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 - Nvidia Turns GPUs Into an Asset Class, the AI Price War Breaks Out, and an AI Agent Goes to War

Aug 23, 202622 minAudio

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! Your Monday roundup of the technology and security stories that mattered to Australian businesses in the week of 10-16 August 2026. Nvidia's $500bn financing platforms (10 Aug). MOUs with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilise third-party capital for AI compute. Why an MOU is not a contract, and the reported - but not company-confirmed - 25% loan backstop. The AI price war (11-14 Aug). Gemini 3.7 Flash three weeks after 3.6 at half the price; GPT-5.6 Luna cut ~80%; Claude Opus 5 at $5/$25 with the Sonnet 5 price rise cancelled; DeepSeek raising prices sharply. What to do about your existing vendor agreement - and why introductory pricing that expires at end-2026 isn't a 2027 budget line. The near-autonomous AI agent attack on Taiwan (12 Aug). Open-source agent frameworks, guardrails bypassed by claiming authorised pen-testing, and an operation that expanded its own scope. Plus OpenAI's Daybreak Blue and Red tiers, and 51 US House Democrats asking questions. The chip money wave (10-13 Aug). TSMC's record July (+44.7% YoY), Intel's upsized $20bn raise, Cisco's $9.3bn of AI infrastructure orders, Applied Materials' record quarter - and why beating expectations still moved three of these share prices down. IBM and OpenAI (13 Aug), Gemini past 1bn users, and Anthropic watermarking Claude output worldwide - the first clearly visible case of EU AI Act compliance being exported globally by default. Security round. August Patch Tuesday (~400 CVEs, one exploited zero-day, CVE-2026-68820, attributed to Lazarus a day later); VMware vCenter CVE-2026-59310 exploited five days after disclosure with 361 victims in 47 countries; SAP Commerce Cloud CVE-2026-58231 exploited three days after patch day; Cisco ASA/FTD CVE-2026-20349; Adobe Commerce CVE-2026-71362 exploited within hours; unpatched GeoServer and Windows Defender ShieldBreak zero-days with no CVE at all; and macOS Screen Sharing CVE-2026-65400 being used to plant Monero miners on internet-exposed Macs. Incidents: the Ceva Logistics breach rippling to Steam, ING, Bol, De Bijenkorf and Ajax; and ransomware taking out doors and HVAC - not clinical systems - at Manitoba's largest hospital. Coming up: Wednesday, the Bendigo Bank cyber penalty. Friday, the Metabase zero-day in full. Visit www.kinsoft.com.au to talk through your security and IT needs. Sources: Nvidia newsroom, CNBC, Bloomberg, TechCrunch, Reuters, VentureBeat, Financial Times, Dream Group research, CyberScoop, The Register, CNN, OpenAI, The Hill, TSMC and Intel investor relations, Cisco, Applied Materials, IBM newsroom, Ars Technica, Axios, SecurityWeek, BleepingComputer, The Hacker News, Zero Day Initiative, Tenable, Rapid7, CISA, Onapsis, Dutch NCSC, The Record, CBC.

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.

Imagine signing this massive enterprise software contract on a Monday. You think you've locked in a great rate for the year. Right. You're feeling good about it. Exactly. And then by Friday, the product you just bought is suddenly like 80 percent cheaper. It's heavily regulated by some new law passed in the EU. And oh, yeah, it's being actively weaponized by North Korean hackers. Which sounds like a nightmare scenario. It really does. But the thing is, that isn't some hypothetical stress test. That is the actual reality of the tech

landscape we are navigating right now. The ground beneath our feet is completely shifting. I mean, the map you were using yesterday is suddenly just obsolete. Yeah, which means the traditional approach of looking at tech news as a series of isolated events. It's a recipe for disaster. We have to stop looking at the individual weather events and, well, start paying attention to the tectonic plates moving underneath. Welcome to our deep dive. Our mission today is to unpack this incredibly dense, rapid -fire technical briefing from Tech Talks with Kinsoft. Yeah,

the August 24th, 2026 briefing. Right. And we are tracking a series of massive structural shifts that are all happening simultaneously. I mean, we're talking about a fundamental rewiring of how artificial intelligence is financed, an absolute cliff dive in inference pricing, the dawn of autonomous cyber attacks, and a frankly brutal Patch Tuesday. And my goal here is to help us connect the dots between these events because, you know, a shift in Wall Street financing for Silicon eventually dictates the security posture of, say, a regional hospital. They are all nodes

