Latest / Tech Talks With Kinsoft / Last Week in Tech – OpenAI Breaks Its Microsoft Lock-In, Big Tech's $650B AI Bill, and the Salesforce Breach Wave

Transcript

Last Week in Tech – OpenAI Breaks Its Microsoft Lock-In, Big Tech's $650B AI Bill, and the Salesforce Breach Wave · Tech Talks With Kinsoft

• 0:00
  Right now, big tech is preparing to drop like
• 0:03
  seven hundred billion dollars just to weave artificial
• 0:07
  intelligence into every single piece of software
• 0:10
  you use. Yeah, which is just a staggering amount
• 0:12
  of money. It really is. But at the exact same
• 0:15
  moment, hackers are bypassing these multimillion
• 0:18
  dollar cybersecurity systems using the digital
• 0:21
  equivalent of. Well, a borrowed master key. Right.
• 0:24
  It's this wild juxtaposition. Exactly. So welcome
• 0:28
  to this custom deep dive tailored specifically
• 0:30
  for you. We are pulling all our insights today
• 0:33
  entirely from the May 4, 2026 edition of Kinsoft
• 0:36
  Tech Weekly. And our mission today is to really
• 0:39
  tear into the hidden costs, the massive shifting
• 0:41
  alliances and the structural vulnerabilities
• 0:44
  hiding right in the center of this current tech
• 0:46
  boom. Yeah, because you have this unprecedented
• 0:47
  capital being poured into the physical infrastructure
• 0:50
  of AI, right? Yeah. But it's being completely
• 0:53
  undermined by back end vulnerabilities in the
• 0:55
  enterprise software that we all rely on every
• 0:57
  single day. Right. And we're going to look at
• 0:59
  both sides of that coin today. Okay, let's unpack
• 1:01
  this. Because to understand the security risks
• 1:04
  of our hyper -connected tech world, we first
• 1:07
  have to look at how the absolute biggest players
• 1:09
  are fundamentally rewiring their partnerships.
• 1:12
  Yeah, the whole foundation of cloud computing
• 1:13
  is shifting right now. Exactly. So the Kinsoft
• 1:16
  brief kicks off by highlighting that OpenAI has
• 1:19
  officially loosened its exclusive cloud ties
• 1:21
  with Microsoft, and they're now live on Amazon's
• 1:24
  AWS. Which is huge. I mean, that signals a...
• 1:27
  major structural shift in how businesses are
• 1:30
  going to access these foundational models. Because
• 1:32
  it used to be just locked down, right? Completely.
• 1:34
  For years, the OpenAI and Microsoft Azure partnership
• 1:37
  was like the defining alliance of the generative
• 1:41
  AI space. If you wanted the absolute best model,
• 1:44
  you had to build inside their specific walled
• 1:47
  garden. I'm looking at this through the lens
• 1:49
  of historical tech shifts, I guess. It reminds
• 1:53
  me a bit of the transition from early lockdown
• 1:55
  mainframe environments to open server standards
• 1:58
  or even just, you know, the universal adoption
• 2:01
  of USB. Oh, that's a good way to look at it.
• 2:04
  We're moving from a proprietary ecosystem to
• 2:06
  this interoperable standard. I guess it's sort
• 2:09
  of like the early days of video streaming. What
• 2:10
  do you mean? Well, remember when every hit show
• 2:13
  was locked to one specific platform and you had
• 2:16
  to buy that exact subscription to watch it? Oh,
• 2:18
  right. Yeah. But now it's like the exclusivity
• 2:21
  era is ending. Open AI moving to AWS is like
• 2:24
  a massive blockbuster movie suddenly being available
• 2:27
  on all channels at once. That makes total sense.
• 2:30
  And, you know, what's fascinating here is why
• 2:32
  this matters from a... purely technical standpoint
• 2:34
  for developers. Right. Because why is being on
• 2:37
  AWS such a game changer? Well, it really comes
• 2:39
  down to friction and physics, honestly, specifically
• 2:42
  data gravity and latency. Data gravity. Okay.
• 2:44
  Impact that for me. So let's say you're a massive
• 2:46
  enterprise company, right? And for the last 10
• 2:49
  years, you've built your entire backend database
• 2:52
  on Amazon AWS. All your customer records, transaction
• 2:56
  history, operational data, it all lives there.
