Latest / Connecting the Dots / Frontier AI Compute, Shifting Rivalries, and Space Infrastructure

Transcript

Frontier AI Compute, Shifting Rivalries, and Space Infrastructure · 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 Docs, 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 looking up at the night sky and,
• 0:18
  you know, knowing that the artificial intelligence
• 0:20
  writing your emails, debugging your software,
• 0:22
  and just organizing your life is physically floating
• 0:25
  in a massive solar -powered server farm 300 miles
• 0:29
  above your head. I mean, it sounds completely
• 0:31
  wild. Right. Like pure science fiction. But today,
• 0:35
  that concept just became an actual documented
• 0:38
  business strategy. So welcome in, everyone. I'm
• 0:40
  Alex. And I'm Morgan. And we are the co -hosts
• 0:43
  of Connecting the Dots, broadcasting right here
• 0:45
  from the SNARFL Solutions Group in sunny Sacramento,
• 0:48
  California. That place to be. Absolutely. So
• 0:50
  it is Thursday, May 7, 2026, and we are unpacking
• 0:54
  a lineup today that is honestly going to fundamentally
• 0:58
  change the physical infrastructure of the Internet
• 1:00
  you use every single day. It really is an incredible
• 1:03
  moment in tech right now. I mean, we are looking
• 1:05
  at unlikely corporate alliances, electricity
• 1:08
  demand so massive. They're literally breaking
• 1:11
  local grids. Breaking them. Yeah. And then the
• 1:13
  very real logistical plans to just shoot gigawatts
• 1:16
  of computing power straight into low Earth orbit.
• 1:20
  Which is just staggering. But before we get to
• 1:22
  the heavy hitters, let's get you caught up on
• 1:24
  what's happening outside your window and in your
• 1:26
  wallet. Looking at the morning star data. It
• 1:29
  is going to be an absolutely gorgeous day here
• 1:31
  in Sacramento. Oh, yeah We're talking sunny skies
• 1:34
  a high of 82 degrees Fahrenheit and a really
• 1:37
  nice light breeze like two to six miles per hour
• 1:39
  Honestly, perfect weather to get outside and
• 1:41
  just take a break from the screens But if you
• 1:44
  are traveling today, you do want to keep an eye
• 1:46
  on the broader map There are a hundred and twenty
• 1:48
  six active US weather alerts right now. Wow,
• 1:51
  that's quite a few Yeah, and about sixty one
• 1:53
  of those are over land. So Definitely check weather
• 1:56
  .gov if you're flying or driving out of state.
• 1:59
  Good call. Now, over on Wall Street, investors
• 2:01
  had a very solid Wednesday. The Dow Jones closed
• 2:03
  up nearly 1%, gaining over 468 points to close
• 2:07
  at 49 ,910 .59. Flirting so closely with that
• 2:12
  50 ,000 mark. So close. And the S &P 500 followed
• 2:15
  that trend, also rising almost 1 % to close at
• 2:17
  7 ,365 .12. If you are watching crypto, Bitcoin
• 2:23
  dipped slightly yesterday, just about half a
• 2:26
  percent, but it's still holding strong, opening
• 2:27
  today at over $81 ,400. So the markets are thrilled.
• 2:32
  Sports fans in the Bay Area, not so much. Rough
• 2:36
  day for baseball yesterday, the San Francisco
• 2:38
  Giants lost, and the Oakland A's dropped their
• 2:41
  game too. But hey, the Kings had the night off,
• 2:43
  so we didn't have to watch them lose as well.
• 2:45
  You always have to look for the silver lining.
• 2:46
  Always. Now, Wall Street might be happy today,
• 2:49
  but if you've noticed your favorite AI chatbot
• 2:52
  freezing up on you lately when you're just trying
• 2:54
  to get a report finished, you are actually feeling
• 2:56
  the real world effects of our main topic today.
