Latest / Tech Talks With Kinsoft / Last Week in Tech – OpenAI's $122B Round, Oracle's Mass Layoffs, and a Backdoor in a Top npm Package

Transcript

Last Week in Tech – OpenAI's $122B Round, Oracle's Mass Layoffs, and a Backdoor in a Top npm Package · Tech Talks With Kinsoft

• 0:00
  What happens when a company that's, you know,
• 0:01
  generating like $2 billion a month is built on
• 0:05
  a foundation so utterly fragile that just a single
• 0:09
  hollowed out block of code could bring the entire
• 0:12
  empire crashing down? Right. It's a terrifying
• 0:14
  thought. I mean, you look at the astronomical
• 0:16
  wealth of the AI boom and it feels completely
• 0:19
  invincible. But then you look beneath the surface
• 0:21
  and, well, the vulnerabilities are just terrifying.
• 0:24
  Yeah. And the thing is, the headlines focus almost
• 0:27
  exclusively on the money. Right. The valuations,
• 0:29
  which I mean, that paints a very lopsided picture
• 0:32
  of what is actually happening in the tech landscape
• 0:34
  right now. Exactly. Because the real story, the
• 0:37
  one that dictates the security of the devices
• 0:38
  you rely on every single day is that friction,
• 0:41
  that friction between this unprecedented growth
• 0:44
  and an incredibly brittle digital foundation.
• 0:49
  So true. And today we are pulling apart a pretty
• 0:52
  dense but fascinating tech update from Kinsoft
• 0:55
  Tech Talks. This is dated April 6th, 2026. It's
• 0:58
  a great source for this. Yeah, it really is.
• 1:01
  And our mission in this deep dive is to unpack
• 1:04
  this specific briefing to help you, the listener,
• 1:07
  understand this profound dual nature of our current
• 1:10
  reality. Right. Because we are basically witnessing
• 1:12
  this. historical reality bending growth that
• 1:17
  is It's completely masking some serious foundational
• 1:21
  cracks. Okay, let's unpack this. Starting with
• 1:24
  the sheer scale of the money. Because, I mean,
• 1:27
  the numbers Kinsoft reports coming out of OpenAI
• 1:29
  are almost difficult to process. Oh, they are
• 1:32
  staggering. They just closed a record funding
• 1:34
  round of $122 billion. I mean, billion, with
• 1:40
  a B. Yeah, which pushes their valuation past
• 1:42
  $850 billion. Wow. And, you know, to put that
• 1:46
  in an economic context, if that valuation were
• 1:48
  a country's GDP, it would comfortably place them
• 1:51
  in the top 25 economies on the entire planet.
• 1:53
  That is just wild. And the usage metrics totally
• 1:56
  back up that valuation. The briefing cites around
• 1:58
  two billion dollars in monthly revenue. Two billion
• 2:01
  a month. Yeah. And close to a billion weekly
• 2:03
  users, which let's actually visualize that for
• 2:05
  a second. Please do, because. It's hard to grasp.
• 2:08
  Right. So a billion weekly users means that roughly
• 2:12
  one eighth of the entire global population is
• 2:14
  interacting with this single company's tools
• 2:16
  every single week. It's just massive. Like if
• 2:19
  you walk down a crowded street anywhere in the
• 2:21
  world, statistically. One out of every eight
• 2:24
  people you pass is actively plugged into this
• 2:27
  specific ecosystem. Wow. It has moved entirely
• 2:30
  past being, you know, just a popular software
• 2:33
  tool. It's becoming a foundational layer of human
• 2:35
  communication and workflow. What's fascinating
• 2:38
  here is the speed of that integration. Because.
• 2:43
  Traditional tech giants like your social media
• 2:45
  monopolies or your legacy search engines, they
• 2:48
  took decades of slow iterative growth to reach
• 2:51
  a billion active users. Decades. Exactly. But
• 2:54
  OpenAI has compressed that timeline into a fraction
• 2:56
  of the space. Right. I mean, when you are looking
• 2:58
  at $2 billion in monthly revenue, you're no longer
• 3:01
  looking at early adopters trying out a novelty.
