Latest / Tech Talks With Kinsoft / Last Week in Tech – The AI Boom's Memory-and-Power Bill, Microsoft's 4,800 Cuts, and a Perfect-10 ColdFusion Flaw

Transcript

Last Week in Tech – The AI Boom's Memory-and-Power Bill, Microsoft's 4,800 Cuts, and a Perfect-10 ColdFusion Flaw · Tech Talks With Kinsoft

• 0:00
  You know when you flip a light switch in your
• 0:01
  living room. You just you expect the light to
• 0:04
  turn on. Right. You definitely aren't thinking
• 0:06
  about the power grid. Exactly. You don't stand
• 0:08
  there visualizing the power plant 50 miles away
• 0:10
  or like the high voltage transmission lines cutting
• 0:13
  through the mountains. You just want the room
• 0:15
  to be bright. Yeah. We've been completely conditioned
• 0:17
  to expect the end result without ever really
• 0:20
  having to think about the mechanism behind it.
• 0:22
  And that expectation has just well, it's permeated
• 0:25
  how we interact with all modern technology. Oh,
• 0:27
  100 percent. Yeah. The infrastructure is by design
• 0:30
  completely. invisible to you. Which is exactly
• 0:33
  the illusion we are shattering today. Because
• 0:36
  we open our laptops, pull out our phones, type
• 0:39
  some complex query into an AI chat bot, and we
• 0:43
  just expect this highly synthesized magical answer
• 0:46
  in milliseconds. And you don't see the massive,
• 0:49
  humming, incredibly strained physical reality
• 0:52
  operating behind the scenes to make that happen.
• 0:55
  Right. So today we're taking a stack of really
• 0:58
  critical tech developments from mid -July 2026,
• 1:01
  primarily anchored by this brilliant briefing
• 1:04
  called Tech Talks with Kinsoft. And we're going
• 1:07
  to actually look at the mechanism. Yeah, we aren't
• 1:08
  just running down a list of tech news today.
• 1:10
  No, not at all. The mission for this deep dive
• 1:13
  is to explore the hidden infrastructure of the
• 1:16
  technology you rely on every single day. We're
• 1:19
  looking at the physical constraints, the ruthless
• 1:21
  corporate maneuvers, the regulatory crackdowns,
• 1:24
  and honestly, the terrifying speed of modern
• 1:27
  security threats. Because, I mean, if you are
• 1:30
  managing any kind of technology stack or business
• 1:32
  strategy right now, understanding that hidden
• 1:35
  infrastructure is really the only way you can
• 1:37
  accurately map out your own risk. And your budget,
• 1:39
  for that matter. So true. The reality of tech
• 1:42
  is just far more physical and far more vulnerable
• 1:45
  than a seamless user interface implies. Absolutely.
• 1:48
  So let's start at the literal foundation. Before
• 1:50
  any software can execute, before any large language
• 1:52
  model can write a line of code or generate an
• 1:55
  image, it fundamentally requires hardware and
• 1:58
  electricity. Yeah, you can't code your way out
• 2:00
  of needing a power source. Exactly. And based
• 2:02
  on three separate financial announcements this
• 2:04
  week, an enormous physical bill is coming due
• 2:08
  for the AI boom. Oh, it's an infrastructure build
• 2:11
  out on a scale we really haven't seen in decades.
• 2:13
  The numbers are almost difficult to process.
• 2:15
  They really are. Let's look at Samsung's preview
• 2:17
  of their second quarter results from July 7th.
• 2:20
  They're projecting an operating profit of roughly
• 2:22
  89 trillion Korean won. Which is what, about
• 2:26
  58 billion U .S. dollars? Yeah, almost 60 billion.
• 2:30
  And that profit is up nearly 19 -fold from a
• 2:32
  year earlier. 19 times. That's just wild. It's
• 2:36
  the highest quarterly profit in their entire
• 2:38
  corporate history. And this isn't from selling
• 2:40
  more TVs or smartphones. It is almost entirely
• 2:42
  driven by the explosive demand for specialized
• 2:45
  high -bandwidth memory chips, or HBM for AI data
• 2:49
  centers. And then a few days later, SK Hynix
• 2:52
  debuts on the U .S. NASDAQ exchange, pulling
• 2:55
  in around $29 billion specifically to fund new
• 2:59
  chip fabrication and packaging plans. Yeah, that
• 3:01
  SK Hynix debut is one of the largest listings
• 3:04
  the U .S. market has ever seen. Massive. It is.