• 2:59
  So we have a lot of ground to cover. Let's unpack
• 3:01
  this. Let's do it. Today's episode is brought
• 3:03
  to you by Strong Coffee, the premium instant
• 3:06
  coffee that tastes like a traditional brew but
• 3:08
  is packed with 15 grams of protein, MCTs and
• 3:11
  adaptogens like ashwagandha. Get clean, sustained
• 3:15
  energy without the jitters. Use promo code SNARFL
• 3:18
  for 20 % off at strongcoffeecompany .com. Okay,
• 3:21
  so to really grasp the magnitude of our first
• 3:25
  story, you have to understand the physical reality
• 3:27
  of the AI boom. Like you hear the word cloud
• 3:30
  and your brain naturally pictures something light,
• 3:32
  fluffy, and invisible. Right. But the cloud is
• 3:35
  heavy. It is made of steel, concrete, and tens
• 3:39
  of thousands of processors that require just
• 3:41
  a mind -boggling amount of electricity to function.
• 3:43
  And right now, the companies building these AI
• 3:46
  tools are hitting a literal physical wall. There
• 3:49
  isn't enough computing power to go around. Yeah,
• 3:52
  the industry calls it the compute crunch. The
• 3:54
  compute crunch, which brings us to this massive,
• 3:57
  unprecedented mega deal between two absolute
• 4:00
  titans in the space, right? Anthropic, the startup
• 4:03
  behind the Claude Chatbot, and SpaceX AI, which
• 4:06
  is Elon Musk's newly merged aerospace and AI
• 4:09
  company. Right. And this deal is entirely about
• 4:12
  easing that crunch. So Anthropic just signed
• 4:14
  an agreement to gain access to Colossus One,
• 4:17
  which is SpaceX AI's flagship data center in
• 4:19
  Memphis, Tennessee. OK. And Anthropic is getting
• 4:22
  access to more than 300 megawatts of computing
• 4:24
  capacity. Let's let's just pause on that number
• 4:27
  because. 300 megawatts is really hard to visualize.
• 4:30
  It is. That is enough continuous electricity
• 4:33
  to power roughly 225 ,000 homes. Like, an entire
• 4:39
  medium -sized city's worth of power. And they
• 4:41
  are using it just to run one facility full of
• 4:45
  computers. Why? Like, what makes these specific
• 4:48
  machines so power hungry? compared to the laptop
• 4:51
  sitting on my desk. Well, it comes down to the
• 4:53
  architecture of the chips. Yeah. So Colossus
• 4:55
  One houses over 220 ,000 of Nvidia's most advanced
• 4:59
  GPUs. Graphics processing units. Exactly. Specifically,
• 5:03
  the H100s, H200s, and the new GB200 accelerators.
• 5:06
  Wait, I want to clarify something for you listening.
• 5:09
  Why graphics processors? I thought those were
• 5:10
  just for rendering video games. Why can't they
• 5:12
  just use thousands of standard computer processors,
• 5:14
  the CPUs? That is the core of the compute crunch
• 5:17
  right there. Think of a standard CPU, like a
• 5:19
  single master baker in a kitchen. OK, a baker.
• 5:22
  That baker can make a beautiful, incredibly complex,
• 5:25
  multi -tiered wedding cake, but they have to
• 5:28
  bake it one careful step at a time. That's serial
• 5:30
  processing. But an AI doesn't need one complex
• 5:33
  cake. When an AI processes human language, it
• 5:37
  needs to bake 10 million simple cupcakes in a
• 5:40
  fraction of a second. Oh, I see. So a GPU is
• 5:42
  basically a massive warehouse filled with 10
• 5:45
  ,000 tiny basic ovens. all baking cupcakes simultaneously.