• 3:04
  Definitely not. That's real sustained usage.
• 3:06
  You are looking at global systemic reliance.
• 3:10
  entire industries are routing their core daily
• 3:13
  operations through a singular choke point. But
• 3:18
  I'm still stuck on the $122 billion funding round.
• 3:22
  When a software company takes in that much liquid
• 3:24
  capital, what are they even buying? Right. They
• 3:27
  are buying office chairs, right? They're buying
• 3:29
  physical infrastructure. They're buying entire
• 3:31
  energy grids and massive server farms just to
• 3:34
  keep the lights on for a billion users. Yeah.
• 3:36
  And that is the exact mechanism driving the broader
• 3:39
  market chaos right now. The competing power required
• 3:42
  to serve a billion users, generative AI, is...
• 3:46
  It's staggeringly expensive. I can only imagine.
• 3:49
  OpenAI is basically building a gravitational
• 3:51
  anomaly. The sheer mass of their capital and
• 3:54
  their compute power warps the entire tech ecosystem
• 3:56
  around them. Because if you are a legacy tech
• 3:58
  giant, you cannot simply sit on the sidelines
• 4:01
  while a competitor captures one -eighth of humanity.
• 4:04
  Right, which means the older companies have to
• 4:06
  pivot. And that pivot has a pretty brutal human
• 4:09
  cost. It really does. The Kinsoft source highlights
• 4:12
  a glaring example of this. Oracle, which is an
• 4:16
  absolute foundational giant in the database world,
• 4:19
  just executed its largest ever layoffs. Yeah,
• 4:22
  that was huge. We are talking about an estimated
• 4:24
  20 to 30 ,000 jobs cut. And the explicit stated
• 4:29
  reason for firing basically the population of
• 4:32
  a small town is to pay for their AI build out.
• 4:36
  It is a ruthless reallocation of capital. Yeah.
• 4:38
  You know, moving it from human resources to physical
• 4:41
  compute infrastructure. So what does this all
• 4:43
  mean? mean? I mean, the Kinsoft update notes
• 4:45
  that after Oracle laid off tens of thousands
• 4:48
  of people, the market actually rewarded the stock.
• 4:50
  Yeah, the stock went up. I really struggle with
• 4:52
  the logic there. How do we live in a reality
• 4:54
  where liquidating the livelihoods of 30 ,000
• 4:58
  human beings makes a company supposedly more
• 5:01
  valuable in the eyes of investors? Well, the
• 5:03
  market operates on cold mechanics and it is currently
• 5:06
  signaling a massive shift in how it values different
• 5:09
  types of corporate assets. OK, break that down
• 5:11
  for me. In the simplest terms, investors are
• 5:13
  looking at operating expenses which are you know
• 5:15
  the are going costs of paying human salaries
• 5:16
  right and they're weighing that against capital
• 5:19
  expenditures which is the money spent on physical
• 5:22
  assets like AI data centers and GPUs ah the hardware
• 5:26
  exactly right now the market views those massive
• 5:30
  data centers as the only viable lottery ticket
• 5:33
  for future survival So investors are essentially
• 5:35
  looking at Oracle and saying, well, we prefer
• 5:37
  you own machines that might build the future
• 5:40
  rather than paying the people who are building
• 5:42
  your present. That is the harsh reality. The
• 5:45
  market sees OpenAI's $850 billion valuation and
• 5:49
  demands that every other company build the infrastructure
• 5:52
  to compete in that arena. Wow. So when Oracle
• 5:55
  announces they are slashing their operating expenses
• 5:58
  to hoard capital for GPUs, Wall Street applauds
• 6:01
  the move as necessary for survival. That's just
• 6:03
  wild. It really is a stark reminder that this
• 6:06
  era of unprecedented wealth generation is happening
• 6:10
  concurrently with massive, disruptive human displacement
• 6:13
  within the exact same sector. It honestly feels
• 6:16
  like watching a company cannibalize itself. Yeah,
• 6:18
  that's a good way to put it. But, you know, while
• 6:21
  companies like Oracle are shedding massive amounts
• 6:23
  of human capital just to afford centralized warehouse
• 6:27
  -sized AI server farms, the technology itself
• 6:30
  is doing something entirely counterintuitive.