• 3:07
  And just to put some technical context behind
• 3:09
  why these specific chips are generating $58 billion
• 3:12
  in profit for Samsung, it really comes down to
• 3:16
  how AI actually computes. Okay, wait on us. So
• 3:19
  traditional computing is largely sequential.
• 3:21
  But training an AI model requires parallel processing,
• 3:25
  meaning you have to shuffle massive, unprecedented
• 3:27
  data sets back and forth between the processor
• 3:30
  and the memory simultaneously. So a standard
• 3:32
  chip just can't keep up. Exactly. Standard memory
• 3:34
  chips create a traffic jam. High bandwidth memory
• 3:37
  is essentially widening a two -lane highway into
• 3:41
  a hundred -lane superhighway. AI data centers
• 3:44
  literally cannot function without it. Wow. And
• 3:47
  then there's the power required to actually run
• 3:49
  that superhighway. Britain's National Grid announced
• 3:52
  a $1 .75 billion investment into a U .S. power
• 3:55
  developer to build a 2 .5 gigawatt facility.
• 3:59
  That's a lot of juice. And here's a kicker. Its
• 4:01
  entire purpose is to feed a single Microsoft
• 4:04
  data center campus in Texas. OK, let's unpack
• 4:06
  this. A typical nuclear power plant generates
• 4:09
  about one gigawatt of power. So we are talking
• 4:11
  about dedicating two and a half nuclear power
• 4:13
  plants worth of electricity to a single cluster
• 4:16
  of buildings. It's like everyone is admiring
• 4:18
  a shiny new sports car, the AI chatbots. But
• 4:22
  nobody is noticing that the industry is fundamentally
• 4:24
  running out of the asphalt and gasoline required
• 4:27
  to actually drive them. That's a great way to
• 4:30
  put it. The constraints have entirely shifted
• 4:32
  for the last 20 years. The tech industry's bottleneck
• 4:35
  was software engineering, right? Like, how fast
• 4:37
  can you write the code? Today, the bottleneck
• 4:40
  is down in the physical layer. Can you secure
• 4:42
  the silicon and can you secure the megawatts?
• 4:45
  So let's bring this down to the ground level
• 4:46
  for you, the listener. If I'm an IT director
• 4:49
  mapping out my three -year hardware budget or
• 4:51
  a CTO planning a major infrastructure overhaul,
• 4:55
  why does a giant Texas data center affect me?
• 4:58
  How does this massive supply suction at the top
• 5:01
  actually hit my spreadsheets? Well, it hits your
• 5:03
  spreadsheets in two very distinct ways, cost
• 5:06
  and time. These massive data center builders,
• 5:09
  you know, your Microsoft, Google's, Amazon's,
• 5:11
  they are quite literally hoovering up the global
• 5:13
  supply of raw materials. Mostly memory and silicon.
• 5:16
  Exactly. When buyers at that scale are absorbing
• 5:19
  all the inventory, the trickle -down effect for
• 5:22
  enterprise hardware is severe. So regular businesses
• 5:25
  pay the price. Yeah, you need to adjust your
• 5:28
  forecasting immediately. Expect memory to be
• 5:31
  significantly pricier across all your server
• 5:33
  and endpoint purchases, and expect lead times
• 5:36
  for network equipment to stretch out. Oh, that
• 5:38
  makes sense. Yeah, if you used to refresh your
• 5:40
  hardware every four years, you honestly need
• 5:43
  to start the procurement process at year three
• 5:44
  now. So the physical constraints of AI are going
• 5:48
  to tax the budgets of companies that aren't even
• 5:50
  building AI. Precisely. And it also changes how
• 5:53
  you evaluate vendor promises. When an enterprise
• 5:56
  software vendor tells you a revolutionary new
• 5:59
  AI feature is coming, like next quarter, you
• 6:01
  have to realize that the honest barrier to that
• 6:04
  release isn't necessarily their software team
• 6:06
  anymore. It's whether they have the server space.