• 5:50
  That's exactly that is parallel processing. It
• 5:53
  is really the only way to train and run these
• 5:55
  large language models. But running 10 ,000 tiny
• 5:59
  ovens at the same time requires an exponential
• 6:01
  amount of electricity compared to just running
• 6:04
  one big oven. And right now, Anthropic's customers
• 6:07
  are demanding more cupcakes than the company
• 6:08
  can possibly bake. Way more. Especially the software
• 6:11
  developers. I mean, Anthropic has an AI programming
• 6:14
  assistant called Claude Code, and according to
• 6:17
  their data, the average developer is practically
• 6:19
  living in this tool. How much time are they spending
• 6:21
  in it? Spending at least 20 hours a week running
• 6:23
  it. Wow. Yeah, and they aren't just asking it
• 6:26
  to write a single line of code. They're having
• 6:28
  it analyze entire software libraries, debug complex
• 6:31
  systems, and run these massive multi -step testing
• 6:34
  environments. And because thousands of developers
• 6:37
  were doing that simultaneously, anthropic systems
• 6:40
  were just choking. Completely choking. Users
• 6:42
  were constantly hitting rate limits. They'd get
• 6:44
  a message saying, essentially, you've baked too
• 6:46
  many cupcakes today. Please come back tomorrow.
• 6:48
  Which is terrible when you have a deadline. Exactly.
• 6:51
  But because of this massive injection of compute
• 6:54
  from the Memphis facility, Anthropic announced
• 6:57
  they're completely overhauling their service.
• 7:00
  They are doubling the rate limit for their paid
• 7:02
  plans and completely dropping peak hour usage
• 7:05
  caps. Which directly impacts you, the listener,
• 7:08
  even if you don't write a single line of code.
• 7:11
  Because the professionals building the banking
• 7:13
  apps, the medical software, the entertainment
• 7:15
  platforms you use every day, they now have faster,
• 7:18
  unrestricted tools. Better tools for them. mean
• 7:21
  faster, more reliable apps on your phone. Right.
• 7:24
  And Anthropic also announced a fascinating new
• 7:26
  feature called Dreaming, which is entirely made
• 7:29
  possible by this new compute capacity. They are
• 7:32
  letting their AI systems effectively sleep between
• 7:35
  user sessions. Wait, how does a mathematical
• 7:38
  model sleep? Well, think about human sleep. When
• 7:41
  we sleep, our brains prune neural pathways and
• 7:45
  solidify memories from the day. OK. Anthropic
• 7:48
  is doing this mathematically. When you log off,
• 7:51
  the AI goes offline and reviews the complex tasks
• 7:54
  it just did for you. It recalculates its neural
• 7:57
  weights, basically adjusting the mathematical
• 8:00
  importance it gives to certain concepts based
• 8:02
  on your specific preferences. So the next time
• 8:05
  you log in, it doesn't have to spend precious
• 8:07
  computing power figuring out your coding style
• 8:09
  all over again. Exactly. It already knows. It
• 8:11
  does its homework overnight so it doesn't have
• 8:12
  to slow down during the test. I love that. But
• 8:15
  it also leads me to a glaring question about
• 8:17
  the business side of this. We know Anthropic
• 8:19
  previously committed, what, billions to Google
• 8:22
  for cloud services. Yes. And they committed billions
• 8:25
  more to Amazon. So if they've already handed
• 8:27
  the biggest tech giants in the world hundreds
• 8:29
  of billions of dollars, like a popular local
• 8:31
  bakery getting so many orders they have to rent
• 8:33
  out an industrial factory, why are they suddenly
• 8:37
  renting warehouse space from Elon Musk? I mean,
• 8:40
  are Google and Amazon literally out of power
• 8:41
  lines? In short, yes, the traditional cloud providers
• 8:45
  have a massive infrastructure backlog. They just
• 8:47
  cannot secure the land, the power permits, and
• 8:50
  the cooling systems fast enough to build data
• 8:52
  centers that meet the current demand. Well, moving
• 8:54
  on to our next story, let's look at the personalities
• 8:57
  making this deal happen. Because a backlog at
• 8:59
  Google doesn't fully explain why Anthropic is
• 9:02
  jumping into bed with SpaceX AI. It doesn't.