• 6:33
  Oh, absolutely. It's shrinking. And this is arguably
• 6:36
  the most pivotal development in the Kinsoft briefing.
• 6:39
  The push toward massive centralization is...
• 6:44
  It's creating an equal and opposite reaction
• 6:46
  toward extreme localization. Right. According
• 6:49
  to the update, Google just released Gemma 4.
• 6:51
  The source describes this as a family of, quote,
• 6:54
  open -weight models that are small enough to
• 6:56
  run fully offline on a smartphone. On a phone.
• 6:59
  Not a massive server farm. Exactly. The Kinsoft
• 7:02
  briefing explicitly states local AI is getting
• 7:04
  genuinely usable. Here's where it gets really
• 7:07
  interesting, because to me, this feels like an
• 7:09
  echo of the late 1970s. Oh, you mean the transition
• 7:12
  from the mainframe to the personal computer?
• 7:14
  Yes. Back then, computing meant these massive,
• 7:17
  room -sized, incredibly expensive mainframes
• 7:20
  that only governments and massive corporations
• 7:22
  could afford. You couldn't just have one in your
• 7:24
  house. Right. You had to send your data to them,
• 7:27
  wait for it to process and get the answer back.
• 7:29
  Then suddenly, the personal computer put that
• 7:32
  processing power directly onto your desk. A total
• 7:34
  paradigm shift. And we were watching the exact
• 7:36
  same trajectory play out with artificial intelligence.
• 7:39
  We are moving from relying on OpenAI's billion
• 7:43
  -dollar cloud servers. to having a highly capable
• 7:46
  intelligence running directly on your phone,
• 7:48
  completely disconnected from the internet. It's
• 7:50
  incredible. But I do need to ask you to clarify
• 7:52
  a term here. The briefing calls Gemma 4 an open
• 7:56
  -weight model. I know what open -source software
• 7:59
  is, where anyone can look at the code. But what
• 8:02
  exactly is an open -weight? Yeah, it is a crucial
• 8:05
  distinction. Because in traditional software,
• 8:08
  the code is a set of explicit instructions written
• 8:11
  by a human. Like a recipe. Exactly. And if you
• 8:14
  make it open source, anyone can read those instructions.
• 8:17
  But an artificial intelligence, like a large
• 8:19
  language model, isn't just a list of instructions.
• 8:22
  Okay, so what is it then? It is a massive neural
• 8:25
  network that has learned by processing mountains
• 8:28
  of data. The weights are the mathematical values
• 8:31
  of the connections within that network. Oh, I
• 8:34
  see. So the code is just the empty skeleton,
• 8:36
  but the weights are the actual brain. It's the
• 8:38
  accumulated knowledge the AI gained during its
• 8:41
  training. That is an excellent way to conceptualize
• 8:43
  it, actually. The code to run an AI is actually
• 8:47
  quite simple. Really? Yeah. The magic is in the
• 8:49
  weights, the billions of tune parameters that
• 8:52
  dictate how the AI understands language and reasoning.
• 8:55
  That makes so much sense. So when Google releases
• 8:58
  Gemma 4 as an open weight model, they aren't
• 9:01
  just giving developers the empty skeleton. They're
• 9:04
  releasing the fully trained brain, allowing anyone
• 9:06
  to download it. modify it, and run it locally
• 9:09
  on their own hardware. Which fundamentally shifts
• 9:12
  the power dynamic. Completely. I mean, if you
• 9:14
  can run a highly capable AI offline on your phone,
• 9:17
  you don't need to send your private data up to
• 9:19
  a multi -billion dollar server farm to get an
• 9:21
  answer. And for you, the listener, this means
• 9:24
  absolute privacy. When you query a cloud -based
• 9:27
  model, your data leaves your device. Right, it
• 9:29
  goes to their servers. Exactly. But with a local
• 9:32
  open -weight model like Gemma 4, your personal
• 9:35
  information never has to traverse the Internet.
