Latest / Connecting the Dots / Alibaba’s Qwen3 Models, the AI Data Center Race, and Spotify’s DJ Comeback

Transcript

Alibaba’s Qwen3 Models, the AI Data Center Race, and Spotify’s DJ Comeback · 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. Good morning and welcome back to Connecting
• 0:18
  the Dots. I'm Alex. And I'm Morgan. It is Wednesday,
• 0:21
  September 24th, 2025. Uh -huh. And, you know,
• 0:25
  looking out the window here in Sacramento, looks
• 0:27
  like we've got increasing clouds today, but still...
• 0:29
  Pretty warm, high of about 87 Fahrenheit expected.
• 0:33
  Yeah, pleasant enough. Now, before we get into
• 0:37
  the tech headlines from our Morningstar report,
• 0:39
  let's take a quick look at the markets from yesterday.
• 0:42
  Right. It was a bit of a mixed picture, wasn't
• 0:44
  it? Definitely. We saw the Dow Jones slip just
• 0:47
  a tiny bit down 0 .12%. The S &P 500 also dipped
• 0:52
  about 0 .27 % down. OK, so a little bit of red
• 0:55
  there. A little bit. But Bitcoin, well Bitcoin
• 0:57
  kind of bucked that trend. It actually saw a
• 0:59
  nice gain up 0 .88%. Interesting. Wish we could
• 1:02
  say the same for the Giants though. Oh, don't
• 1:04
  remind me. Yeah. Tough loss yesterday, September
• 1:06
  23rd against the Cardinals. Real nail -biter
• 1:09
  8 to 9. Yeah, that one stung. But hey, onwards
• 1:13
  and upwards, right? We've got some pretty huge
• 1:15
  tech stories to unpack. Absolutely. But first,
• 1:18
  let's talk about one of the great companies helping
• 1:20
  us bring you the show. We're supported by Pins
• 1:23
  and Aces. That's Pins and Aces, yeah. They're
• 1:25
  a fantastic premium golf apparel and accessory
• 1:29
  store. If you're looking for, you know, high
• 1:31
  quality gear, stuff that looks good and performs
• 1:33
  well on the course. They got you covered. They
• 1:35
  really do. From apparel to head covers, ball
• 1:38
  markers, all that stuff. And we've got a great
• 1:40
  deal for you. Use our promo code SNARFL. That's
• 1:43
  S -N -A -R -F -U -L. SNARFL when you check out.
• 1:47
  And what does that get you? It gets you 21 %
• 1:49
  off your entire purchase. That's a really solid
• 1:51
  discount. 21%. Nice, yeah. Just head over to
• 1:54
  their website, pinsandaces .com. That's pinsandaces
• 1:58
  .com, use code SNARFL. Awesome. Okay, let's dive
• 2:02
  into our first major story. This one's about
• 2:04
  the global AI infrastructure race. It's really
• 2:07
  heating up and OpenAI's Stargate project is front
• 2:10
  and center. Okay, Stargate. The scale here is
• 2:13
  just... It's almost hard to grasp, isn't it?
• 2:16
  It really is. We're talking about OpenAI planning
• 2:19
  five new data centers just in the U .S., locations
• 2:23
  popping up in Texas, New Mexico, Ohio, the Midwest.
• 2:27
  And the power consumption. Get this, nearly seven
• 2:30
  gigawatts combined. Seven gigawatts. Okay, so
• 2:33
  to put that in perspective for everyone listening,
• 2:35
  think about a large nuclear power plant. That's
• 2:39
  roughly one gigawatt. So OpenAI's plan needs
• 2:41
  the power equivalent of about seven of those
• 2:44
  nuclear reactors, just running constantly. Exactly.
• 2:46
  It shows you the bottleneck for AI isn't just
• 2:49
  the chips anymore. It's the sheer amount of electricity
• 2:51
  needed to power them and keep them cool. Right.
• 2:54
  And the way they're financing this is also pretty
• 2:56
  interesting. It tells a story in itself. Yeah.