• 6:09
  Exactly. It's whether their cloud provider has
• 6:11
  the physical server capacity to host it. So when
• 6:13
  product roadmaps slip this year, the physical
• 6:16
  limit of the grid is very likely the hidden reason
• 6:19
  why. has always preached this gospel of infinite
• 6:23
  digital scaling. And now they were just violently
• 6:26
  hitting the hard limits of the physical world.
• 6:28
  They really are. But this pivot to AI isn't just
• 6:32
  sucking up electricity and silicon. It is actively
• 6:35
  cannibalizing other departments within these
• 6:37
  very same tech giants. We are moving from the
• 6:40
  physical reallocation of resources to the human
• 6:43
  reallocation. Which brings a completely different
• 6:45
  kind of instability to the enterprise landscape,
• 6:48
  particularly regarding the vendors that almost
• 6:50
  every business relies on. Yeah, let's look at
• 6:52
  Microsoft. On July 6th, they cut around 4 ,800
• 6:55
  jobs, which is a bit over 2 % of their workforce.
• 6:59
  The gaming division took the absolute heaviest
• 7:01
  blow. Xbox shed roughly 20 % of its staff, and
• 7:05
  they sold off or spun off several studios. That's
• 7:07
  a huge cut for one division. Yeah, the internal
• 7:09
  memo from Xbox leadership was incredibly blunt.
• 7:12
  They stated, our business today is not healthy.
• 7:16
  But they didn't just cut gaming. Microsoft used
• 7:19
  this round of layoffs to fundamentally reshape
• 7:21
  its commercial sales and consulting organizations.
• 7:24
  Here's where it gets really interesting. Microsoft
• 7:26
  just posted quarter after quarter of record -breaking
• 7:29
  revenue. Right, they're making money hand over
• 7:31
  fist. Exactly. This is not a company in distress
• 7:33
  trying to avoid bankruptcy. It's like aggressively
• 7:36
  pruning a perfectly healthy, mature oak tree
• 7:39
  just so one specific branch AI and cloud can
• 7:43
  grow to a massive, monstrous size. What's fascinating
• 7:46
  here is that you're watching a deliberate, ruthless
• 7:48
  reallocation of capital in real time. In a high
• 7:52
  interest rate environment, even a giant like
• 7:54
  Microsoft doesn't have infinite cash to throw
• 7:57
  at every project. Right. So they are looking
• 7:59
  at their portfolio, pulling money, attention
• 8:00
  and highly skilled personnel out of slower growing
• 8:03
  legacy areas and pouring absolutely everything
• 8:06
  into AI infrastructure. They are reshaping the
• 8:09
  entire corporate organism to win one specific
• 8:11
  arms race. But if I am an enterprise customer.
• 8:14
  relying on Microsoft's commercial consulting
• 8:16
  arm to architect my cloud migration, what is
• 8:19
  the ripple effect of that internal pruning on
• 8:22
  my business? The ripple effect is a sudden loss
• 8:25
  of institutional knowledge and relationship continuity.
• 8:28
  Enterprise technology at scale is well, it's
• 8:31
  heavily reliant on relationships. Yeah, the people
• 8:33
  you talk to. Exactly. When a vendor fundamentally
• 8:36
  restructures its commercial sales and support
• 8:38
  teams to change a new strategic priority, the
• 8:41
  trusted account manager who understands your
• 8:44
  specific compliance needs might just vanish.
• 8:47
  The Tier 3 support structure you're used to navigating
• 8:50
  can be completely rewritten overnight. Wow, so
• 8:53
  you can't just assume the organizational chart
• 8:55
  of your vendor is a... static resource. Exactly.
• 8:58
  You need to map out your dependencies now, know
• 9:00
  exactly who your critical contacts are, and have
• 9:03
  a clear documented escalation path for how to
• 9:06
  get enterprise support if your entire account
• 9:08
  team is suddenly reallocated to an AI initiative,
• 9:11
  or just manage out of the company entirely. Now,
• 9:14
  with tech giants ruthlessly consolidating power,
• 9:18
  money, and focus to dominate this new era of
• 9:21
  cloud computing, you'd think they're entirely
• 9:23
  unstoppable. But that kind of aggressive consolidation
• 9:26
  makes governments incredibly nervous. Oh, definitely.
