Latest / Connecting the Dots / AI Investment Reality, Market Corrections, and Investor Ripple Effects

Transcript

AI Investment Reality, Market Corrections, and Investor Ripple Effects · Connecting the Dots

• 0:00
  Today's tech news moves fast, but understanding
• 0:03
  it shouldn't be hard. That's why Snarful 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 handing your engineering team
• 0:17
  a corporate credit card in January, and by April...
• 0:22
  finding out they have completely blown through
• 0:25
  the entire year's software budget. Yeah, that
• 0:27
  sounds like a pretty catastrophic accounting
• 0:29
  error. It really does. But that is exactly what
• 0:32
  just happened at Uber. And the culprit behind
• 0:36
  that massive bill is artificial intelligence.
• 0:39
  Wow. Yeah. So today we were looking at the severe
• 0:42
  reality check hitting the tech world and why
• 0:45
  the blank check era of AI is officially over.
• 0:49
  Welcome to Connecting the Dots, broadcasting
• 0:51
  daily from Snarfful Solutions Group right here
• 0:53
  in Sacramento, California. I'm Morgan. And I'm
• 0:56
  Alex. Today is Friday, June 26, 2026. And before
• 0:59
  we get into this massive shift in the tech landscape,
• 1:02
  let's get you caught up on what's happening outside
• 1:04
  your window and in your wallet. Definitely. Locally
• 1:07
  here in Sacramento, we are looking at a beautiful
• 1:09
  sunny day. We've got a high of 82 degrees Fahrenheit
• 1:12
  with winds blowing around 9 to 15 miles per hour.
• 1:16
  Sounds perfect. But you know, if you are listening
• 1:18
  from outside the capital region, you will want
• 1:20
  to stay a bit more vigilant. Oh, for sure. Yeah,
• 1:22
  we are currently tracking 318 active U .S. weather
• 1:26
  alerts right now, and 200 of those are actually
• 1:28
  happening over land. That is a lot of alerts.
• 1:30
  It is. So please make sure to head over to weather
• 1:33
  .gov to grab the specific safety details for
• 1:35
  your local area. Now looking at the markets,
• 1:38
  we have some really fascinating movement. The
• 1:40
  Dow Jones closed on Thursday at 51 ,920 .62.
• 1:45
  Right. That is a slight drop of 88 .40 points
• 1:48
  for the day. But I mean, if you zoom out, the
• 1:50
  index is actually still up over 348 points for
• 1:53
  the week. Not bad at all. Not at all. The S &P
• 1:55
  500 closed at 7 ,357 .49, which is a drop of
• 1:59
  47 .42 points. And checking in on crypto, Bitcoin
• 2:03
  opened this Friday morning around 59 ,000. But
• 2:07
  it has since dipped a bit to 58 ,737 .34. You
• 2:12
  know, the underlying narrative behind those numbers
• 2:13
  is basically a story of two different economies
• 2:15
  right now. How so? Well, we are seeing intense
• 2:18
  volatility in the tech sector, which, by the
• 2:20
  way, ties directly into our main story today.
• 2:22
  But the non -tech sectors, they are actually
• 2:23
  having a phenomenal week. Oh, interesting. Yeah,
• 2:26
  the traditional non -tech companies are stepping
• 2:28
  up and doing the heavy lifting right now. They
• 2:30
  are keeping the broader market relatively stable
• 2:32
  while Silicon Valley works through some some
• 2:36
  pretty severe growing pains. That makes a lot
• 2:38
  of sense. And hey, we have some great news on
• 2:40
  the sports front to wrap up the week, too. Oh,
• 2:42
  yes, we do. The Athletics beat the San Francisco
• 2:45
  Giants 9 -6 on Thursday. Such a good game. It
• 2:49
  really was an incredible performance, especially
• 2:52
  from Jeff Hartleap. I mean, he really stepped
• 2:54
  up to secure the win as the pitching hero of
• 2:57
  the game. Yeah, just a fantastic game to watch.