• 2:59
  But suddenly, you want to use OpenAI's models
• 3:01
  to analyze all that data. When OpenAI was strictly
• 3:05
  tied to Microsoft's Azure, you had to literally
• 3:07
  route your workflows out of AWS across the public
• 3:11
  internet into Azure, process the query, and then
• 3:14
  send it all the way back. Which I imagine introduces
• 3:16
  latency and probably some really hefty data egress
• 3:19
  fees. Exactly. I mean, every single query takes
• 3:22
  milliseconds, and those add up to real performance
• 3:25
  bottlenecks. Plus, you're paying a premium. just
• 3:28
  to move your own data between competing clouds.
• 3:30
  It's just clunky multi -cloud management. Right.
• 3:32
  So by launching natively on AWS, OpenAI is essentially
• 3:36
  allowing developers to bring the AI model directly
• 3:39
  to where the data already lives. No more cross
• 3:42
  -cloud routing. So it reduces the latency, cuts
• 3:44
  down on the architecture complexity, and just
• 3:46
  removes the friction of deploying top -tier AI.
• 3:50
  Yeah. It transitions AI from being this premium
• 3:52
  add -on for a specific ecosystem into just, well...
• 3:56
  Basic utility infrastructure. You plug into it
• 3:58
  wherever you happen to be building. Which is
• 3:59
  amazing for the user, but that ubiquitous run
• 4:02
  -anywhere infrastructure, I mean, that cannot
• 4:04
  be cheap to build. Oh, not at all. I'm looking
• 4:06
  at the capital expenditure numbers in this Kinsoft
• 4:09
  report, and I swear they look like typos. The
• 4:12
  numbers are staggering. I mean, the combined
• 4:14
  AI spend for the year across the major tech firms,
• 4:18
  we're talking Alphabet, Amazon, and Meta, is
• 4:22
  pushing toward $600 to $700 billion. $700 billion.
• 4:28
  That's with a B. That is not a software budget.
• 4:31
  No. That is literally the gross domestic product
• 4:34
  of a mid -sized country. Right, because they
• 4:36
  aren't just writing code anymore. They are buying
• 4:38
  physical reality. What do you mean by physical
• 4:41
  reality? Well, that $700 billion is going towards
• 4:43
  securing millions of specialized GPUs. They're
• 4:47
  designing and building custom silicon. Constructing
• 4:49
  these massive data center campuses. Right. The
• 4:51
  actual building. Yeah, exactly. And outfitting
• 4:54
  them with specialized liquid cooling systems
• 4:56
  because these new racks run so hot. They're out
• 4:58
  there negotiating directly with regional power
• 5:01
  grids just to get the gigawatts of electricity
• 5:03
  to turn it all on. Wow. It's basically a modern
• 5:06
  industrial revolution, but it's happening in
• 5:08
  server farms. But here is where I'm really confused
• 5:11
  by the market reaction noted in the source, because
• 5:13
  the Kinsoft update points out that when Meta
• 5:16
  announced their increased AI capital spending,
• 5:19
  their shares actually fell on the news. Yeah,
• 5:21
  they took a hit. If these companies are quite
• 5:24
  literally building the physical infrastructure
• 5:26
  of the future and they actually have the cash
• 5:28
  on hand to do it, why are investors panicking
• 5:32
  and punishing Meta's stock instead of celebrating
• 5:34
  the innovation? I mean, isn't the golden rule
• 5:37
  of tech to build the moat before anyone else
• 5:39
  can? It's a completely logical question, but
• 5:42
  it runs headfirst into the brutal timeline of
• 5:45
  Wall Street. Which is what? Next quarter? Exactly.
• 5:47
  It's the tension between capital expenditure,
• 5:49
  CapEx, and return on investment, or ROI. Wall
• 5:53
  Street loves a moat, sure, but they demand profit
• 5:56
  margins today. When Meta or Alphabet say they're
• 5:59
  spending hundreds of billions on data centers,
• 6:01
  investors are immediately doing the math on hardware
• 6:03
  depreciation. Meaning those expensive chips they're
• 6:06
  buying won't last forever. Right. A cutting -edge
• 6:08
  GPU cluster has a shelf life of maybe three to
• 6:11
  five years before it's basically obsolete. Oh,
• 6:13
  wow. Only three to five years. Yeah. So if you
• 6:16
  spend $100 billion today, you have a very narrow
• 6:19
  window to generate a massive return before you
• 6:22
  just have to replace all the hardware again.
• 6:24
  So the investors look at that and say, OK, we
• 6:27
  see the astronomical costs. Now, how exactly
• 6:30
  does a slightly smarter chatbot increase our
• 6:33
  ad revenue this quarter to offset those costs?