• 9:05
  The rivalries in the tech world are f - And just
• 9:08
  a few months ago, in February, Elon Musk was
• 9:11
  publicly attacking Anthropic on X. Like he called
• 9:15
  their corporate policies misanthropic and declared
• 9:17
  that their models were evil. It was a scorched
• 9:20
  earth public takedown. Absolutely. But fast forward
• 9:22
  to this week, and the narrative is completely
• 9:25
  reversed. Musk posted that he spent time with
• 9:27
  Anthropic's senior leadership to understand their
• 9:30
  safety protocols. His new verdict. He stated,
• 9:33
  no one set off my evil detector. and praise their
• 9:37
  competence. Okay, I have to push back playfully
• 9:39
  on this whole meeting of the mind's narrative.
• 9:42
  Are we really supposed to believe that Elon Musk
• 9:44
  simply had a nice chat with their leadership
• 9:46
  and decided to hand over his most valuable servers
• 9:49
  for the good of humanity? Why? I mean this isn't
• 9:51
  some classic enemies to lovers movie trope. This
• 9:54
  is about a highly anticipated IPO. Right. You
• 9:57
  were looking right past the PR and seeing the
• 9:59
  balance sheet. SpaceX AI is reportedly seeking
• 10:02
  an initial public offering as soon as next month.
• 10:05
  Next month. Yeah. And when a company goes public,
• 10:08
  institutional investors on Wall Street don't
• 10:10
  care about evil detectors. They care about revenue
• 10:13
  multipliers. And building Colossus One was an
• 10:16
  astonishing capital expense. They bragged about
• 10:19
  building it in a record 122 days on a former
• 10:22
  Electrolux factory site in Memphis. But laying
• 10:25
  thousands of miles of fiber optics and buying
• 10:27
  220 ,000 Nvidia GPUs, that burns through billions
• 10:31
  of dollars in cash. Now that it's built, Wall
• 10:34
  Street is going to demand to see how that massive
• 10:36
  furnace actually generates profit. Exactly. And
• 10:39
  the internal reality at XAI has been turbulent.
• 10:42
  Their proprietary chatbot, Grok, has reportedly
• 10:45
  seen declining usage metrics. They simply don't
• 10:48
  have the user base to justify hoarding 300 megawatts
• 10:50
  of compute all to themselves. Plus, there was
• 10:54
  a massive talent exodus leading up to the SpaceX
• 10:56
  merger. Eleven co -founders and about 80 staff
• 11:00
  members quit. The irony of that talent drain
• 11:03
  is just amazing to me. One of XAI's former co
• 11:07
  -founders, Ross Nordeen, left the company earlier
• 11:09
  this year and joined Anthropic. Right. And his
• 11:11
  publicly stated reason for leaving, he wanted
• 11:14
  to focus on compute. Now the company he fled
• 11:16
  is acting as the landlord, leasing its compute
• 11:19
  back to the company he joined. It just highlights
• 11:21
  how incredibly fluid this industry is. I mean,
• 11:23
  if your own software isn't utilizing your billion
• 11:25
  dollar hardware, the smartest business pivot
• 11:27
  is to rent out the hardware. SpaceX AI is strategically
• 11:30
  pivoting to become a cloud provider. They did
• 11:33
  the exact same thing recently with another AI
• 11:35
  coding startup called Cursor. It's a brilliant
• 11:37
  way to dress up the financials right before asking
• 11:39
  public investors for money. They get to pitch
• 11:42
  themselves not just as an AI creator, but as
• 11:44
  a foundational infrastructure, the unavoidable
• 11:47
  toll road for the entire artificial intelligence
• 11:50
  industry. Musk even framed it by comparing it
• 11:52
  to his aerospace division, noting that SpaceX
• 11:55
  regularly launches satellites for rival aerospace
• 11:58
  companies using fair terms and pricing. So he
• 12:02
  is positioning his company as the inescapable
• 12:05
  logistics provider for the future of tech. But
• 12:08
  there is a massive catch to this toll road. Let's
• 12:11
  look closer because we have to talk about the
• 12:12
  collateral damage happening on the ground and
• 12:14
  what happens when Earth literally runs out of
• 12:16
  room for these machines. Yeah, tucked away in
• 12:19
  the press releases for this Memphis deal was
• 12:21
  a detail that sounds like pure science fiction.