• 9:37
  It decentralizes the intelligence. That sounds
• 9:40
  like a huge win. It is, however. As we embed
• 9:43
  these complex, locally running systems deeper
• 9:46
  into our personal devices, the basic software
• 9:48
  components those systems rely on become an incredibly
• 9:52
  lucrative target. Uh -oh. Which forces us to
• 9:55
  look at the darker side of this Kinsoft briefing.
• 9:58
  Unfortunately, yes. Because we've established
• 9:59
  that all of this incredible technology relies
• 10:02
  on a foundation of basic software building blocks.
• 10:05
  And the briefing details what it calls a nightmare
• 10:08
  scenario regarding those blocks. And it really
• 10:10
  is a nightmare. Attackers linked to North Korea
• 10:13
  executed a massive supply chain attack by placing
• 10:16
  a backdoor into a JavaScript package called Axios.
• 10:19
  Axios, yeah. And this isn't some obscure piece
• 10:21
  of software. Axios is one of the... the most
• 10:23
  widely used tools in the developer world with
• 10:26
  over 100 million weekly downloads. The scale
• 10:30
  of 100 million weekly downloads means this single
• 10:33
  package is embedded in the software of banks,
• 10:36
  hospitals, tech startups, and massive enterprises.
• 10:40
  Wow. You know. To visualize how terrifying this
• 10:43
  is, imagine you are a construction company building
• 10:46
  a new skyscraper. You wouldn't forge your own
• 10:49
  steel beams from scratch or mix your own cement,
• 10:52
  right? No, of course not. You'd buy them. You'd
• 10:54
  buy those standardized build blocks from a trusted
• 10:57
  supplier. Well, software development works the
• 10:59
  same way. Developers don't write every single
• 11:02
  line of code from scratch. They use prepackaged
• 11:04
  blocks, like Axios, to handle standard, repetitive
• 11:08
  functions. It saves time. So what Kinsoft is
• 11:10
  describing here is the equivalent of discovering
• 11:12
  that the factory supplying the steel beams for
• 11:15
  every single new building in your city has been
• 11:17
  secretly hollowing them out and packing them
• 11:20
  with explosives before delivery. That is a terrifying
• 11:22
  image. And nobody noticed because the beams looked
• 11:25
  completely normal on the outside. But I have
• 11:27
  to push back here. Okay, go ahead. If Axios is
• 11:30
  open source and 100 million people are using
• 11:33
  it, how does a state -sponsored hacker just slip
• 11:36
  malicious code into it without anyone noticing?
• 11:39
  Shouldn't someone have seen the explosives inside
• 11:41
  the steel beams? That's the logical question.
• 11:44
  Yeah. But if we connect this to the bigger picture,
• 11:47
  it exposes a massive vulnerability in how open
• 11:51
  source software is maintained. How so? It relies
• 11:54
  on the many eyes theory. The idea that because
• 11:57
  the code is public, Bugs and malicious additions
• 11:59
  will be spotted quickly. Right, because everyone
• 12:02
  is looking at it. But reality is that many of
• 12:04
  these foundational packages, even ones with 100
• 12:07
  million downloads, are maintained by a handful
• 12:09
  of unpaid volunteers doing it in their free time.
• 12:12
  Wait, seriously? Unpaid volunteers? Yes. They
• 12:16
  are essentially burning out trying to manage
• 12:17
  an infrastructure the whole world relies on.
• 12:20
  Oh my god. And sophisticated attackers, like
• 12:22
  state -sponsored groups, prey on that burnout.
• 12:25
  They don't usually hack in and steal the code.
• 12:27
  They use social engineering. And they manipulate
• 12:29
  people. Exactly. They pose as helpful developers.
• 12:33
  They submit dozens of legitimate, helpful updates
• 12:36
  to the code over months, building trust with
• 12:39
  the exhausted maintainers. Playing the long game.
• 12:42
  Yes. Once they have earned the status of a trusted
• 12:45
  contributor, they slip in a seemingly innocuous
• 12:48
  update that actually contains an invisible back
• 12:51
  door. Oh, wow. So the call is literally coming
• 12:54
  from inside the house. The maintainers themselves
• 12:57
  are tricked into approving the malicious code
• 12:59
  because they trust the person submitting it.
