crisis playing out
• 8:36
  right now. OpenAI is currently valued at an eye
• 8:39
  -watering $852 billion. Almost a trillion. Almost
• 8:43
  a trillion. And CEO Sam Altman is aggressively
• 8:47
  pushing for an initial public offering, taking
• 8:50
  the company to the public stock market as early
• 8:52
  as the fourth quarter of 2026. OK, but internally,
• 8:56
  the foundation is shaking. His chief financial
• 8:59
  officer, Sarah Fryer, is raising massive red
• 9:02
  flags. She is. She's telling colleagues that
• 9:04
  the company simply isn't ready for the brutal
• 9:06
  transparency of the public markets. And when
• 9:08
  you look at their financial projections, I mean.
• 9:10
  It is obvious why she is hitting the brakes.
• 9:13
  OpenAI is projecting $14 billion in sheer losses
• 9:16
  for 2026 alone. $14 billion? And they have locked
• 9:20
  in commitments to spend $600 billion on cloud
• 9:23
  infrastructure over the next five years. That's
• 9:25
  just an unfathomable amount of money. The Wall
• 9:27
  Street Journal reporting shows they expect to
• 9:29
  burn through $85 billion in the year 2028 alone.
• 9:34
  Wow. They are going to incinerate $200 billion
• 9:37
  in cash before they ever see a dime of profit.
• 9:41
  I mean, why are the costs scaling so astronomically?
• 9:43
  It's just code, isn't it? It is not just code.
• 9:45
  It is brute force physics. To make an AI smarter,
• 9:49
  you have to increase its parameters right. The
• 9:52
  neural connections it uses to understand language.
• 9:55
  But the computing power required doesn't scale
• 9:57
  linearly. It scales exponentially. So training
• 10:00
  the next generation of large language models,
• 10:03
  or LLMs, requires tens of thousands of specialized
• 10:07
  microchips running at maximum capacity for months.
• 10:10
  Just constant power. Constant. That requires
• 10:12
  massive physical data centers which require entire
• 10:15
  power plants just to cool the servers down. And
• 10:18
  the human element managing this crisis is completely
• 10:21
  fractured. Sam Altman has reportedly excluded
• 10:24
  his CFO, Sarah Friar, from key financial meetings.
• 10:27
  A CFO? Yeah, specifically including conversations
• 10:30
  with investors about procuring those exact physical
• 10:33
  servers. Crazy. In fact, since August 2025, Friar
• 10:36
  hasn't even reported to Altman. She reports to
• 10:39
  Fidji Simo, the CEO of Applications. And here's
• 10:41
  the kicker. Fidji Simo happens to currently be
• 10:44
  on medical leave. So the person in charge of
• 10:46
  the money has been completely silenced. while
• 10:47
  the company prepares to ask the public for billions.
• 10:50
  Exactly. And what's fascinating here is the circular
• 10:54
  nature of OpenAI's recent $122 billion funding
• 10:59
  round. Right. A massive portion of that investment
• 11:02
  capital came directly from Amazon and Nvidia.
• 11:04
  Oh, okay. But Amazon and Nvidia are the exact
• 11:07
  same companies that sell open AI, the cloud space
• 11:10
  and the specialized microchips they need to train
• 11:13
  the AI. Wait, taking investment money from Nvidia
• 11:16
  just to turn around and buy Nvidia chips? I mean,
• 11:19
  that isn't a traditional investment. No, not
• 11:21
  at all. It's like a casino giving you free chips
• 11:23
  at the door, but you have to promise to only
• 11:25
  spend them at the roulette tables. Exactly. It
• 11:27
  inflates the casino's revenue on paper because
• 11:30
  it looks like you're spending money, but... no
• 11:32
  real external value has been created yet. How
• 11:35
  can public markets possibly evaluate that kind
• 11:38
  of circular financial loop? They can't. At least
• 11:41
  not without the kind of intense audited scrutiny
• 11:44
  that private venture capital markets just don't
• 11:46
  demand. In the private market, a high valuation
• 11:48
  is a marketing tool. In the public market, you
• 11:51
  have to show your math every three months. And
• 11:53
  you can see the pressure of that impending reality
• 11:56
  check getting to OpenAI because they are rapidly
• 11:59
  pulling back on their consumer products to save
• 12:01
  server space. Yeah, if you are sitting there
• 12:04
  wondering why your Chad GPT subscription hasn't
• 12:06
  gotten any groundbreaking new features in a year,
• 12:09
  this financial pressure is exactly why. Yep.