• 2:59
  All right, let's unpack this. Today's episode
• 3:01
  is brought to you by Strong Coffee, the premium
• 3:03
  instant coffee that tastes like a traditional
• 3:05
  brew, but is packed with 15 grams of protein,
• 3:09
  MCTs, and adaptogens like ashwagandha. Get clean,
• 3:12
  sustained energy without the jitters. Use promo
• 3:14
  code SNARFL for 20 % off at strongcoffeecompany
• 3:17
  .com. So we are looking closely today at a really
• 3:22
  massive financial reckoning inside the artificial
• 3:24
  intelligence industry. Yeah, reckoning is the
• 3:26
  right word for it. Over the last couple of years,
• 3:28
  AI has basically been treated like magic by corporate
• 3:31
  boards. The mandate was just, you know, implemented
• 3:34
  everywhere at any cost. Right. Just get it done.
• 3:36
  Exactly. But the bill is finally coming due.
• 3:40
  And the sticker shock is sending shockwaves from
• 3:42
  small local startups straight up to the global
• 3:45
  stock markets. Because the core issue here is
• 3:48
  a fundamental misunderstanding of what computing
• 3:50
  actually costs in the AI era. For the last two
• 3:54
  years, companies have operated under a strategy
• 3:56
  known in the developer world as token maxing.
• 4:00
  Which, I mean, that sounds like a cryptocurrency
• 4:02
  scam, doesn't it? It really does sound like one.
• 4:04
  But it was actually the prevailing corporate
• 4:06
  strategy. A token is just the basic unit of data
• 4:10
  an AI processes like a syllable and a word. Right.
• 4:13
  Every single time you prompt an AI, and every
• 4:16
  time it generates a response, it consumes tokens.
• 4:19
  And those tech giants charge you by the token.
• 4:21
  The meter is always running. Always. Under token
• 4:24
  maxing, companies literally told their developers
• 4:27
  to prioritize speed and capability over cost.
• 4:30
  Just build it fast. Yeah, maximize the tokens
• 4:32
  to build new tools quickly and just sort of assume
• 4:35
  the return on investment would magically appear
• 4:37
  later. And well, that strategy worked when venture
• 4:39
  capital money was essentially free, right? When
• 4:42
  the goal was just getting a product to market.
• 4:44
  But fast forward to today and the financial realities
• 4:47
  are catching up fast. When we look at standard
• 4:51
  cloud computing like fetching an email or loading
• 4:54
  a web page, the cost is literally fractions of
• 4:57
  a penny. You are just retrieving existing data.
• 4:59
  Right. It's super cheap. But generative AI is
• 5:02
  entirely different. Every single time you ask
• 5:04
  a frontier model to write an email, it is running
• 5:07
  billions of mathematical parameters through massive
• 5:10
  server farms just to predict the next word. The
• 5:13
  compute power required is staggering. Which brings
• 5:16
  us right back to Uber. They've fully embraced
• 5:18
  that token maxing philosophy to build out their
• 5:21
  customer service bots and internal tools. And
• 5:23
  how did that go? Well, the sheer volume of computing
• 5:26
  power they needed burned through their entire
• 5:28
  annual AI budget in four months. Four months!
• 5:31
  That is insane! It is. They have now been forced
• 5:34
  to implement these strict spending tiers, where
• 5:37
  some teams are capped at just $1 ,500 a month.
• 5:40
  Wow. That is a hard pivot. But what really fascinates
• 5:44
  me is how the smaller companies are reacting
• 5:46
  to this. Take this AI startup called Lindy, led
• 5:49
  by CEO Flo Crivello. Okay. They built their entire
• 5:53
  platform on Anthropic's Claude model, which is
• 5:56
  one of those massive premium AI brains. But the
• 6:00
  API costs were literally bleeding them dry. So
• 6:03
  they had to make a change. A huge one. To stop
• 6:06
  the bleeding, Lindy had to completely overhaul
• 6:08
  their infrastructure. They migrated 100 % of
• 6:11
  their traffic away from Claude and switched to
• 6:14
  an open -weight Chinese alternative called DeepSeek.