• 13:01
  And that is why a supply chain attack is a nightmare
• 13:04
  scenario. You could have the most secure corporate
• 13:07
  network on the planet protected by biometric
• 13:09
  locks and elite cybersecurity teams. Fort Knox.
• 13:12
  Right. But if the foundational building blocks
• 13:15
  you used to write your internal software were
• 13:17
  compromised at the source, all of those downstream
• 13:20
  defenses are completely useless. Because the
• 13:22
  enemy is already inside. The enemy is already
• 13:24
  inside your walls, smuggled in by your own developers.
• 13:27
  The briefing issues a very pointed warning here.
• 13:30
  Businesses must know their dependencies. You
• 13:33
  have to audit the building blocks you are bringing
• 13:34
  into your environment. Now, if you are listening
• 13:36
  to this, it is very easy to compartmentalize
• 13:39
  that threat. You might think, well, I'm not an
• 13:42
  enterprise developer. I don't build JavaScript
• 13:44
  applications. A supply chain attack on Axios
• 13:47
  is a corporate problem, not a me problem. Assuming
• 13:50
  immunity based on your technical expertise is
• 13:53
  a very dangerous posture in the current landscape.
• 13:55
  Absolutely. And the Kinsoft briefing aggressively
• 13:59
  shatters that assumption, bringing the threat
• 14:01
  straight down from the enterprise server level
• 14:03
  directly to your personal laptop. Yeah, this
• 14:06
  next part is wild. They highlight a separate
• 14:09
  critical vulnerability. Apple recently had to
• 14:12
  patch a flaw in Spotlight, which is, you know,
• 14:14
  the built -in search feature on Macs. Right,
• 14:16
  the little magnifying glass. Exactly. This flaw
• 14:19
  could expose files in a user's downloads folder.
• 14:22
  But the detail that the source flags as highly
• 14:24
  concerning is that it could expose cached AI
• 14:27
  data. Cached AI data, yeah. The update uses this
• 14:30
  as a direct command to keep your personal machines
• 14:32
  updated. But let's break down the mechanics of
• 14:35
  this. Why is our local AI... leaving cached data
• 14:38
  lying around for a search bar flaw to expose
• 14:40
  in the first place. Well, it comes back to how
• 14:43
  local models, like the ones we discussed with
• 14:45
  Google's Gemma 4, actually operate on a personal
• 14:48
  device. To be useful, An AI assistant running
• 14:51
  on your phone or your Mac needs context. It has
• 14:55
  to ingest your personal documents, your recent
• 14:57
  emails, and your private notes to formulate an
• 15:00
  intelligent answer. Wow. So it's reading my life
• 15:03
  to help me manage my life. Correct. But processing
• 15:06
  all that raw data requires massive amounts of
• 15:09
  computing power and battery life. Makes sense.
• 15:12
  To conserve resources, the system doesn't want
• 15:14
  to reread your entire hard drive every time you
• 15:16
  ask a question. So it creates a cache. a temporary,
• 15:20
  highly compressed memory bank of the intimate
• 15:22
  data it has recently processed. Oh, I get it.
• 15:25
  It is leaving breadcrumbs of its thought process.
• 15:28
  Exactly. And those breadcrumbs are pure, synthesized,
• 15:30
  highly sensitive information. Which is terrifying
• 15:32
  if it gets exposed. Yes. When a foundational
• 15:35
  everyday tool like Apple Spotlight has a vulnerability
• 15:38
  that accidentally exposes those specific cache
• 15:41
  files, the stakes change dramatically. Keeping
• 15:43
  your operating system patched. as Kinsoft strongly
• 15:46
  advises, is no longer about stopping a generic
• 15:50
  virus from slowing down your web browser. It
• 15:52
  is about actively protecting the raw, intimate
• 15:55
  context that your local AI tools are constantly
• 15:58
  digesting in the background. The paradigm of
• 16:01
  personal security has completely shifted. I mean,
• 16:04
  we used to worry about someone intercepting our
• 16:06
  credit card numbers. Right. Now we have to worry
• 16:09
  about a glitch in our desktop search bar accidentally
• 16:11
  exposing the private thoughts and unfinished
• 16:13
  documents we've been feeding to our local AI
• 16:16
  assistant. Yeah. As the technology becomes more
• 16:18
  personalized and integrated, the vulnerabilities
• 16:21
  become incredibly intimate. It forces a total
• 16:24
  reevaluation of what we consider sensitive data
• 16:27
  on our personal devices. It really does. Which
• 16:30
  brings this entire deep dive full circle. We
• 16:33
  started by looking at the dizzying astronomical
• 16:36
  heights of this tech boom. The big numbers. Right.