• 12:23
  Anthropic and SpaceX AI officially expressed
• 12:26
  interest in partnering to develop orbital AI
• 12:29
  compute capacity. Orbital? Like, they are seriously
• 12:31
  discussing building gigawatt -scale data centers
• 12:34
  that float in low Earth orbit. Yes. To understand
• 12:38
  why they are looking at space, you have to look
• 12:40
  at the traffic jam they are causing in Tennessee.
• 12:42
  Because getting a supercomputer the size of Colossus
• 12:45
  1 running in 122 days is an engineering marvel,
• 12:49
  but the local power grid in Memphis could not
• 12:52
  magically support a sudden 300 megawatt spike
• 12:56
  in demand. So how did they turn the surfers on?
• 12:59
  They had to brute force the power supply by bringing
• 13:01
  in mobile gas turbines and massive banks of Tesla
• 13:04
  MegaPak batteries. Oh wow! And the local residents
• 13:07
  are furious. There have been numerous complaints
• 13:10
  about the relentless noise and the localized
• 13:12
  air pollution from these gas turbines running
• 13:14
  constantly. Just last month, environmental protesters
• 13:18
  crashed a SpaceX investor gathering over this
• 13:21
  exact issue. Well, that makes sense. It's like
• 13:23
  someone moved an incredibly loud, power hungry
• 13:25
  generator into your living room. Eventually the
• 13:28
  noise, the heat and the sheer footprint become
• 13:30
  unsustainable and you just want to stick the
• 13:31
  generator out in the shed in the backyard. Right.
• 13:33
  Except for the AI industry, the only shed big
• 13:35
  enough left is low Earth orbit. The joint press
• 13:38
  release was surprisingly blunt about this. They
• 13:41
  stated that the compute required to train the
• 13:43
  next generation of AI is outpacing what terrestrial
• 13:46
  power, land, and cooling can deliver on the timelines
• 13:50
  that matter. Earth is simply running out of localized
• 13:53
  resources for this specific type of hyper -concentrated
• 13:56
  demand. Hold on, I have to stop you there. I
• 13:59
  understand running out of land and grid power,
• 14:02
  but cooling? Shooting servers into space sounds
• 14:05
  like an absolute thermodynamic nightmare. It
• 14:07
  does I know space is cold, but it's a vacuum
• 14:09
  right it acts like a giant yeti thermos Heat
• 14:13
  gets trapped if half the energy of a data center
• 14:16
  on earth is just running the air conditioning
• 14:19
  How do you cool a server in a vacuum where there's
• 14:22
  no air to blow over the heat sinks? That is a
• 14:24
  phenomenal catch and it's the biggest engineering
• 14:26
  hurdle they face. You are entirely correct. You
• 14:29
  cannot use convection cooling in space. A fan
• 14:31
  does nothing in a vacuum. Exactly. So while the
• 14:33
  ambient environment is cold, shedding the intense
• 14:36
  heat generated by hundreds of thousands of GPUs
• 14:39
  is incredibly difficult. So how do they plan
• 14:41
  to do it? They have to design massive radiational
• 14:44
  cooling fins. The heat has to be captured via
• 14:46
  liquid cooling loops directly on the chips, transported
• 14:49
  out to these giant external panels, and literally
• 14:52
  radiated out into the void of space as infrared
• 14:55
  light. That is a massive engineering challenge.