• 6:16
  And did it work? Yeah. Crevella noted that their
• 6:19
  cost curve basically crashed to the floor, saving
• 6:22
  the company millions. He characterized the switch
• 6:24
  as a literal matter of survival. We should probably
• 6:26
  clarify what open -weight actually means here
• 6:29
  because it is so central to why this move saved
• 6:31
  them so much money. Yeah. Let's break that down.
• 6:33
  When you use a closed model like Claude or a
• 6:37
  chat GPT, you are paying a toll to anthropic
• 6:41
  or open AI every single time data passes through
• 6:46
  their proprietary black box. Right. An open weight
• 6:49
  model means the underlying architecture, the
• 6:52
  actual trained parameters or weights of the neural
• 6:54
  network. Those are made available to the public.
• 6:57
  So no toll booth. Exactly. You don't have to
• 6:59
  pay a toll for the software. You just rent your
• 7:01
  own servers and run the model yourself. I mean,
• 7:03
  it is the difference between eating every meal
• 7:05
  at a Michelin star restaurant versus buying the
• 7:08
  ingredients and cooking the meal in your own
• 7:10
  kitchen. That's a great way to put it. You still
• 7:12
  pay for the groceries and the gas bill, but you
• 7:14
  aren't paying the restaurant's massive markup.
• 7:17
  And this move by Lindy represents a much broader
• 7:19
  industry shift away from token maxing toward
• 7:23
  something called model routing. Right, model
• 7:25
  routing. Yeah, I look at model routing like,
• 7:27
  well, like buying a Ferrari. OK, I'm listening.
• 7:29
  If you buy a Ferrari, you use it to drive on
• 7:31
  the Autobahn. You don't buy a Ferrari just to
• 7:34
  drive to the end of your driveway to check the
• 7:35
  mail. No, that would be completely ridiculous.
• 7:38
  Right. But for the last two years, companies
• 7:40
  have been doing exactly that, using incredibly
• 7:42
  expensive, massive AI brains to do basic, repetitive
• 7:47
  tasks. like extracting a date from a calendar.
• 7:50
  That's just such a waste of resources. Model
• 7:52
  routing is the realization that you should build
• 7:54
  software that acts like a traffic cup. It routes
• 7:57
  the simple basic tasks to a cheap open weight
• 8:00
  model and only sends the highly complex logic
• 8:03
  problems to the expensive frontier models. And
• 8:06
  that traffic cut mechanism is essential now because
• 8:09
  the capital required to run those frontier models
• 8:11
  is completely reshaping the global economy. It
• 8:14
  really is. Yeah. And it is creating a severe
• 8:16
  fracture in the tech sector, which analysts are
• 8:19
  currently calling the K -shaped AI market. OK,
• 8:22
  we usually hear about a K -shaped recovery in
• 8:24
  economics, right? Yeah. Where the wealthy recover
• 8:26
  quickly. So the line going up while the working
• 8:28
  class struggles, the line going down. Exactly.
• 8:31
  We are seeing that exact same divergence right
• 8:33
  now on Wall Street, but it is entirely based
• 8:36
  on whether a company is buying AI infrastructure
• 8:38
  or selling it. Right, so the downward slope of
• 8:41
  that K is occupied by the software model makers
• 8:45
  and big tech giants. Ones that are buying the
• 8:47
  infrastructure. Yep. Microsoft and Meta have
• 8:50
  actually entered a literal bear market, falling
• 8:52
  20 % from their recent peaks. And the rest of
• 8:55
  the magnificent seven companies like Amazon and
• 8:57
  Google, they are in correction territory, down
• 9:00
  at least 10%. Wow. They are the ones spending
• 9:02
  billions upon billions of dollars just to build
• 9:05
  massive AI data centers. And then the upward
• 9:08
  slope of the K, belongs to the hardware companies,
• 9:10
  the ones actually selling the computer chips
• 9:12
  to fill those data centers. Exactly. Micron,
• 9:15
  which is one of the world's leading memory chip
• 9:17
  makers, they just saw their stock surge nearly
• 9:20
  16 % on stellar earnings. I mean, it feels like
• 9:24
  we were watching a replay of the 1849 gold rush.