in the same expanding network. I want to speak directly to you listening right now. Whether you are a CFO staring down a ballooning cloud computing bill or a CSO losing sleep over new threat vectors or even just an IT manager trying to keep your hardware refresh cycle on track, this briefing contains granular details. that are absolutely going to dictate your priorities this week. So let's jump straight into the money. Let's do it. Because the economics of the AI boom just got a major rewrite. NVIDIA just signed Memorandum of Understanding with six absolute

titans of Wall Street. We're talking Apollo, BlackRock, Blackstone. Brookfield, Goldman Sachs, and KKR. The stated goal here is to mobilize $500 billion for AI infrastructure. Jensen Huang actually went on television and explicitly called his chips an investable asset. That phrase, investable asset, it represents a profound shift in language and strategy. How so? Well, think about how massive compute clusters have historically been funded.

If you wanted to build a data center full of GPUs, you paid cash off your own balance sheet, or you borrowed against your corporate credit rating. Right, which works perfectly fine if you are Microsoft or Amazon. Exactly. But if you're a smaller cloud provider, a specialized neocloud, or some AI startup burning through venture capital, you simply do not have the balance sheet to drop billions of dollars in cash on hardware. That old model completely breaks down. OK, let's unpack this because there's a crucial mechanism at play here. These are no use, right? Not committed capital yet. Correct. But according

to the reporting around this, NVIDIA is agreeing to backstop up to 25 percent of these loans. That is $125 billion in potential exposure. Yeah, it's massive. So if I'm understanding this correctly, is Nvidia basically acting like a car dealership offering a guaranteed trade -in value just to convince the banks to approve your massive car loan? That analogy hits the nail on the head. By guaranteeing the residual value of the chips being used as collateral, NVIDIA is actively underwriting the resale value of its own product.

Wow. If a startup defaults on their loan, the bank knows NVIDIA will buy the chips back at a set floor price. That reduces the risk for Wall Street, making the financing math work. So the build out of AI infrastructure is no longer constrained by the cash these tech companies have on hand. Not at all. It is now entirely fueled and Which naturally leads to the big question for anyone listening. How does this Wall Street maneuvering directly affect my monthly cloud bill? Like if my company is running AI compute, what does this massive debt fueled build out

actually mean for my operational expenses? It creates a massive tug of war on your invoice. On one side, this influx of financed capital means an astronomical amount of physical capacity is being built. More servers coming online increases supply, which traditionally drives the price of running inference, you know, generating the AI responses down. Right. Supply and demand. But on the other side of that rope, that capacity is being built by entities that now have massive debt service to cover every single month. That pressure creates an artificial floor, keeping

prices from going to absolute zero. And right now, the supply side of that tug of war is winning by a landslide. The floor is completely falling out on AI model costs. Oh, absolutely plummeting. Look what happened with Google. They launched Gemini 3 .7 Flash just three weeks after launching version 3 .6. Three weeks. And they halved the price down to $0 .75 per million input tokens and $3 .75 for output. OpenAI slashed the price of GPT -5 .6 Luna by 80%. Anthropic cut Claude

Opus 5 by half. The unit economics of the entire industry are in freefall. There is one massive exception in the data, right? Yeah. DeepSeek v4 Pro, their prices actually shot up with reports showing an 1100 % increase on certain complex workloads. Wait, really? 1100 %? Yeah. But the broader trend, particularly in the U .S. market, is a fierce race to the bottom. It's driven by enterprises aggressively capping their bills and using alternative open weight models to force those price cuts. So the immediate action item

for you as a listener is clear. If you signed an AI vendor agreement in early 2026, the unit economics have moved aggressively in your favor. Get your account manager on the phone and renegotiate. Or if you shelved an internal project six months ago because running the inference was going to bankrupt you, you need to rerun those build versus buy calculations today. The math has completely changed. But, and there's a big but here, if these highly capable models are getting incredibly fast and practically free for legitimate businesses,

we have to acknowledge the dark side of that equation. Right. They are also suddenly very fast, practically free tools for threat actors. Exactly. If threat actors can spin up thousands of operations for pennies, the immediate bottleneck in a cyber attack isn't software anymore. Take the research published this week by the Israeli security firm Dream. Yeah, this was fascinating. They detailed an attack targeting Taiwanese government systems that occurred in early July. And the attackers didn't even use bespoke, custom -coded