• 6:35
  Exactly. They want the payoff right now. But
• 6:38
  enterprise adoption of AI is actually moving
• 6:41
  quite slowly. So there's this huge gap between
• 6:44
  building a trillion dollar infrastructure and
• 6:47
  finding a business model that actually pays for
• 6:49
  it. Precisely. And. If we connect this to the
• 6:52
  bigger picture, this is why this matters directly
• 6:55
  to you, the listener, because tech companies
• 6:58
  are not going to just eat a $700 billion bill
• 7:00
  out of the goodness of their hearts. No, of course
• 7:02
  not. That cost is going to travel downstream.
• 7:04
  The Kinsoft update warns that this level of spending
• 7:06
  dictates the pricing models we are going to see
• 7:09
  over the next two years. Right. So we're looking
• 7:11
  at the end of the subsidized, limitless AI experimentation
• 7:13
  phase. Yep. We're going to see aggressive paywalls,
• 7:17
  subscription price hikes for enterprise software
• 7:19
  across the board, tools that, you know, integrated
• 7:23
  AI for free just to build user habits are going
• 7:26
  to start charging per query. It's inevitable.
• 7:29
  I mean, the infrastructure is just too expensive
• 7:30
  to run at a loss forever. It really is. But this
• 7:33
  massive financial pressure brings us to a really
• 7:35
  critical interception in the report, because
• 7:37
  to justify this $700 billion investment to create
• 7:41
  that ROI. These AI models have to actually be
• 7:45
  useful to businesses. Right. And an AI model
• 7:48
  is totally useless if it doesn't have context.
• 7:51
  It needs to feed on a company's internal data
• 7:53
  to do anything valuable. Which forces companies
• 7:56
  to connect their AI tools directly to their central
• 7:58
  nervous system. Exactly. The AI boom is actively
• 8:02
  requiring companies to build these hyper interconnected
• 8:04
  data architectures. Every single tool must talk
• 8:08
  to every other tool. And that integration explosion
• 8:11
  is actively multiplying. the vulnerabilities
• 8:13
  highlighted in the second half of this report.
• 8:15
  Yeah, this is where things get really scary.
• 8:17
  The source points to a massive ongoing cybersecurity
• 8:19
  crisis. The shiny hunter Salesforce theft campaign.
• 8:23
  Yeah, this is where the physical reality of the
• 8:26
  data centers meets the fragile reality of the
• 8:30
  software layer. According to the Kinsoft update,
• 8:32
  recent victims of this campaign include some
• 8:35
  major entities. Amtrak had millions of customer
• 8:38
  records exposed. Pitney Bowes and Vimeo were
• 8:41
  also hit. Huge names. But the fascinating and
• 8:43
  honestly terrifying part is how the threat actors
• 8:46
  got in. These companies were not hacked directly.
• 8:49
  Their primary authentication wasn't breached
• 8:51
  at all. They were hit through connected third
• 8:54
  -party access to their CRM data. Right. And we
• 8:57
  really have to stop thinking of a CRM like Salesforce
• 8:59
  as just a digital address book. A modern CRM
• 9:03
  operates as the operational hub for enterprise
• 9:05
  data. Right. It's everything. It handles billing,
• 9:08
  marketing pipelines, customer communications,
• 9:10
  and now it's the primary data feed for enterprise
• 9:12
  AI endpoints. To make all of those functions
• 9:15
  work automatically, the CRM has to constantly
• 9:18
  talk to dozens, sometimes hundreds, of external
• 9:21
  applications. And the way they talk to each other
• 9:23
  is through application programming interfaces,
• 9:26
  right? APIs. Exactly. When you connect, say,
• 9:29
  a third -party marketing analytics tool to your
• 9:32
  CRM, you don't give the marketing tool your actual
• 9:36
  password. You generate an API token, like a digital
• 9:40
  trusted key. That token allows the third -party
• 9:43
  app to completely bypass primary authentication
• 9:46
  and just pull the data it needs continuously,
• 9:49
  machine to machine. Okay, it's like building
• 9:52
  an impenetrable billion -dollar fortress to protect
• 9:55
  your most valuable assets. You have multi -factor
• 9:57
  authentication, biometric scanners, lasers, a
• 9:59
  full security team. Right. But... You hire a
• 10:02
  guy to come in twice a week to water the decorative
• 10:05
  plants in the lobby. And just to make his job
• 10:07
  easier, you give him a master key that bypasses
• 10:10
  the biometric scanners at the back door. Is a
• 10:12
  perfect way to visualize it. The hackers, they
• 10:14
  do not spend months trying to defeat your multi
• 10:16
  -factor authentication. They just wait until
• 10:19
  the plant guy connects to a public Wi -Fi network
• 10:21
  at a coffee shop and they steal his key. You
• 10:23
  didn't get hacked. Your plant guy got hacked.