• 9:26
  Oh, totally. During the gold rush, the prospectors
• 9:29
  panning for gold in the river mostly went broke.
• 9:31
  The people who actually made generational wealth
• 9:33
  were the merchants selling the picks, the shovels,
• 9:36
  and the denim jeans. Selling the tools. Yeah,
• 9:39
  the AI software companies look like the modern
• 9:42
  -day prospectors burning through cash, while
• 9:44
  the semiconductor companies are getting rich
• 9:47
  selling the shovels. Because the market valuations
• 9:49
  for those software giants were built entirely
• 9:51
  on the promise of future efficiency, investors
• 9:54
  rewarded them for laying the groundwork. But
• 9:56
  the patience is wearing thin. Very thin. Now
• 9:59
  Wall Street wants to see actual bottom -line
• 10:01
  profit, and the math just isn't adding up yet
• 10:04
  because the infrastructure is so incredibly expensive.
• 10:07
  It's astronomical. It is. Building an AI data
• 10:10
  center doesn't just require advanced processing
• 10:13
  chips. It requires unimaginable amounts of high
• 10:16
  bandwidth memory. The neural networks need to
• 10:20
  hold massive amounts of data in short term memory
• 10:22
  just to generate those responses in real time.
• 10:25
  Which, by the way, is causing real world collateral
• 10:28
  damage for the average consumer. Yeah, it hits
• 10:30
  home. We are seeing this demand for memory chips
• 10:32
  bleed directly into the products you buy. Apple
• 10:35
  just had to raise the retail prices on MacBooks
• 10:37
  and iPads, specifically because they are facing
• 10:40
  severe memory shortages. That is wild. The global
• 10:43
  supply of these components is finite. So you
• 10:46
  are paying more for a tablet at the local electronics
• 10:48
  store simply because a massive data center in
• 10:52
  Virginia is buying up all the memory to run a
• 10:54
  corporate chatbot. Unbelievable. And when Apple
• 10:57
  announced those price hikes, their stock tumbled
• 10:59
  over 6%. They are caught right in the crossfire
• 11:02
  of this k -shaped divide and the ripple effects
• 11:05
  of that capital burn rate are shaking the very
• 11:08
  foundation of the industry Which leads us to
• 11:10
  the biggest prospector in the river suddenly
• 11:13
  reconsidering its next move Yeah The extreme
• 11:15
  costs and the resulting tech queasiness on Wall
• 11:18
  Street have forced the biggest name in AI to
• 11:21
  hit the brakes on going public open AI The company
• 11:24
  that effectively launched this entire generative
• 11:26
  AI frenzy. The New York Times broke the rumor
• 11:29
  that OpenAI is leaning heavily toward delaying
• 11:31
  its initial public offering until 2027. Wow.
• 11:34
  And it all traces back to this exact market volatility.
• 11:37
  Right. CEO Sam Altman has a very specific target
• 11:40
  in mind. He wants the company valued at $1 trillion.
• 11:43
  That is a massive target. Huge. But there are
• 11:47
  very real internal fears that if they open their
• 11:50
  accounting books to a jittery Wall Street right
• 11:52
  now, investors will basically balk at the sheer
• 11:56
  scale of their computing expenses and they will
• 11:58
  miss that trillion dollar mark entirely. You
• 12:01
  really cannot overstate the psychological impact
• 12:03
  of just the rumor of this delay. I believe it.