• 14:58
  So if cooling is actually harder and launching
• 15:01
  heavy steel servers on rockets is astronomically
• 15:04
  expensive, why are they taking this seriously?
• 15:06
  Because of the power generation. Space offers
• 15:09
  near limitless uninterrupted solar energy. If
• 15:12
  you position the orbit correctly, You don't have
• 15:15
  to deal with nighttime, you don't have cloud
• 15:16
  cover, and you don't have weather. You have constant,
• 15:19
  100 % efficient solar radiation to power these
• 15:22
  massive server farms. Oh, wow. And just as importantly,
• 15:25
  you have zero impact on the terrestrial power
• 15:27
  grid. No noise complaints, no gas turbines polluting
• 15:30
  neighborhoods. I saw a great quote about this
• 15:32
  from Ryan Mallory, the CEO of a terrestrial data
• 15:35
  center operator called FlickCentral. He said,
• 15:38
  the fact that serious companies are even discussing
• 15:41
  compute capacity in space tells you how aggressively
• 15:44
  the market is searching for scale. It proves
• 15:47
  this isn't just Elon Musk tweeting a crazy idea,
• 15:50
  you know. This is a legitimate exploration of
• 15:52
  the final frontier of computing logistics. And
• 15:55
  if any single entity can pull this off. It is
• 15:58
  SpaceX. The documentation noted that they are
• 16:00
  the only organization on the planet with the
• 16:03
  launch cadence to make this an engineering reality
• 16:06
  rather than just a research concept. Right. I
• 16:08
  mean, they already managed thousands of Starlink
• 16:09
  satellite launching orbital data centers is kind
• 16:12
  of the next logical step to monopolize the physical
• 16:15
  infrastructure of the Internet. It is wild to
• 16:16
  think about. Yeah. Before we wrap up, a quick
• 16:18
  thanks to Old Glory, an iconic music and sports
• 16:21
  fan merch store. Perfect for the techie in your
• 16:23
  life. Use promo code SNARFL for 15 % off at OldGlory
• 16:27
  .com. So when we step back and look at these
• 16:30
  three massive shifts together, a very clear,
• 16:33
  slightly overwhelming picture emerges of where
• 16:36
  our digital lives are heading. It really all
• 16:38
  starts right in the palm of your hand. The everyday
• 16:40
  apps we rely on to write emails, generate code,
• 16:43
  and manage our workflows require so much raw
• 16:46
  computing power that former bitter rivals are
• 16:49
  teaming up. Yeah, they're sharing massive supercomputers,
• 16:52
  pivoting their business models into infrastructure
• 16:54
  toll roads, and seriously planning to shoot data
• 16:57
  centers into outer space just to keep our digital
• 17:01
  world running. Which leaves us with a profound
• 17:04
  question to consider. If the physical infrastructure,
• 17:07
  the brains of our artificial intelligence, eventually
• 17:10
  lives entirely in orbit, powered by the sun,
• 17:13
  how will that change who controls the future
• 17:16
  of the internet? When the world's most powerful
• 17:18
  computers are no longer tied to the laws or the
• 17:21
  power grids of any one specific country, the
• 17:23
  balance of global power is going to shift in
• 17:25
  ways we are only just beginning to map out. A
• 17:28
  completely borderless, orbital intelligence run
• 17:30
  by private infrastructure companies? It is definitely
• 17:33
  something for you to mull over the next time
• 17:35
  you wait a few seconds for an app to load. We're
• 17:37
  here to help at Snarful, so reach out on our
• 17:38
  website if you have any more questions. That's
• 17:44
  a wrap on today's top stories. Remember, tech
• 17:47
  doesn't have to be complicated, and we're here
• 17:49
  to help connect the dots. If you've got questions
• 17:52
  or want to learn more, visit us at snarful .com.
• 17:55
  Thanks for listening, and we'll catch you next
• 17:58
  time.