• 2:58
  It's not just OpenAI writing a giant check. They're
• 3:00
  bringing in massive partners. Oracle, for instance.
• 3:03
  OK. What's Oracle's role? Well, Oracle is actually
• 3:05
  paying for and building three of these five centers.
• 3:08
  The report mentions they're using, and I quote,
• 3:11
  interesting new corporate structures and wise
• 3:13
  of doing financing. Interesting. So Oracle builds
• 3:16
  it, funds it. Then what? OpenAI just rents the
• 3:20
  compute power. Essentially, yes. OpenAI purchases
• 3:23
  the computing power from Oracle. It's almost
• 3:25
  like... Like a massive utility deal Oracle acts
• 3:29
  like the power company and open AI is the main
• 3:32
  customer So Oracle becomes kind of a specialized
• 3:35
  landlord or power provider for AI compute pretty
• 3:38
  much It helps open AI manage the immense upfront
• 3:41
  cost and risk and remember this is part of a
• 3:43
  much bigger picture open AI Amazon Google meta
• 3:47
  Microsoft They're collectively spending over
• 3:50
  325 billion dollars on this stuff this year Wow
• 3:54
  325 billion. Yeah. So finding creative ways to
• 3:56
  finance is key. SoftBank is doing something similar
• 3:59
  for the other two sites through SB Energy raising
• 4:02
  debt from banks. So it's like a land grab, but
• 4:04
  for electricity and compute capacity. That's
• 4:07
  a good way to put it. And you can't talk about
• 4:08
  this race without mentioning the competition,
• 4:10
  right? Elon Musk's company XAI. Ah yes, Colossus
• 4:14
  2. Their approach seems a bit different though.
• 4:16
  It is. While OpenAI is spreading at 7 GW across
• 4:20
  5 sites, XAI is aiming for, like, the world's
• 4:23
  first single supercomputer that breaks the gigawatt
• 4:25
  barrier. One massive machine. Right. Aiming for
• 4:28
  1 million GPUs in one place, they're pouring,
• 4:31
  what, over 40 billion dollars into just that
• 4:34
  project. So - Distributed versus consolidated.
• 4:37
  Two different bets on how to win the AI race.
• 4:40
  Exactly. Both incredibly expensive, both demanding
• 4:43
  huge amounts of power, and, you know, the U .S.
• 4:47
  government is watching this closely, framing
• 4:48
  it as vital for American leadership in AI. And
• 4:51
  for people on the ground, OpenAI says these projects
• 4:54
  will create over 25 ,000 on -site jobs. That's
• 4:57
  a significant local impact. Absolutely. It's
• 5:00
  tech infrastructure, but it's also becoming major
• 5:02
  economic development. Okay, so the hardware arms
• 5:04
  race is undeniable. But you need the software
• 5:06
  to make all that power useful, right? Precisely.
• 5:09
  Which brings us to our next story, Alibaba. They're
• 5:12
  making some really aggressive moves in open source
• 5:15
  AI. Yeah, I saw this. They just released their
• 5:17
  Quen3VL Vision models. Correct. And these are
• 5:20
  not small models. The big one, the flagship,
• 5:22
  is 235 billion parameters. That's huge. And it's
• 5:26
  open source. Open source, really. That's a direct
• 5:28
  challenge to guys like Google and their proprietary
• 5:30
  models, isn't it? It absolutely is. Alibaba's
• 5:33
  claiming these models, they have different versions,
• 5:36
  instruct and thinking can match or even beat
• 5:39
  Google's latest Gemini 2 .5 Pro on some key tests
• 5:43
  for perception and reasoning. Bold claim. Yeah.
• 5:46
  But what's the really new thing here? What's
• 5:48
  the aha moment? It's this thing they're calling
• 5:51
  the visual agent capability. This is the game
• 5:55
  changer, I think. Visual agent. Break that down.
• 5:57
  We've had AI that can see images for a while.
• 6:00
  Right, but this goes way beyond just describing
• 6:02
  what's in an image. The visual agent can actually
• 6:04
  operate software. It can understand and interact
• 6:07
  with graphical user interfaces, GUIs on computers
• 6:11
  and mobile devices. Wait, operate them? You mean
• 6:13
  like click buttons, fill in forms, navigate menus?
