Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

Tech Talks With Kinsoft

Tech Talks With Kinsoft artwork

Last Week in Tech – China's Open-Source Shock, TSMC's Record Quarter, and a 570-Flaw Patch Tuesday

Jul 19, 202619 minAudio

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! Your Monday catch-up on the week in tech, with a security lens. China's Moonshot AI releases Kimi K3 - the largest open-weight model ever, at 2.8 trillion parameters - and chip stocks shudder in a DeepSeek-style sell-off. The AI build-out rolls on regardless: TSMC posts a record quarterly profit, up 77 per cent, and adds another 100 billion US dollars to its Arizona plans, a day after ASML lifts its full-year outlook. The first Nvidia H200 chips ship to China under new licences even as Nvidia halves its approved Asian buyer list. And a report says Meta is weighing leasing Anthropic up to 10 billion dollars of compute - early talks only, and we flag it as such. On the security desk: Microsoft's biggest-ever Patch Tuesday fixes 570 flaws including two zero-days under active attack, a separate perfect-storm SharePoint flaw (CVE-2026-58644) draws a three-day CISA patch deadline, and the US, EU and UK launch their first joint cyber-sanctions package against the Russian ransomware ecosystem. Wondering what any of this means for your patch queue or your AI plans? Visit www.kinsoft.com.au to talk through your security and IT needs. Sources: Bloomberg; CNBC; Axios; SCMP; SEC filings; DigiTimes; Reuters; CNN Business; BleepingComputer; CrowdStrike; CISA; Rapid7; SecurityWeek; The Hacker News; The Record; Chainalysis.

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

So this week, a tech company basically gave away a multi -trillion parameter AI model entirely for free. Just completely free. Right. And it immediately triggered this massive market panic. I mean, it wiped billions of dollars off semiconductor stocks. It really did. But, you know, here is the profound irony of the whole thing. The exact same hardware companies that investors thought were losing out. They just turned around and posted record shattering multibillion dollar profits. It's totally wild. It is a complete

paradox. Like the financial markets are reacting to the sudden collapse in the price of software while completely misreading the physical reality of what it actually takes to power that software. Yeah. And that paradox is exactly what we're looking at today. Welcome to this deep dive. Today, we are taking a massive stack of source material, specifically a comprehensive July 2026 tech industry brief. It's a dense one, too. Oh, very dense. And our mission here is to cut through the noise of all the AI hype and the cybersecurity panic. We want to help you understand the hidden,

interconnected forces that are actually shaping the digital infrastructure that you rely on every single day. Exactly. And we're going to do that by tracing this really fascinating causal chain. We're going to start at the very top layer of... the tech world, like the cutting -edge AI models, and follow the money. Follow the money and the physical hardware supply chains, right? All the geopolitical drama, tracing it all the way down to the vulnerable legacy servers, just sitting in business closets around the world. Okay, let's unpack this. We have to start with the software layer because it is radically altering the economics

of artificial intelligence. It's a complete shift. So on July 16th, a Beijing -based company called Moonshot dropped a new model. It's called Kimi K3. The specs on this release are almost hard to comprehend. They really are. We are talking about 2 .8 trillion parameters and a 1 million token context window. What's fascinating here is just the sheer scale of engineering required to reach 2 .8 trillion parameters. Yeah, it's massive. To put that in perspective, I mean, just a couple of years ago, a 400 billion pyramidal

model was considered... absolute frontier tech. A model of this new size is likely using a massive mixture of experts' architecture. Which basically means it doesn't run the whole thing at once, right? Right, exactly. It routes your query to highly specialized subnetworks rather than firing up the entire multi -trillion parameter brain all at the same time. That makes sense. And what about that one million token context window? Oh, that's huge. In practical terms, that means like a front end developer can dump their entire company's code base plus maybe three distinct

programming textbooks straight into the prompt. Oh, wow. Yeah. And the AI just holds all of it in its working memory simultaneously. That is massive utility. But here is the real kicker for me. It is open weight. Meaning it is entirely free to download and run. Exactly. Moonshot has made this the largest open source model ever released to the public. And they are claiming it beats leading U .S. models across a variety of benchmarks. Which is a bold claim. It is. And, you know, I always look at vendor supplied benchmarks with a critical eye. But wait, the

briefing notes that in blind independent testing, developers actually preferred Kimi K3 for front end coding tasks. Yeah, they really did. utility is undeniably real. It is very real. And the financial markets recognized it instantly, but they didn't celebrate it. They panicked. Big time. The source material compared the market reaction to the deep seat panic we saw back in early 2025. The major semiconductor index dropped over 4 percent. Ouch. Yeah. And giants like NVIDIA