malware. They used open -source, general -purpose AI frameworks, specifically Hermes and OpenClaw. And they bypassed the AI's safety guardrails using a surprisingly simple framing technique. They just primed the system to act as if it were conducting authorized penetration testing. Wow. Just like that. Yeah. And once they bypassed the guardrails, the aliens ran autonomous learning cycles. And here's where I really want to break down the mechanics, because they didn't just run a static script. No, not at all. These agents are constantly checking CVE databases, which

are public catalogs of known software vulnerabilities, and then cross -referencing that data with code snippets on GitHub to figure out how to compile custom exports on the fly for the specific target in front of them. Think about the traditional threat model. Historically, a security team assumes that the most expensive part of a breach is the human attacker's time. Right. Because highly skilled hackers don't work for free. They have to sleep and they can really only focus on one complex problem at a time. Exactly. This attack fundamentally inverts those economics. It is

like facing a burglar who doesn't just jiggle your doorknob, but can instantly read the architectural blueprint of your house, pull the manufacturing specifications of your deadbolt, and analyze the patrol schedule of the neighborhood security guard all at the exact same time. And then synthesize a custom key in seconds. Right. I do want to push back on the hype just a little bit here, though. The briefing specifically notes this was near autonomous. and still required human tuning. So, are the headlines that are out there

screaming about Terminator -style autonomous hacking overstating the current reality? Like, is this really a terrifying new weapon, or just a glorified, highly efficient script? Well, it is not a fully autonomous self -aware system. You are absolutely correct to highlight that nuance. But treating it as just a glorified script is dangerous. Okay, why? Because it is an unprecedented multiplier. An AI agent reading a database and trying exploits is not a fundamentally new class of weapon. But when that agent can scan, adapt, and iterate thousands of times a minute for fractions

of a cent, the attacker's time is no longer the limiting factor. And this isn't just happening in the shadows with rogue actors. That same week, OpenAI launched Daybreak Blue and Red. Right, models explicitly built and released for offensive security work. They noted that GPT -5 .6 cyber independently found two previously unknown Chrome V8 bugs. It found zero days entirely on its own. Which naturally caught the attention of lawmakers. Over two dozen U .S. House Democrats wrote to Sam Holtman, and another 22 wrote to Dario Amode

at NPROPIC, demanding answers about these AI agents breaking out of test environments. Regardless of your political leanings, we're not taking a side on the politics here. But this bipartisan panic shows that lawmakers across the board are realizing the regulatory framework is miles behind the actual capability of the technology. Absolutely. So if threat actors and enterprise defenders are spinning up thousands of these autonomous AI agents per minute, the physical constraints of the real world come into play. Someone has

to power that compute. Right. The hardware. You would think the companies building the fiscal chips and the hardware infrastructure would be invincible right now. But Wall Street is telling a completely contradictory story. Let's look at the hardware paradox. The demand signal is undeniably astronomical. TSMC, you know, the massive foundry that actually manufactures the chips, had a record July. We are talking 14 .5 billion U .S. dollars, up almost 45 percent year on year. And Intel saw so much demand for a share offering that they upsized it from $15 billion

to $20 billion in a matter of hours. But here's where it gets wild. Look at Cisco and Applied Materials. Yeah. Cisco beat their expectations, pulling in $17 .3 billion in revenue, heavily driven by massive hyperscaler AI orders. But when they announced that... Their shares fell 4 .6%. And Applied Materials set an all -time record with $9 .12 billion in revenue, and their shares fell 5%. Make that make sense for me. They're literally breaking records and printing money, and Wall Street punishes them. It comes

down to two major factors, the concept of being priced for perfection and the reality of the infrastructure lag. Okay, let's start with priced for perfection. Wall Street has baked in such astronomical, compounding expectations for AI growth that merely having a record -breaking quarter is viewed as a failure. The market's tolerance for anything less than pure exponential acceleration is zero. And what about that infrastructure lag? Why is TSMC rewarded while Cisco is punished? Because building a data center happens in phases. The hyperscalers, Google, Meta, Microsoft, they