• 6:16
  Exactly. Imagine you have a repetitive task like
• 6:18
  processing expense reports online. You could
• 6:21
  show the visual agent how to do it once. And
• 6:23
  it could then just... do the next hundred by
• 6:25
  itself. Navigating the website, clicking submit,
• 6:27
  everything. That's the idea. It moves AI from
• 6:29
  being a passive observer to an active participant,
• 6:32
  an autonomous digital worker for certain tasks.
• 6:34
  That's a massive potential productivity boost.
• 6:37
  Okay. Yeah, I can see that being huge for businesses.
• 6:40
  Yeah. There was something about video, too. A
• 6:42
  massive context window. Uh -huh. Up to one million
• 6:45
  tokens. Which in practical terms means you could
• 6:48
  feed this thing like an entire feature -length
• 6:51
  movie a whole movie Yeah, and then ask it super
• 6:54
  specific questions. What was the character wearing
• 6:57
  in the scene at the 53 minute mark or Pinpoint
• 7:01
  the exact second the plot twist was revealed
• 7:03
  It could potentially find that information because
• 7:06
  it can hold and understand the context of hours
• 7:09
  of video. That's incredible. So Why give this
• 7:11
  away? They released it under the Apache 2 .0
• 7:14
  license, right? That's very permissive. It is,
• 7:16
  and that's the strategic brilliance, you could
• 7:18
  argue. By making it open source under Apache
• 7:20
  2 .0, any company can use it, modify it, build
• 7:23
  on top of it commercially without paying Alibaba
• 7:26
  a dime in licensing fees. So they're trying to
• 7:29
  become the default foundation for open source
• 7:31
  vision AI, undercutting the Western proprietary
• 7:34
  options by being free and powerful. That looks
• 7:37
  like the play. They flood the market with a high
• 7:39
  -quality, free alternative. Developers start
• 7:42
  using it, building tools around it. And suddenly,
• 7:45
  Alibaba's tech is embedded everywhere in the
• 7:48
  open -source ecosystem. It's a way to gain global
• 7:50
  influence and market share, even if they also
• 7:53
  have their own massive closed -weight model,
• 7:55
  the Quen 3 Max. Clever. A long game strategy.
• 8:00
  Okay, so we've got the hardware race and the
• 8:02
  software battle lines being drawn. Let's shift
• 8:04
  gears to something maybe a bit more immediate
• 8:06
  for music lovers and creators. Spotify. Yes.
• 8:11
  Good news for DJs, finally. Spotify is bringing
• 8:13
  back direct integration with third -party DJ
• 8:16
  software for premium subscribers. Oh, that's
• 8:18
  huge. They got rid of that back in, what, 2020?
• 8:21
  And people were not happy. No, they weren't.
• 8:23
  It was likely due to tricky music rights issues,
• 8:25
  but it was a major pain point for DJs who had
• 8:27
  built their libraries and playlists in Spotify.
• 8:29
  So what does this mean now? It means if you're
• 8:32
  a premium subscriber in one of the 51 markets
• 8:34
  where this is rolling out, you can access your
• 8:36
  entire Spotify library, all your tracks, your
• 8:40
  playlists directly inside professional DJ software.
• 8:43
  We're talking Rekordbox, Serato, Algorithms DJ.