and Micron lost serious ground on paper. But hold on. This is the part that doesn't make sense on the surface to me. If an incredibly powerful model is now free to download, wouldn't tens of thousands of companies rush to download it? You would think so. Right. And wouldn't they need to buy more NVIDIA chips and memory to actually run a 2 .8 trillion parameter behemoth on their own servers? So why did the chip stocks drop? Well, it's because the market is terrified of software commoditization. Okay. Unpack that a bit. Sure. So if frontier level AI capabilities

are suddenly just free, the massive profit margins of enterprise software companies start to look very fragile. And those companies currently buying the bulk of these GPUs. Exactly. Investors fear that if the software layer can't maintain high prices, their capital expenditure budgets for buying hardware will eventually dry up. Oh, I see. It's a psychological reaction to the cost floor of AI just violently collapsing. But giving away an AI model of this size isn't purely altruistic, right? There has to be a catch. Oh, absolutely.

The catch here is data governance. Right. So if you're listening to this and you run a U .S. financial firm or maybe a health care provider, You can't just casually spin up a Chinese origin open weight model on your network without setting off massive compliance alarms. Not at all. And looking purely at the enterprise mechanics here and setting aside the intense political debate around the U .S.-China tech race, which I want to be clear, we are remaining entirely impartial on the data provenance and regulatory obligations of running these models, are just an absolute

minefield. Exactly. We are just impartially porting the geopolitical context from the source material here. The fact is being free to download does not mean it is free of legal or regulatory friction. Which perfectly explains. why that initial stock market panic over the hardware companies was entirely disconnected from physical reality. Like, Kimi K3 tanked chip stocks on paper because the software became free. But when you look at the physical ground the company is actually manufacturing the silicon, the dynamic is the exact opposite.

If we connect this to the bigger picture, the software getting cheaper is actually making the hardware infinitely more lucrative. How so? Because no matter how free the weights are, you cannot execute code on thin air. You need the chips. The numbers are just staggering. On July 15th, ASML beat their expectations and raised their AI demand outlook. And the very next day, TSMC reported their fifth consecutive record quarterly profit. We're talking $40 billion in revenue, with net income up 77 % year over year. And here's

the metric that should make everyone listening sit up and take notice. Two -thirds of TSMC's revenue is now coming from high -performance computing. That's AI, right? Mostly, yes. For a decade, smartphones drove TSMC's profits. That era is officially over. The pivot to AI hardware is absolute. Plus, they announced another $100 billion in capital investment for their Arizona facilities. Which pushes their total committed U .S. spending to $265 billion. It's like the ultimate goal. Gold Rush. Moonshot is giving

away the gold maps for free, but TSMC and ASML are the ones making billions selling the shovels. That's a perfect way to put it. So if you are sitting in a boardroom right now managing an IT budget, mapping out your server storage or memory needs for the next two years, the takeaway is pretty stark. What's the takeaway? You are directly competing with AI hyperscalers for the exact same advanced silicon wafers. Oh, wow. That is the crucial point here. The physical supply chain dictates your everyday IT budget. A $100 billion fab plant takes years to build.

Yeah, you can't just spin that up overnight. Exactly. The demand is here today, but the supply won't catch up for a very long time. You basically have to assume a massive, sustained squeeze on supply and prices. So plan refresh cycles early. Very early. You need to plan them much earlier than you traditionally would because the hardware is heavily constrained. But even if you had the budget, you... can't just buy these chips freely, right? The geography of compute is totally locked down. It is. We are seeing a massive geopolitical game of musical chairs dictating global GPU distribution.

Yeah, the supply is tightly controlled by export restrictions. Like on July 14th, a U .S. commerce official testified that the first batch of NVIDIA's highly coveted H200 chips are finally shipping to China under a new licensing regime. Right, but it's just a trickle. Just a trickle. Reuters reported that more Chinese firms, including including a unit of ZTE, were approved as buyers. But the mechanics of this are so fascinating. Oh, totally. Because on the exact same day that news came out, NVIDIA internally purged over half of its

authorized Asian customers from a whitelist. Yeah, they are trying to stop this massive gray market where chips are bought in one country and then secretly diverted through third -party routes. It's such a strange dance. It's like opening the front door a tiny crack while simultaneously building half the side windows shut. That is exactly what it's like. And just to maintain our framework here, we are neutrally reporting the actions of the U .S. Commerce Department, NVIDIA and the subsequent supply chain friction. Right. We aren't taking any stance on the validity

or the. Politics of the export controls themselves. Exactly. But we do need to talk about the impact because the administrative burden this places on hardware distribution is immense. It really is. And that friction isn't just high -level international politics. It directly impacts the listener. How does it hit the day -to -day IT teams? Well, this geopolitical turn dictates where the world's GPUs end up. If your company is trying to spin up a new application and you're waiting, you know, six months for cloud compute capacity. Which happens a lot right now. All

the time. Or if your vendor keeps inexplicably pushing back their AI roadmap, this export control tug of war is a massive underlying reason for that. It creates total scarcity in the cloud. And that extreme compute scarcity is forcing fierce corporate. rivals into these almost surreal arrangements. The briefing highlights a major single -source press report. This was relayed by The Times, Reuters, and CNN, claiming that Meta is considering leasing its own compute power to Anthropic. The maker of the Claude AI models.