• 12:06
  The entire global tech economy has been banking
• 12:09
  on the OpenAI IPO to validate the last two years
• 12:13
  of massive spending. When the market leader flinches,
• 12:17
  the global supply chain panics. The international
• 12:19
  fallout was instantaneous and severe, wasn't
• 12:21
  it? Very severe. SoftBank, the massive tech investment
• 12:24
  conglomerate led by Masayoshi Sun, they had placed
• 12:27
  heavy bets on OpenAI. Following the rumor, SoftBank
• 12:30
  shares tanked over 12%. Ouch. Yeah. Japan's Nikkei
• 12:34
  index erased its recent record highs, dropping
• 12:36
  over 4%. But the most dramatic reaction was actually
• 12:39
  in South Korea, where the KOSPI index fell 5
• 12:41
  .8 % and literally tripped circuit breakers.
• 12:45
  For context, tripping a circuit breaker means
• 12:47
  the stock exchange literally pulls the emergency
• 12:49
  break and halts all trading for 20 minutes just
• 12:52
  to prevent a complete market freefall. That is
• 12:55
  wild. I mean, a 20 -minute halt in an entire
• 12:57
  nation's stock market because of an IPO rumor
• 13:00
  in California sounds like an extreme overreaction.
• 13:03
  Why is South Korea so hypersensitive to OpenAI's
• 13:06
  timeline? Well, it comes down to the composition
• 13:09
  of their economy. Nearly half the total value
• 13:11
  of the South Korean KOSPI index is tied up in
• 13:14
  just two companies, Samsung and SK Hynex. And
• 13:17
  their primary export is the exact high bandwidth
• 13:20
  memory chips we were just talking about. They
• 13:22
  are the central suppliers for the global AI infrastructure.
• 13:25
  Oh, wow. Yeah. We also saw Kioxia, a major Japanese
• 13:29
  memory maker, slide over 11%. If OpenAI hints
• 13:33
  that they might need to slow down their timeline,
• 13:35
  the hardware suppliers who depend on their endless
• 13:37
  server purchases immediately crater. This leaves
• 13:40
  a massive lingering question for anyone trying
• 13:42
  to understand where this industry is heading.
• 13:44
  If the crown jewel of the AI boom is scared to
• 13:47
  face public investors, is the AI bubble finally
• 13:50
  bursting? Or is the industry just putting on
• 13:52
  its big kid pants and learning how to run a sustainable
• 13:55
  business? The underlying data strongly suggests
• 13:58
  it is the latter. Okay, so no burst bubble. No,
• 14:01
  the technology itself is not failing. And demand
• 14:04
  is certainly not slumping. In fact, enterprise
• 14:07
  demand is surging right now. That is good news.
• 14:09
  It is. The revenue run rates for these top tier
• 14:12
  companies are staggering. Anthropic is currently
• 14:16
  looking at a $47 billion annualized run rate.
• 14:20
  OpenAI is pacing closer to $25 billion. Wait,
• 14:24
  let's pause on that term run rate because I think
• 14:26
  it is crucial for understanding the disconnect
• 14:28
  here. Sure. Saying a company has a $25 billion
• 14:31
  run rate. Sounds like they have $25 billion just
• 14:35
  sitting in the bank. But a run rate is just taking
• 14:37
  their current monthly revenue and multiplying
• 14:40
  it by 12, assuming demand stays perfectly flat.
• 14:43
  It is a snapshot of revenue, not profit. Precisely.
• 14:46
  The top line revenue is fantastic. The problem
• 14:48
  is the bottom line. Right. If you are taking
• 14:50
  in 25 billion dollars a year, but you are spending
• 14:52
  30 billion on server costs, electricity and hardware
• 14:55
  just to keep the lights on, you have a fundamentally
• 14:58
  broken business model. Exactly. The industry
• 15:00
  isn't suffering from a lack of interest. It is
• 15:02
  just desperately trying to fix its cost structure
• 15:04
  so it can actually keep the money it makes. Which
• 15:06
  circles us right back to the Lindy story. How
• 15:09
  do the tech giants fix their margins? when the
• 15:12
  technology is inherently so expensive to run,
• 15:15
  they have to build smaller, cheaper versions
• 15:17
  of their own products. And they are doing exactly
• 15:20
  that. Big tech incumbents like Google and Amazon
• 15:23
  recognize that if they want AI integrated into
• 15:25
  every small business on earth, they cannot rely
• 15:28
  solely on the massive frontier model. It's just
• 15:31
  too expensive. Right. So they are aggressively
• 15:34
  building lighter weight models. Google just rolled
• 15:37
  out Gemini 3 .5 Flash, which offers solid performance
• 15:40
  at half or even a third of the price of their
• 15:43
  flagship model. The engineering behind those
• 15:45
  lightweight models is fascinating to me. They
• 15:47
  use a process called distillation. Right, distillation.