• 9:30
  Regulators are pushing back against monopolistic
• 9:32
  behaviors and they are finally proving they can
• 9:35
  draw blood. Yeah. The regulatory landscape, particularly
• 9:38
  in the European Union, has crossed a very significant
• 9:41
  threshold this month. Right. So on July 2nd,
• 9:43
  the European Union's highest court threw out
• 9:46
  Google's final appeal against a massive antitrust
• 9:49
  fine. We are talking about 4 .1 billion euros.
• 9:53
  So roughly 4 .7 billion U .S. dollars. Huge number.
• 9:57
  The core issue was how Google bundled its search
• 10:00
  app and Chrome browser onto Android phones, essentially
• 10:03
  forcing device manufacturers to pre -install
• 10:05
  them if they wanted access to the Google Play
• 10:07
  Store. Which is a textbook example of leveraging
• 10:10
  dominance in one market, the Android operating
• 10:12
  system, to choke out competition in another market
• 10:14
  like search and web browsing. Just by making
• 10:16
  it the default. Exactly. By forcing manufacturers
• 10:20
  to set Chrome and Google search as the default
• 10:23
  out of the box, they capitalized on user inertia.
• 10:27
  most users simply will not change the default
• 10:29
  browser on a new phone i know i don't right so
• 10:32
  that effectively locked competing search engines
• 10:35
  out of the mobile ecosystem before the user even
• 10:38
  turned the device on okay i understand the mechanics
• 10:40
  of the monopoly argument there but i have to
• 10:43
  push back on the impact 4 .7 billion dollars
• 10:46
  is a staggering number to you and me But for
• 10:49
  a titan like Alphabet, a company that generates
• 10:52
  tens of billions of dollars in profit every single
• 10:55
  quarter, isn't this ultimately just an expensive
• 10:58
  parking ticket? Like, hasn't this traditionally
• 11:00
  been the acceptable cost of doing business for
• 11:02
  big tech? You break a rule, you capture a market,
• 11:05
  you pay a fine a decade later, and you keep your
• 11:07
  dominance. So that has historically been the
• 11:09
  playbook, yeah. But this ruling signals that
• 11:11
  the parking ticket era is over. The dollar amount
• 11:14
  grabs the headlines, but the legal finality is
• 11:17
  what actually matters here. This specific case
• 11:19
  has been dragging through the appeals process
• 11:21
  since 2018. Google employed some of the most
• 11:25
  expensive legal teams on earth to fight this,
• 11:28
  and they lost their final appeal. The courts
• 11:30
  are firmly... undeniably backing the regulators.
• 11:34
  So the strategy of tying up regulators in infinite
• 11:36
  litigation is failing. It is failing and the
• 11:39
  fines are sticking. For you, the listener, this
• 11:42
  fundamentally changes the risk calculus. If your
• 11:45
  organization operates in Europe or if you're
• 11:47
  developing a product strategy that relies on
• 11:49
  aggressively bundling software to force user
• 11:51
  adoption, you have to recognize that the regulatory
• 11:54
  bodies now have real unavoidable teeth. You can't
• 11:57
  just stall them forever. Exactly. You can no
• 11:59
  longer assume that your legal department can
• 12:01
  just litigate away anti -competitive behavior.
• 12:04
  The blueprint for prosecuting big tech in the
• 12:07
  EU is now finalized and legally binding. So while
• 12:10
  these mega corporations are fighting regulators
• 12:12
  over market control in the courts, the actual
• 12:14
  day -to -day software infrastructure that underpins
• 12:17
  all of these companies is constantly crumbling
• 12:19
  under our feet. Yeah, the velocity of the security
• 12:22
  landscape right now is frankly terrifying. It
• 12:25
  really is. Yeah. Let's talk about Adobe ColdFusion.
• 12:27
  On June 30th, Adobe patched a batch of flaws
• 12:30
  in this web application platform. One of them,
• 12:33
  cataloged as CVE -2026 -48282, scored a perfect
• 12:38
  10 out of 10 on the CVSS severity scale. Which
• 12:41
  is as bad as it gets. Right. For context, a 10
• 12:44
  out of 10 usually means an attacker requires
• 12:45
  zero authentication, no username, no password,
• 12:48
  to remotely execute their own malicious code
• 12:50
  on your server from across the Internet. And
• 12:53
  according to the Kinsoft briefing, security researchers
• 12:55
  observed attackers actively exploiting this specific
• 12:58
  flaw within two hours of Adobe publicly disclosing
• 13:01
  it. Two hours. That is just two hours. The U
• 13:03
  .S. cybersecurity agency CISA immediately added
• 13:06
  it to their known exploited vulnerabilities catalog
• 13:08
  and ordered all federal agencies to patch it.