Right. And the deal is rumored to be worth up to $10 billion over two years. Now, I have to put a massive asterisk on this. Please do. We have to be clear. Nothing is agreed upon. They could walk away tomorrow. This is just a rumor. Right. Just a rumor for now. But just the logic of it is... I mean, Meta is spending well over $100 billion on AI infrastructure this year alone just to win the AI race. Why on earth would a giant like Meta lease its precious, hard -fought compute to a direct rival like Anthropic? It

comes down to the economics of what we call compute landlords. A few landlords. Yeah. Meta might be realizing that the raw infrastructure they've built is actually more valuable than the software running on it. Okay, wait, explain that. Well, building a massive GPU cluster isn't just about buying 100 ,000 chips. It's about the incredibly complex InfiniBand networking and the power infrastructure required to make them talk to each other without idle time. So it's not just renting out a spare bedroom. It's an entirely different business model. Exactly. It's almost like Meta built a massive solar farm for their own headquarters,

realized they generated more power than they could use 2047 during their training runs, and decided to become the local utility company for their biggest rival. That's spot on. Renting out excess capacity at a premium guarantees billions in high margin revenue. It basically insulates them, even if their own AI software models face setbacks. Which is a smart hedge. Very smart. If Meta is transitioning from just being an AI builder to becoming an AI landlord, it tells you everything you need to know about the market right now. And for the rest of us. For the rest

of us, it means when giants fight over capacity to this degree, we just have to accept that AI compute will stay. Okay, so we've spent the first half of this deep dive talking about the cutting edge. It's the fun stuff. Yeah, trillion parameter open models, multi -billion dollar chip fabs, massive compute landlords. But what is all of this corporate data actually sitting on? What is the actual physical foundation supporting this frantic digital economy? It is legacy infrastructure.

And that foundation is severely cracking. It is crumbling. Here's where it gets really interesting. Because on July 14th, Microsoft shipped its biggest patch Tuesday ever. That was huge. We are talking about 570 vulnerabilities fixed in a single day. Just a staggering number. And 59 of them were rated critical. Two were zero days already being actively exploited in the wild. One gave attackers administrative privileges through Active Directory Federation services, but the absolute top priority was a separate flaw in SharePoint server. Yes,

cataloged as CVE -2026 -58644. A mouthful. It is, but it scores a 9 .8 out of 10 on the severity scale. Let's break down the mechanics of why a 9 .8 is so catastrophic. Because this specific vulnerability allows an attacker to execute malicious code on an unpatched on -premises SharePoint server with no login required. Right. Zero authentication. From a network security perspective, a pre -authentication remote code execution or RCE is the ultimate

nightmare. So wait, the attacker doesn't even need a stolen password. Nope. They don't need to trick an employee with a phishing email to bypass multi -factor authentication. No, they don't need any of that. Because of how the server parses incoming network requests before it even checks who the user is, the attacker just sends a specifically cracked packet to the server's public -facing door. And then what? And then they are instantly executing code at the system level. That is deeply unsettling. You are completely exposed just by being plugged into the Internet.

Exactly. And because this was a zero day, attackers were exploiting it before Microsoft even had a patch ready. Yeah, they were breaking in, stealing server machine keys and planting malware to establish persistence. Which is why CISA. The U .S. cyber agency added this to their known exploited list on July 16th and gave federal agencies an unusually brutal deadline to patch it. Just three days. Three days and federal IT is light speed. It really is. So for you listening, here's the immediate takeaway. If your company's SharePoint lives

in the cloud on Microsoft 365, you are immune to this specific flaw. That's the good news. But if you run an on -premises SharePoint server. specifically the 2016 -2019 or subscription editions, you have to patch immediately. And we must stress the difference here between merely patching and actively hunting for a pre -existing compromise. What's the difference? Well, because attackers were already exploiting this in the wild. Applying the patch today just locks the window they used to get in. But they might already be inside.