• 10:25
  But the thieves still get to walk right into
• 10:27
  your vault completely undetected because the
• 10:29
  system recognizes the key as legitimate. And
• 10:33
  that token bypasses all the expensive security
• 10:35
  layers you put on the front door. This exposes
• 10:38
  the fundamental structural flaw of modern business
• 10:41
  technology. Because the CRM is specifically designed
• 10:44
  to integrate to be useful, every single one of
• 10:47
  those API tokens is a potential entry point.
• 10:50
  Wow. The very interconnectedness that makes these
• 10:53
  cloud tools so powerful is exactly what makes
• 10:56
  them fragile. If a hacker compromises a popular
• 10:59
  third -party analytics plugin, they don't just
• 11:02
  hack the plugin company, they instantly gain
• 11:04
  lateral movement into the CRM data of every single
• 11:07
  enterprise that installed that plugin. It's a
• 11:09
  cascading failure. One trusted integration can
• 11:12
  spill data across dozens of companies, and the
• 11:14
  blast radius of this vulnerability is staggering.
• 11:17
  The Kinsoft brief proves this isn't just a Salesforce
• 11:20
  issue and it isn't an isolated incident in the
• 11:22
  U .S. No, it's global. Right. This is a systemic
• 11:25
  architectural problem. They mention a Hungarian
• 11:28
  media giant that suffered a massive breach through
• 11:31
  similar vectors. Eight and a half terabytes of
• 11:34
  payroll and financial files were stolen and leaked.
• 11:36
  You really have to stop and consider what eight
• 11:38
  and a half terabytes of text and spreadsheets
• 11:40
  actually represents. It's hard to even picture.
• 11:43
  Right. Because that is not just a list of emails.
• 11:45
  That is years of operational history. Millions
• 11:48
  of highly sensitive financial documents, employee
• 11:51
  salaries, internal questions. corporate accounts,
• 11:53
  banking details, all ripped out through a back
• 11:56
  -end door. Just massive amounts of data. And
• 11:59
  the collateral damage isn't just financial. The
• 12:01
  source also notes a Swiss radiology network that
• 12:04
  was hit by ransomware through similar vectors.
• 12:07
  Yeah, that one is rough. We are talking about
• 12:08
  deeply personal, highly sensitive medical imaging
• 12:11
  and patient diagnostics locked down and held
• 12:13
  hostage. And the recurring thread tying all these
• 12:16
  global attacks together is connected apps and
• 12:18
  CRM data. The threat actors have realized that
• 12:21
  targeting the primary... enterprise is the hard
• 12:23
  way in. Targeting the forgotten, under -secured
• 12:27
  third -party integration is the easy way in.
• 12:29
  So what does this all mean? We have a tech industry
• 12:32
  demanding we connect everything to AI to justify
• 12:35
  their $700 billion hardware spend, and an ecosystem
• 12:39
  where those very connections are actively exposing
• 12:42
  healthcare networks and terabytes of financial
• 12:45
  data. It's a massive contradiction. How do businesses
• 12:47
  actually protect themselves without just, you
• 12:50
  know, unplugging from the internet completely?
• 12:53
  The mandate cannot just be go back to 1995. No,
• 12:56
  you can't go back to pen and paper. but you also
• 12:58
  cannot operate with blind trust in the ecosystem
• 13:01
  anymore. This raises an important question about
• 13:04
  security posture. The core actionable advice
• 13:07
  from the Kinsaw source revolves around adopting
• 13:10
  a zero -trust mindset for your integrations.
• 13:13
  Okay, zero -trust. It means staying patched,
• 13:15
  obviously, but more importantly, staying deeply
• 13:17
  skeptical of your own architecture. Staying skeptical,
• 13:20
  basically assuming the plant guy is eventually
• 13:22
  going to lose his key. Precisely. You have to
• 13:25
  operate under the assumption that your third
• 13:27
  -party vendors will eventually be compromised.
• 13:29
  It's just a matter of when. Because API sprawl
• 13:43
  is a very real thing. Oh, absolutely. A marketing
• 13:46
  team will install a specific app for a three
• 13:49
  -month campaign, grant it full read -write access
• 13:52
  to the entire Salesforce database, and then completely
• 13:56
  forget to uninstall it when the campaign ends.