• 13:10
  A 10 out of 10 severity weaponized and actively
• 13:13
  exploited in two hours. That is the brutal reality
• 13:16
  of modern cyber warfare. So what does this all
• 13:19
  mean? Because I am looking at that two hour window.
• 13:22
  And the operational math just doesn't work for
• 13:24
  most IT departments. It's like a burglar finding
• 13:27
  a highly detailed blueprint to your house and
• 13:29
  then robbing it before the ink on the blueprint
• 13:32
  is even dry. That's a really good analogy. If
• 13:35
  an exploit is live in two hours, the traditional
• 13:38
  enterprise mindset of we'll test the patch and
• 13:40
  deploy it during the maintenance window next
• 13:42
  weekend. is literally a death sentence for your
• 13:44
  network. Oh, it completely invalidates legacy
• 13:47
  patching schedules. And what makes cold fusion
• 13:49
  particularly dangerous is that it's often legacy
• 13:52
  middleware. Meaning it's kind of forgotten about.
• 13:55
  Exactly. It's the connective tissue lurking behind
• 13:57
  older, half -forgotten web applications that
• 14:00
  no one on the current IT staff actually built.
• 14:02
  If you run software like that anywhere on your
• 14:05
  perimeter, you have to automate your patching.
• 14:07
  If you genuinely cannot patch a 10 out of 10
• 14:10
  vulnerability within hours of disclosure, Your
• 14:12
  only other option is to sever that server from
• 14:15
  the public Internet until you can. The grace
• 14:18
  period between disclosure and disaster is gone.
• 14:20
  And it's not just older middleware getting hit.
• 14:22
  The giants themselves are bleeding. The briefing
• 14:26
  notes that on July 7th, the massive global consulting
• 14:29
  and IT services firm Accenture confirmed they
• 14:33
  suffered a security incident. A cyber criminal.
• 14:36
  on a dark web forum claimed to be selling a massive
• 14:39
  cache of stolen Accenture data. Yeah, I saw that.
• 14:42
  Now, Accenture stated the matter was isolated,
• 14:45
  that they immediately remediated the source,
• 14:48
  and that there was no impact on their operations
• 14:51
  or their clients' systems. Okay, if we connect
• 14:54
  this to the bigger picture, we need to model
• 14:56
  some critical thinking regarding how we consume
• 14:58
  threat intelligence. Right, because there's always
• 15:00
  a lot of noise. Yes. When dealing with a high
• 15:03
  -profile breach notification, you have to surgically
• 15:06
  separate the confirmed facts from the threat
• 15:08
  actor's claims. I see the distinction, because
• 15:10
  the cybercriminal claimed they had extracted
• 15:12
  35 gigabytes of data, specifically highlighting
• 15:15
  that they had internal source code and a batch
• 15:17
  of cloud access keys. Precisely. The confirmed
• 15:20
  fact is that a perimeter breach occurred at Accenture.
• 15:24
  The unverified claim is the 35 gigabytes of source
• 15:27
  code and keys. The journalists who broke the
• 15:30
  story couldn't verify the full scope of that
• 15:32
  data dump. So they might just be bluffing to
• 15:34
  drive up the price. Exactly. And this is the
• 15:38
  crucial takeaway for the listener. The core lesson
• 15:40
  holds, regardless of whether it was 35 gigabytes
• 15:43
  or 35 megabytes. Okay, let's pause and clarify
• 15:46
  the technical threat there. What? When we say
• 15:48
  cloud access keys, we are basically talking about
• 15:51
  machine -to -machine master passwords, right?
• 15:53
  Pretty much, yeah. These are tokens that allow
• 15:55
  automated scripts or software to bypass human
• 15:57
  multi -factor authentication and go directly
• 16:00
  to cloud infrastructure. That's exactly it. They
• 16:03
  are the keys to the kingdom. And if a global
• 16:05
  giant with immense world -class security resources
• 16:08
  like Accenture can potentially leak source code
• 16:10
  and access tokens, your third -party suppliers
• 16:13
  absolutely can too. Wow. Yeah. It is the ultimate
• 16:17
  supply chain risk. Think of it like giving a
• 16:19
  valet the keys to your corporate vehicle. You
• 16:22
  might have the best car alarm in the world, but
• 16:24
  if the valet company's unmonitored lockbox gets
• 16:26
  pried open, your car is still getting stolen.