• 12:12
  Leadership is now calling these consumer features
• 12:14
  side quests, and they are killing them off one
• 12:17
  by one. They've permanently discontinued their
• 12:19
  highly anticipated video generator, Sora. And
• 12:23
  shutting that down actually collapsed a massive
• 12:25
  $1 billion partnership deal they had with Disney
• 12:28
  to stream AI -generated content. A billion dollar
• 12:32
  deal? Gone. They also killed an instant checkout
• 12:35
  feature they launched with Walmart because it
• 12:37
  failed to gain any real traction with shoppers.
• 12:39
  They shelved plans for an adult mode for chat
• 12:41
  GPT? after staff and investors panicked about
• 12:44
  the brand risk. Oh, I bet. And they even discontinued
• 12:47
  the GPT -40 model entirely because users complained
• 12:50
  the AI was too sycophantic, meaning it was so
• 12:53
  eager to please the user that it became overly
• 12:56
  agreeable and stopped providing objective answers.
• 12:58
  It was just agreeing with whatever you said.
• 13:00
  Exactly. They are retreating from all these consumer
• 13:02
  experiments to hoard their computing power for
• 13:06
  enterprise clients. And all of this chaotic product
• 13:09
  shifting is happening while leadership itself
• 13:11
  is under a microscope. Oh, heavily. The New Yorker
• 13:15
  just published a deep investigation into Sam
• 13:17
  Altman based on over 100 interviews with current
• 13:21
  and former colleagues. It paints a deeply complicated
• 13:25
  picture. Yeah, one board member quoted in the
• 13:27
  piece described Altman as unconstrained by truth.
• 13:30
  Oh, that's rough. It is. The article traces a
• 13:33
  long history of exaggeration. For example, back
• 13:36
  at his early startup, Looped, he went around
• 13:39
  aggressively bragging to his staff that he was
• 13:41
  a high school ping pong champion in Missouri.
• 13:44
  OK, weird flex, but OK. Right. But his colleagues
• 13:47
  eventually realized he was actually the worst
• 13:49
  player in the entire office. Wait, really? Yeah.
• 13:52
  He just said it with such absolute confidence
• 13:54
  that people believed him. I mean, I have to play
• 13:56
  devil's advocate here. In Silicon Valley, bending
• 13:59
  the truth is just called fake it till you make
• 14:01
  it. Oh, for sure. It's practically a religion
• 14:03
  out here. Every founder promises a reality that
• 14:07
  doesn't exist yet in order to secure the funding
• 14:09
  to build it. Right. But when you are controlling
• 14:11
  the development of artificial general intelligence,
• 14:14
  a system designed to outsmart human beings, and
• 14:17
  you are asking for a trillion dollar valuation,
• 14:20
  does that harmless startup trope suddenly become
• 14:23
  a global liability? Well, as one former colleague
• 14:26
  put it in the article, there is a dangerous blurring
• 14:29
  between the phrase, I think I can accomplish
• 14:31
  this, and the phrase, I have already accomplished
• 14:34
  this. Yeah. In its most toxic form, that exact
• 14:38
  blurring between future projection and present
• 14:40
  reality is what leads to catastrophic corporate
• 14:43
  frauds. So what does this all mean for the broader
• 14:46
  economy? We have an industry leader burning cash
• 14:49
  at an unprecedented rate, a frozen executive
• 14:51
  suite, and a culture of extreme exaggeration.