• 13:58
  That happens all the time. And then five years
• 14:00
  later, that abandoned app gets bought by some
• 14:03
  shady holding company or its servers get breached,
• 14:05
  and that trusted key is still sitting there completely
• 14:08
  active. That forgotten app is a ticking time
• 14:10
  bomb. The mitigation strategy is active auditing.
• 14:13
  You have to review those integrations regularly
• 14:15
  and relentlessly revoke access for anything that
• 14:18
  is not mission critical today. You must minimize
• 14:21
  the machine to machine doors into your core data.
• 14:24
  The brief notes that if you have never formally
• 14:26
  audited your CRMs connected apps, visiting kinsoft
• 14:30
  .com .au is a highly recommended starting point
• 14:33
  to map out how to get that process moving. It's
• 14:35
  the digital equivalent of revoking old building
• 14:38
  passes. But instead of stopping a former employee
• 14:41
  from like using the corporate. at gym, you are
• 14:44
  closing a back door that could cost you millions
• 14:46
  in liability. It is absolute essential hygiene
• 14:49
  for this era of cloud computing. So bringing
• 14:52
  all these threads together, we have covered some
• 14:54
  massive structural shifts today. We started with
• 14:57
  the physical layer, the unprecedented $700 billion
• 14:59
  CapEx race to build the infrastructure of the
• 15:03
  AI era. Right. And how open AI shift to AWS signals
• 15:06
  a move toward ubiquitous multi -cloud standardizations.
• 15:10
  But we also unpacked the tension of that reality,
• 15:12
  right? How the pressure for ROI on that hardware
• 15:15
  is going to force aggressive monetization onto
• 15:17
  the consumer. Definitely. And how the drive to
• 15:19
  connect enterprise data to... these AI models
• 15:21
  is creating an absolute golden age for cybercriminals.
• 15:24
  The Shiny Hunters campaign is the perfect case
• 15:26
  study for that. Bypassing primary defenses by
• 15:29
  targeting the trusted API tokens of third -party
• 15:31
  CRM integrations. Ripping through Amtrak, Vimeo,
• 15:35
  and global healthcare networks. Not by breaking
• 15:38
  down the walls, but by using borrowed keys. Which
• 15:41
  leaves you with a very clear, immediate takeaway.
• 15:44
  Do not wait for a third -party vendor breach
• 15:46
  to dictate your Friday night. Audit your CRM's
• 15:49
  connected apps today. Find out exactly who holds
• 15:52
  the digital keys to your data environment and
• 15:54
  revoke access for any integration that doesn't
• 15:57
  strictly need it to function. Yeah, you have
• 15:59
  to be proactive. And as the source mentioned,
• 16:01
  kinsoft .com .au is the place to start if you
• 16:04
  are staring at your dashboard wondering how to
• 16:05
  even untangle that web. It really is the only
• 16:08
  way to manage the risk. Yeah. But, you know,
• 16:10
  looking at the sheer scale of both the spending
• 16:12
  and the vulnerability. This leaves me with a
• 16:14
  final thought I want to pose to you, the listener.
• 16:16
  Let's hear it. We are watching tech giants spend
• 16:19
  nearly a trillion dollars to weave artificial
• 16:22
  intelligence and connected APIs into every single
• 16:25
  layer of our professional lives. The default
• 16:28
  assumption is that the future is hyperconnected.
• 16:30
  But as these API integrations become exponentially
• 16:33
  riskier and the financial cost of securing and
• 16:36
  maintaining them skyrockets, are we fast approaching
• 16:39
  a breaking point? Oh, that's interesting. Right.
• 16:42
  Will we soon see a future where major enterprises
• 16:44
  run the math and decide that the risks and costs
• 16:47
  of being hyperconnected are just too high? Could
• 16:50
  the next big contrarian trend in tech actually
• 16:53
  be companies deliberately disconnecting their
• 16:56
  systems, pulling their data off the public cloud,
• 16:59
  and intentionally siloing their operations just
• 17:02
  to survive? Wow. An intentional retreat from
• 17:05
  the cloud, rebuilding the physical air -gapped
• 17:07
  walls because the digital API doors are just
• 17:10
  too fundamentally dangerous to manage. Exactly.
• 17:12
  That completely upends the narrative of the last
• 17:14
  decade, and it is a fascinating thing to ponder
• 17:16
  as we watch these tech giants build their massive
• 17:19
  data. We will leave you with that thought to
• 17:21
  chew on. Thank you so much for joining us on
• 17:23
  this custom deep dive into the hidden currents
• 17:26
  of the 2026 tech landscape. Stay patched, stay
• 17:29
  skeptical, and mind your digital keys.