• 16:38
  Open AI absorbing $122 billion to fuel a valuation
• 16:43
  that rivals national economies, demanding a billion
• 16:46
  users every single week. Incredible scale. And
• 16:49
  we explored the immense gravitational pull of
• 16:52
  that growth, looking at how Oracle liquidated
• 16:55
  tens of thousands of jobs just to afford the
• 16:58
  physical infrastructure to stay relevant, a ruthless
• 17:01
  calculation that the market actively rewarded.
• 17:03
  We also examined the counter movement, you know,
• 17:06
  the democratization of that immense power through
• 17:09
  open weight models like Gemma 4, which shrink
• 17:12
  the capabilities of a massive server farm down
• 17:15
  to a device that fits in your pocket and runs
• 17:16
  entirely offline. But hovering above all of that
• 17:19
  incredible innovation is the sobering reality
• 17:21
  of a foundation it all sits on. We have state
• 17:23
  -sponsored actors playing the long game, using
• 17:26
  social engineering to slip invisible backdoors
• 17:28
  into universally trusted software blocks like
• 17:31
  Axios. And we have everyday flaws in our personal
• 17:34
  operating systems threatening to expose the deeply
• 17:37
  intimate data our local AI models are constantly
• 17:40
  processing. It is a landscape of unparalleled
• 17:43
  technological power built on an undeniably fragile,
• 17:46
  invisible foundation. Yeah. And as we wrap up,
• 17:49
  I really want to leave you with a final thought
• 17:51
  that connects these two realities. Oh, what's
• 17:53
  that? Well, we just learned that Axios, an open,
• 17:56
  trusted software building block used by millions,
• 17:58
  was successfully weaponized at its source. Yes,
• 18:01
  the hollowed -out steel beams. Exactly. And at
• 18:04
  the same time, we are celebrating the arrival
• 18:06
  of open -weight AI models like Gemma 4. These
• 18:09
  models are essentially the new, incredibly complex
• 18:12
  building blocks that we are rushing to download
• 18:14
  directly to our personal phones. Oh, I see where
• 18:17
  you were going with this. The question we must
• 18:18
  ask ourselves is this. As these opaque, pre -trained
• 18:22
  AI models become the foundational dependencies
• 18:25
  of the next decade, how do we ensure they aren't
• 18:29
  hiding the exact same kind of invisible, state
• 18:32
  -sponsored backdoors? Wow. I mean, when the building
• 18:34
  block isn't just a few lines of readable JavaScript,
• 18:37
  but an incredibly dense neural network weighing
• 18:39
  gigabytes. How do we actually verify what is
• 18:42
  hidden inside the weights? That takes the nightmare
• 18:44
  scenario and just amplifies it to an entirely
• 18:47
  new level. It really does. It perfectly echoes
• 18:49
  the final advice from the Kinsoft Tech Talks
• 18:51
  briefing. If you have never paused to ask yourself
• 18:54
  what is actually inside your software supply
• 18:56
  chain, today is the day to start asking. Absolutely.
• 19:00
  We are all building on top of this solid gold
• 19:03
  skyscraper, but we need to remember the swamp
• 19:06
  underneath it. As you navigate this highly capable
• 19:09
  but deeply vulnerable digital landscape, take
• 19:12
  that advice to heart. Stay patched. Stay patched,
• 19:15
  stay skeptical, and never stop questioning what
• 19:17
  is actually hidden inside the tools you trust
• 19:19
  every single day.