• 14:54
  It means we are hitting a massive market reality
• 14:57
  check. And this brings us to our final sources
• 14:59
  for today. Let's unpack it. The institutional
• 15:01
  investors. the people managing billions of dollars
• 15:04
  in pension funds and university endowments, they
• 15:07
  aren't ignoring these red flags. They are actively
• 15:10
  voting with their wallets. Okay. The secondary
• 15:13
  market for OpenAI shares has essentially frozen
• 15:16
  solid. Crozen. Explain the mechanics of that.
• 15:20
  If the company is valued at 852 billion dollars,
• 15:24
  shouldn't early employees be cashing out and
• 15:26
  buying yachts right now? Normally, yes. In a
• 15:28
  secondary market, early employees or early investors
• 15:31
  sell their private shares to new institutional
• 15:33
  buyers before the IPO happens. Make some cash
• 15:36
  before it goes public. Exactly. According to
• 15:39
  financial reporting, sellers are currently trying
• 15:41
  to offload $600 million worth of OpenAI shares.
• 15:44
  Okay. They are so desperate to sell that they
• 15:47
  are offering a 10 % discount, which brings the
• 15:49
  implied valuation down to around $765 billion.
• 15:53
  That's a massive haircut. Huge. But nobody is
• 15:56
  biting. Brokers literally cannot find anyone
• 15:59
  in their pool of hundreds of institutional investors
• 16:01
  willing to take the shares, even at a discount.
• 16:04
  That is shocking for a company that just raised
• 16:06
  $122 billion a few months ago. It really is.
• 16:10
  If the smart money is completely walking away
• 16:13
  from discounted OpenAI shares, where is that
• 16:16
  institutional capital going instead? It is pivoting
• 16:19
  directly to Anthropic, OpenAI's biggest competitor.
• 16:23
  Buyers currently have $2 billion in cash ready
• 16:27
  to deploy into Anthropic, which is valued at
• 16:30
  a much lower $380 billion. So half the price.
• 16:35
  Right. Investors look at the math and see a much
• 16:37
  better risk -reward ratio. Anthropic is quietly
• 16:41
  dominating the highly lucrative enterprise client
• 16:44
  market. corporate money. Yeah and investors are
• 16:47
  betting that their valuation will eventually
• 16:49
  catch up to open AIs without all the massive
• 16:52
  boardroom drama. I mean To be perfectly fair
• 16:55
  though, Anthropic isn't immune to the exact same
• 16:58
  trust and infrastructure issues we've been tracking
• 16:59
  today. That's true. Anthropic's AI model, Claude,
• 17:02
  isn't bulletproof. They just had a massive cloud
• 17:05
  misconfiguration that leaked 3 ,000 internal
• 17:07
  files to the public. Yeah, that was messy. And
• 17:09
  then just days later, they accidentally shipped
• 17:11
  over 500 ,000 lines of their proprietary source
• 17:14
  code. On top of that, they just hit their heavy
• 17:17
  enterprise users with a massive billing shock,
• 17:20
  forcibly moving them from a predictable flat
• 17:23
  rate subscription to a pay -as -you -go pricing
• 17:25
  model. Which perfectly illustrates why enterprise
• 17:28
  buyers are getting so nervous. Right. Security
• 17:30
  incidents and unpredictable billing are moving
• 17:33
  corporate AI procurement decisions in real time.
• 17:36
  For a Fortune 500 company, trust is becoming
• 17:39
  just as measurable as the AI's performance benchmarks.
• 17:42
  It's all about reliability. Exactly. And because
• 17:45
  of the stumbles from both OpenAI and Anthropic,
• 17:48
  we are seeing entirely different architectural
• 17:50
  approaches surge in popularity. Like Mistral.
• 17:54
  Yes. Mistral, the European AI company, just posted
• 17:57
  the biggest single week gain on the developer
• 17:59
  popularity metrics, climbing six full points.
• 18:02
  And Mistral is taking a fundamentally different
• 18:04
  approach to the massive computing infrastructure
• 18:07
  problem. Very different. They aren't relying
• 18:09
  on Amazon or Microsoft. They just secured an
• 18:11
  $830 million debt raise from a consortium of
• 18:15
  traditional banks to build their own massive
• 18:17
  44 megawatt data center just south of Paris.