• 16:29
  Man. So if I am a CTO, how do I defend my architecture
• 16:33
  against a breach that happens on my supplier's
• 16:36
  network? I can't patch Accenture servers. You
• 16:39
  know, you can't. You build defense in depth on
• 16:41
  your own side of the wall. The defensive moves
• 16:43
  are entirely unglamorous, but they are essential.
• 16:46
  First, you mandate strict automated rotation
• 16:49
  of your cloud keys and access tokens. So they
• 16:51
  expire quickly. Right. If a token gets stolen
• 16:54
  in a third -party supply chain breach, you want
• 16:56
  to ensure it expires before the attacker can
• 16:58
  figure out what it unlocks. Second, you lock
• 17:01
  down your code repositories. Implement strict
• 17:03
  identity access management so that only specific...
• 17:09
  That makes a lot of sense. And finally, review
• 17:12
  your vendor contracts today. Ask your major IT
• 17:15
  suppliers point blank what their breach notification
• 17:17
  protocols are. If their lockbox gets pried open,
• 17:21
  how many hours will pass before they notify you?
• 17:23
  You must secure those answers before an incident
• 17:26
  happens, not while you are reading about it on
• 17:28
  a security blog. Okay, so assume the supply chain
• 17:31
  breach is inevitable and aggressively limit the
• 17:33
  shelf life of what can be stolen. is highly actionable.
• 17:37
  Now, we've talked about billion dollar physical
• 17:39
  power constraints, 10 out of 10 software vulnerabilities
• 17:42
  and complex supply chain breaches. But our final
• 17:46
  story highlights that the most consistently successful
• 17:48
  attack vector on the planet doesn't rely on cutting
• 17:51
  edge zero day code at all. No, it doesn't. It
• 17:53
  relies entirely on human psychology. Exactly.
• 17:56
  The human mind remains the most exposed, unauthenticated
• 17:59
  attack surface in any enterprise. On July 9th,
• 18:02
  Interpol announced the results of an operation
• 18:04
  called First Light 2026. This was huge. It was
• 18:07
  a massive, highly coordinated law enforcement
• 18:09
  crackdown spanning 97 countries. They made over
• 18:12
  1 ,500 arrests globally and intercepted $293
• 18:16
  million in illicit stolen money. Just an incredible
• 18:20
  haul. And what kind of high -tech digital wizardry
• 18:23
  were these syndicates using? It wasn't remote
• 18:25
  code execution. It was everyday scams, phishing,
• 18:27
  romance cons, investment fraud, and overwhelmingly
• 18:30
  business email compromise. or BEC. Yeah, BEC
• 18:34
  is a nightmare. This is where a crook impersonates
• 18:37
  a CEO or a trusted vendor, creates a false sense
• 18:40
  of urgency, and simply asks an employee to route
• 18:43
  a payment to a fake invoice. $293 million intercepted
• 18:47
  globally, fundamentally sourced from variations
• 18:49
  of criminals just politely or urgently asking
• 18:52
  people to hand over their money. Which brings
• 18:54
  up a fascinating philosophical contradiction
• 18:56
  to me. With all this incredibly complex technology
• 18:59
  we've explored today, the AI data centers, the
• 19:01
  multi -cloud architectures, the sophisticated
• 19:04
  software flaws, why is rudimentary social engineering
• 19:07
  still the undisputed workhorse of global fraud?
• 19:10
  Are we, as humans, just the unpatchable bug in
• 19:12
  the system? This raises an important question
• 19:14
  about how threat actors actually operate. And
• 19:17
  the answer is rooted in pure economics. Hacking
• 19:20
  is a business. Right. They want the best ROI.