• 15:50
  Instead of training a massive neural network
• 15:52
  on the entirety of human knowledge, so it can
• 15:54
  write Python code, compose a sonnet, and pass
• 15:57
  the bar exam, they use the massive model to train
• 16:00
  a smaller, highly focused model. Which is so
• 16:02
  much more efficient. Exactly. You end up with
• 16:05
  a streamlined AI that might only be good at summarizing
• 16:08
  customer service emails, but it does so using
• 16:11
  a fraction of the compute power. They created
• 16:13
  the most complex technology in human history,
• 16:15
  and now they're racing to figure out how to sell
• 16:17
  a discount specialized version of it just to
• 16:20
  survive the math. Before we wrap up, a quick
• 16:22
  thanks to Old Glory, an iconic music and sports
• 16:25
  fan merch store. Perfect for the techie in your
• 16:28
  life. Use promo code SNARFL for 15 % off. oldglory
• 16:33
  .com. So to pull all these threads together,
• 16:36
  what we are witnessing is the definitive end
• 16:39
  of the AI blank check era. It's over. The honeymoon
• 16:43
  phase where companies could throw unlimited capital
• 16:44
  at token maxing and just hope the profit would
• 16:46
  figure itself out. is over. Businesses are demanding
• 16:49
  ruthless efficiency now. And Wall Street is reflecting
• 16:52
  that exact new reality. Investors are rewarding
• 16:55
  the hardware manufacturers who are generating
• 16:57
  actual cash flow today while aggressively punishing
• 17:00
  the software giants who are burning through capital.
• 17:03
  And the global markets are absorbing the turbulence
• 17:05
  as industry leaders like OpenAI accept that they
• 17:09
  may need to delay their grand public debuts until
• 17:11
  they can prove their business model is actually
• 17:14
  sustainable. It is setting up to be a highly
• 17:16
  turbulent few months for tech stocks, for sure,
• 17:19
  but ultimately it is a necessary correction for
• 17:21
  the long -term health of the industry. Absolutely.
• 17:24
  But as we leave you today, there is a lingering
• 17:26
  thought worth pondering. We have talked extensively
• 17:30
  about how these massive frontier models are becoming
• 17:33
  too expensive for the average startup to run.
• 17:35
  Right. And we noted that the only entities with
• 17:38
  the infrastructure and capital to offer cheap,
• 17:41
  reliable, lighter weight alternatives at a global
• 17:43
  scale are the established big tech incumbents,
• 17:46
  companies like Microsoft, Google, and Amazon.
• 17:48
  The big guys. Exactly. Yeah. So if the sheer
• 17:50
  cost of cutting edge AI forces everyone from
• 17:53
  local bakeries to multinational logistics firms
• 17:56
  to rely exclusively on the streamlined models
• 17:59
  provided by big tech. Yeah. Are we accidentally
• 18:02
  building a future where just three companies
• 18:03
  control the basic digital logic of the entire
• 18:06
  global economy? That is a very intense thought.
• 18:09
  It really is something to keep in mind the next
• 18:11
  time you use a chat bot to route a simple email.
• 18:14
  We're here to help at Snrful, so reach out on
• 18:16
  our website if you have more questions. That's
• 18:21
  a wrap on today's top stories. Remember, tech
• 18:24
  doesn't have to be complicated, and we're here
• 18:26
  to help connect the dots. If you've got questions
• 18:29
  or want to learn more, visit us at snarful .com.
• 18:33
  Thanks for listening and we'll catch you next
• 18:35
  time.