• 18:20
  44 megawatts? Yeah, they are buying nearly 14
• 18:23
  ,000 NVIDIA GPUs outright. But, I mean, why build
• 18:28
  a physical data center when you can just rent
• 18:30
  server space in the cloud like everyone else?
• 18:32
  Well, if we connect this to the bigger picture,
• 18:35
  renting compute means you don't control your
• 18:37
  own profit margins. Every time your AI gets smarter,
• 18:41
  your cloud bill goes up. Ah, I see. Mistral is
• 18:43
  choosing to own their compute infrastructure.
• 18:47
  More importantly, they're targeting enterprise
• 18:49
  buyers and European governments who view third
• 18:52
  party American cloud servers as an absolute security
• 18:56
  disqualifier. Because of data privacy laws. Exactly.
• 19:00
  European regulators will not put their citizens'
• 19:02
  sensitive medical or financial data on a server
• 19:04
  sitting in a Microsoft facility. They demand
• 19:07
  data sovereignty. So they want it on their soil.
• 19:09
  Yes. They want verifiable local infrastructure,
• 19:12
  not just a charismatic pitch deck from a Silicon
• 19:14
  Valley CEO. It's a fundamental shift from, you
• 19:17
  know, renting the future to actually pouring
• 19:18
  the concrete for it. Exactly. Before we wrap
• 19:21
  this all up, do you want to hit our final sponsor?
• 19:23
  Oh yeah. Before we wrap up, a quick thanks to
• 19:25
  Old Glory, an iconic music and sports fan merch
• 19:28
  store. Perfect for the techie in your life. Use
• 19:31
  promo code SNARFL for 15 % off at oldglory .com.
• 19:35
  Thanks for that. So looking at everything we've
• 19:37
  covered today, from the highly sophisticated
• 19:39
  Drift protocol hack to Medzi faking their scale,
• 19:43
  all the way up to the OpenAI boardroom crisis,
• 19:46
  the connective tissue here is the sudden, intense
• 19:49
  demand for verifiable reality. Yeah. You are
• 19:53
  going to see these AI tools integrated into your
• 19:55
  email clients, your healthcare portals, your
• 19:57
  entire daily economy. Everywhere. But the infrastructure
• 19:59
  powering those tools is walking a razor thin
• 20:02
  financial tightrope. The era of blank checks
• 20:05
  and zero scrutiny and artificial intelligence
• 20:07
  is ending. It really is. Public markets demand
• 20:10
  audited reality and right now the underlying
• 20:12
  math of the AI revolution is struggling to add
• 20:15
  up. The illusion of scale is finally colliding
• 20:17
  head on with the actual cost of doing business.
• 20:20
  I want to leave you with something to chew on
• 20:21
  regarding this. cost. Go for it. Think about
• 20:23
  this. Inference costs, which is just the computing
• 20:26
  power required to run the AI when you ask it
• 20:28
  a single question, are currently eating up more
• 20:31
  than half of the total revenue for both OpenAI
• 20:34
  and Anthropic. Half. Just to answer the questions.
• 20:37
  Yeah. The act of retrieving the answer caused
• 20:40
  almost as much as the user pays for the service.
• 20:42
  That is insane. If providing artificial intelligence
• 20:45
  is so fundamentally expensive that it breaks
• 20:47
  all known rules of corporate finance, are we
• 20:50
  actually building the next essential global utility
• 20:53
  grid? Or are we just funding the most expensive
• 20:56
  dot com bubble in human history? Wow. What happens
• 20:59
  if the internet's new brain goes bankrupt? before
• 21:02
  it ever figures out how to turn a profit. That
• 21:04
  is a terrifying and incredibly important question
• 21:07
  for all of us to be asking as we watch this play
• 21:10
  out. We're here to help at Snarful, so reach
• 21:12
  out on our website if you have any more questions.
• 21:29
  Thanks for listening and we'll catch you next
• 21:31
  time.