Exactly. If they compromised you yesterday, they are already sitting in your living room. You have to explicitly hunt through your system logs for signs of stolen machine keys or existing compromise. Okay, but let's play this out. Let's say a business misses that brutal three -day deadline. They don't hunt for the compromise. What happens? Invariably, the ransomware gangs get in. They encrypt the data, steal a copy, and demand a massive payout. Right. But the entire endgame of how a business handles ransomware fundamentally shifted this week. Yes. On July

13th, the U .S., the EU, and the U .K. launched a massive coordinated cyber sanctions package. But they didn't just target a few random affiliates like they usually do. No, they targeted the structural plumbing of the Russian ransomware ecosystem. They went after the administrator of the TrickBot and Conti operations. A single individual who allegedly received over $300 million personally. Wild. They also sanctioned a bulletproof hosting provider called MediaLand that sheltered gangs like LockBit for a decade. They targeted specialized

password -stealing crews and the VPN encryption services these criminals use to hide their tracks. They are basically mapping out the specialized supply chain of cybercrime, initial access brokers, negotiators, hosting providers. It's a huge shift in tactics by law enforcement. They aren't just trying to catch the burglars anymore. They are bulldozing the burglars' safe houses and freezing their bank accounts. By attacking the infrastructure, yes. But this raises an important question. What would your organization actually do tomorrow if you were hit? Because these new sanctions

make paying a ransom legally radioactive, right? Yes. This is the crucial danger for the listener here. If your company gets locked up and your executives decide to quietly pay the ransom just to get your business operational again, you are stepping into a legal minefield. Because of OFAC. Exactly. OFAC, the U .S. Sanctions Enforcement Agency, operates on strict liability. Meaning what exactly? Meaning it does not matter if you didn't know the hacker was sanctioned. If your ransom payment ends up in the crypto wallet of a sanctioned entity, which is highly likely now

that they've sanctioned all the core infrastructure providers, your organization could be breaking the law. You could face catastrophic fines or federal prosecution simply for trying to buy your own data back. Yes. The safety net of just paying the ransom and calling it a cost of doing business is entirely gone. So you can't pay them, but you can't survive without your data. What is the actual solution? The only solution is the unglamorous, boring IT work. It is having tested backups. Heavily tested offline backups.

Yes. Physically isolated from your main network. And it requires a rehearsed incident response plan. You need a pre -agreed decision framework so your board of directors isn't trying to decipher international sanctions law while staring down a 48 -hour countdown timer. That makes a lot of sense. So what does this all mean? It's a lot to take in. It is. When you step back and look at the whole board, we're basically watching a massive collision. We have the plunging cost of AI software models like Kimi K3 crashing headfirst into severe geopolitical hardware bottlenecks.

Driven by export controls and immense silicon demand at TSMC. Exactly. And that collision is creating this frantic digital economy built on top of incredibly vulnerable, heavily targeted legacy servers. It creates a really striking reality when you think about it. If the actual software and AI models are trending toward being entirely free and open, but the physical chips and the security infrastructure to protect your data are becoming impossibly expensive and legally fraught, are we entering an era where the true currency of the tech world is no longer code?

Wait, if it's not code, what is it? It might be pure physical infrastructure and cyber fortresses, the hardware and the walls protecting it. Wow, that is a brilliant thought to leave on. The code might be free. But the concrete foundation costs billions and is under constant attack. Thank you for joining us for this deep dive. As you go about your week, remember to look past the shiny AI headlines. Check your patches, test your backups, and stay curious. We'll catch you next time.

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. Iris Ohyama - A Suspicious File in the Cloud, 7.7 Million Customers at Risk, and an Online Store That Went Dark

    Nov 19, 202620 minAudio

    Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! Japanese home goods company Iris Ohyama (online store operator Iris Plaza) found a suspicious file in an external...

    Credit the people on this episode

  2. Who Gives A Crap - A Marketing Supplier Breached, Six Days to Tell the Client, and Customer Data That Didn't Need to Be There

    Nov 17, 202619 minAudio

    Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! Australian online retailer Who Gives A Crap told customers on 7 November 2026 that their information was accessed...

    Credit the people on this episode

  3. Last Week in Tech - An Outlook Zero-Day That Leaks Your Password Hash, Chrome's Eighth Exploited Bug, and Nvidia Without Jensen Huang

    Nov 15, 202621 minAudio

    Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! Last week in tech, 7 to 14 November 2026. Microsoft's November Patch Tuesday (10 November) fixes well over a...

    Credit the people on this episode

  4. Chainalysis - From "Non-Sensitive" to "All Users": How a Breach Notice Changed in Five Days

    Nov 12, 202619 minAudio

    Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! Chainalysis disclosed on 30 October 2026 that an external actor using "offensive agentic techniques" accessed...

    Credit the people on this episode

  5. NSW Health - A Dental Voucher Website, Four Gigabytes Taken, and the Providers Whose Details Were Exposed

    Nov 10, 202623 minAudio

    Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! NSW Health reported unauthorised access to its Oral Health Fee For Service Scheme website on 29 October 2026;...

    Credit the people on this episode