• 8:48
  The big ones. Wow, okay, so that makes prepping
• 8:50
  sets, mixing tracks, even using features like
• 8:53
  Automix for parties way, way easier if your music
• 8:56
  world revolves around Spotify. Exactly. It just
• 8:59
  removes a ton of friction. It's a big win for
• 9:01
  usability and acknowledges how important that
• 9:03
  creative community is. Connecting this back,
• 9:06
  this isn't just about restoring an old feature,
• 9:08
  is it? There's likely some AI infrastructure
• 9:10
  angle here, too. Well, certainly related. Alongside
• 9:12
  this, Spotify is also rolling out smart filters
• 9:15
  for premium users in seven regions, including
• 9:17
  the U .S. Smart filters. What do they do? They
• 9:19
  let you sort your stuff, music, podcasts, audiobooks,
• 9:22
  not just by genre, but by things like mood or
• 9:25
  activity. So you could filter for, say, funky
• 9:27
  music or party vibes or focus tracks. Ah, OK,
• 9:31
  so that uses AI power categorization, pulling
• 9:34
  from all that data and compute power we talked
• 9:36
  about earlier to help you find the right thing
• 9:38
  faster in their huge catalog. Precisely. The
• 9:41
  massive investment in infrastructure, like those
• 9:43
  Stargate centers, ultimately enables these kinds
• 9:46
  of smarter, more personalized user features on
• 9:49
  the apps you use every day. So the gigawatts
• 9:51
  in the desert help filter my funky playlist at
• 9:54
  home. Got it. That's the connection. Macroscale
• 9:57
  enables the micro features. Really interesting.
• 10:00
  Okay, great updates there. Before we wrap things
• 10:02
  up, let's give a shout out to another great sponsor,
• 10:05
  SCONI. Yeah, that's SCONI. If you need premium
• 10:08
  golf shoes this time, SCONI is the place to look.
• 10:11
  They focus specifically on engineering footwear
• 10:13
  that gives you comfort, style, and performance
• 10:15
  on the course. And guess what? Same code works
• 10:18
  here too. Use SNARFL, S -N -A -R -F -U -L, SNARFL
• 10:22
  at checkout. And for SCONI, that code gets you.
• 10:25
  15 % off your entire purchase. So great shoes,
• 10:28
  great discount. 15%. Can't beat that. Check them
• 10:30
  out at sconi .com. That's S -K -O -N -I dot com.
• 10:34
  Code SNARFL for 15 % off. All right, so let's
• 10:36
  quickly pull the threads together from today.
• 10:38
  Yeah, let's recap. We started with the sheer
• 10:40
  scale of the AI infrastructure build out. OpenAI's
• 10:44
  Stargate. the gigawatt challenge, and partners
• 10:47
  like Oracle basically turning compute into a
• 10:50
  financed utility. Right. Then we looked at the
• 10:53
  software side with Alibaba's aggressive open
• 10:55
  source play, giving away powerful vision models
• 10:58
  like the Quen 3 VL with its visual agent capability
• 11:02
  to try and dominate that landscape. And finally,
• 11:05
  we saw how these big tech undercurrents surface
• 11:07
  in consumer apps. like Spotify bringing back
• 11:10
  crucial DJ integration and rolling out AI -powered
• 11:13
  smart filters, making the user experience better.
• 11:16
  It really shows how interconnected it all is,
• 11:18
  doesn't it? The multi -billion dollar race for
• 11:21
  power and chips directly shapes the features
• 11:23
  you interact with daily. Yeah, the future of
• 11:26
  how we use everything digital is really being
• 11:28
  forged in these massive power -hungry data centers
• 11:30
  right now. It makes you think. What's that? Well,
• 11:33
  you mentioned Oracle acting like a utility provider
• 11:35
  for compute. in Alibaba giving away the software.
• 11:38
  It raises a question, maybe something for you
• 11:40
  to chew on. In this future, who holds the ultimate
• 11:43
  power? Is it the companies designing the AI models
• 11:47
  or the companies controlling the underlying infrastructure,
• 11:50
  the electricity, the cooling, the physical space?
• 11:54
  That's a fantastic question. Is the value in
• 11:56
  the algorithm or in the power plug? Definitely
• 11:58
  something to keep an eye on. We're here to help
• 12:00
  it's narful. So reach out on our website if you
• 12:02
  have any more questions. That's a wrap on today's
• 12:08
  top stories. Remember, tech doesn't have to be
• 12:11
  complicated, and we're here to help connect the
• 12:13
  dots. If you've got questions or want to learn
• 12:16
  more, visit us at snarful .com. Thanks for listening,
• 12:20
  and we'll catch you next time.