• 19:23
  Exactly. Tricking a stressed, multitasking executive
• 19:27
  into paying a fraudulent invoice is fundamentally
• 19:30
  easier, cheaper, and yields a higher return on
• 19:33
  investment than spending six months trying to
• 19:36
  brute force a corporate firewall. That makes
• 19:38
  total sense. And these BEC attacks have evolved
• 19:40
  too. They aren't just spoofing email addresses
• 19:42
  anymore. They are compromising legitimate internal
• 19:45
  email threads, watching the cadence of the conversation
• 19:48
  and inserting the fake invoice at the exact moment
• 19:51
  a payment is expected. inherently want to be
• 19:54
  helpful. They want to resolve urgent issues quickly
• 19:56
  and attackers ruthlessly weaponize that empathy
• 19:59
  and urgency. Wow. So if you can't push a software
• 20:02
  update to the human brain, what is the defense?
• 20:05
  How do you secure an organization against its
• 20:07
  own empathy? You don't patch the human, you patch
• 20:09
  the process. The single most effective control
• 20:12
  against business email compromise is incredibly
• 20:14
  boring, but it works flawlessly if enforced.
• 20:18
  What is it? You institute an unbreakable rule.
• 20:21
  If a payment detail, a routing number, or a bank
• 20:24
  account changes on an invoice, you must verify
• 20:27
  that change through a secondary out -of -band
• 20:30
  channel before moving a single cent. A secondary
• 20:33
  out -of -band channel. Meaning, under no circumstances
• 20:36
  do you just hit reply to the email asking for
• 20:39
  the money. Never. You pick up the phone and call
• 20:42
  a number you already have saved on file for that
• 20:44
  supplier. You don't call the new phone number
• 20:46
  listed in the suspicious email signature. You
• 20:49
  use a known, trusted, pre -existing channel to
• 20:52
  verbally confirm the change. Oh, wow. It's that
• 20:54
  simple. It adds exactly two minutes of friction
• 20:57
  to the account's payable process, and it stops
• 20:59
  millions of dollars in corporate fraud dead in
• 21:02
  its tracks. It is remarkably sobering how the
• 21:04
  simplest administrative processes are often the
• 21:07
  only effective firewall we have. Well, we have
• 21:10
  covered some incredibly dense territory today
• 21:12
  on this deep dive. We started with the massive
• 21:15
  gigawatt level physical constraints of the AI
• 21:18
  boom. The memory and megawatts fundamentally
• 21:20
  required just to keep the lights on. The literal
• 21:22
  founding. Exactly. We explored Microsoft's ruthless
• 21:25
  human capital reallocation to feed that very
• 21:28
  same infrastructure. We unpacked how the EU's
• 21:31
  antitrust regulators are finally proving their
• 21:33
  fines have real unavoidable teeth. We looked
• 21:36
  at the terrifying two -hour exploitation window
• 21:38
  of software vulnerabilities and the reality of
• 21:41
  third -party supply chain breaches. It's been
• 21:43
  a heavy list today. It really has. And finally,
• 21:46
  we ended on the enduring, unpatchable vulnerability
• 21:49
  of human psychology and the economics of social
• 21:52
  engineering. Because the infrastructure of technology
• 21:55
  is not a clean digital cloud. It is a highly
• 21:58
  complex physical machine, deeply intertwined
• 22:01
  with raw materials, corporate strategy, regulatory
• 22:04
  friction, and human behavior. And understanding
• 22:07
  that messy, hidden infrastructure, staying critically
• 22:10
  informed about the mechanisms behind the headlines,
• 22:12
  is truly the best defense you can have against
• 22:14
  a rapidly shifting tech landscape. Thank you
• 22:16
  for joining us today, for going past the shiny
• 22:18
  interfaces, and for taking this journey with
• 22:20
  us. We've spent a lot of time today talking about
• 22:23
  two distinct problems, how AI development is
• 22:26
  currently constrained by massive physical power
• 22:28
  needs, and separately, how cyber attacks are
• 22:31
  moving with terrifying speed, weaponizing flaws
• 22:34
  in just two hours. But I want to leave you with
• 22:37
  a final thought to mull over. What happens when
• 22:40
  those two isolated forces inevitably collide?
• 22:43
  Ooh, what do you mean? What happens when generative
• 22:45
  AI is fully unleashed by threat actors to optimize
• 22:48
  and automate those rapid cyber attacks? and they
• 22:51
  point those automated attacks directly at the
• 22:53
  strained physical power grids and data centers
• 22:55
  that AI relies on to exist. It is a technological
• 22:58
  snake eating its own tail, and the fallout of
• 23:00
  that collision is something every business leader
• 23:02
  needs to be preparing for right now.