buy the GPUs first. So TSMC, who fabricates the silicon, sees the immediate revenue spike. Okay, that makes sense. But you don't immediately buy all the networking gear to connect those chips until the facility is actually being built out. Cisco operates in that networking layer, which naturally lags behind the initial silicon purchase. Oh, I see. Furthermore, Applied Materials makes the massive complex machines that fabricate the chips. They sell to TSMC. Their peak happened even earlier in the cycle when the foundries

were expanding. So Wall Street looks at Cisco and applied materials and assumes their hyper growth phase has already peaked, even if their current revenue is at an all time high. That context is crucial. So if you're a listener running IT for a midsize company, why do you care if Wall Street is throwing a tantrum over a stock price? Because financial volatility eventually translates directly into supply chain volatility. When these hardware companies get punished by the market. Despite record revenues, it impacts their internal capital expenditure plans, their

hiring, their factory expansions. They might delay building a new networking equipment factory? Exactly. For the listener, this Wall Street drama eventually shows up on your desk as completely unpredictable lead times for your own company's router and server hardware refreshes. So you have a hardware market defined by volatile lead times and AI models moving at breakneck speed while being actively weaponized. How on earth is a standard, heavily regulated business supposed to safely implement AI? under these conditions?

Well, they don't do it alone anymore. We are seeing a massive enterprise reality check. Look at the strategic partnership announced between IBM and OpenAI. They are embedding GPT 5 .6 into IBM Consulting Advantage and standing up a practice of thousands of consultants. Meaning the days of an enterprise just buying an API key off a website and handing it to their developers are ending? Exactly. For regulated industries like finance, healthcare, and government, systems integrators like IBM are the new gatekeepers. distribution route for frontier AI into the enterprise

is now through a massive consulting firm with a certified practice that can handle the liability and the implementation strategy. And the rules governing that implementation are getting incredibly complex. Google Gemini just passed 1 billion monthly users. Crucially, 63 % of those interactions are voice and 20 % involve live camera or screen sharing. Which is a massive shift in how people use AI. Right. Meanwhile, Antropic is now watermarking all claw -generated text worldwide using a standard called C2PA, and they are doing this to comply

with Article 50 of the EU AI Act, which went into effect on August 2nd. I want to break down how C2PA actually works because it isn't just a visible logo stamped at the bottom of a document. Right. It's way more technical than that. C2PA relies on cryptographic asset hashing. It mathematically binds the origin data, proving the content was generated by an AI directly into the metadata of the file itself in a way that travels securely with the asset, even if it gets copied or moved. Which brings up a fascinating geographic friction point. If I'm a listener running an IT department

in Chicago, managing an industry, internal HR bot for my employees, why am I suddenly subject to the EUAI Act? You are experiencing regulatory export. For a global vendor like Impropic, it is a logistical nightmare to maintain one set of untracked models for the United States and a completely separate, cryptographically watermarked set of models for Europe. So they just do it for everyone. It is simply cheaper and more efficient to build one system that complies with the strictest global standard, in this case, Brussels, and

apply it everywhere. So that mid -level IT manager in Chicago now has to ensure their internal HR systems can handle and process C2PA cryptographically hashed data simply because the base model they are renting complies with European law. The regulation exports itself globally through the software architecture. That is wild. And let's revisit those live camera stats for Gemini. If 20 % of a billion users are interacting visually, That completely shatters how most companies handle data security. If your company's acceptable use policy for AI only covers what employees are

allowed to type into a text box, that policy is already obsolete. You have employees pointing their phone cameras at proprietary engineering schematics or confidential whiteboard meetings and asking the AI to analyze them in real time. It represents a massive, unmonitored expansion of the corporate attack surface. Which brings us to the immediate, boots -on -the -ground reality for security teams. While executives are busy strategizing over AI consulting contracts and EU compliance, the traditional IT house is currently on fire. Yeah, let's talk about that. Let's look

at the fallout from a frankly brutal patch Tuesday, which brought around 400 fixes. One of them really stands out. CVE -2026 -6820 -0. It's a flaw in a Windows component called WinSOC. Microsoft rated it a CDSS 7 .0, categorizing it as important. Let's define CVSS for a moment. That is the common vulnerability scoring system. A 7 .0 out of 10 usually means a vulnerability is serious, but perhaps requires the attacker to already have some level of access or it's difficult to execute. Right. So an IT team might logically schedule

that patch for the following week's maintenance window. But the very next day, the security firm Checkpoint revealed that this exact vulnerability was already being actively exploited in the wild. North Korea's Lazarus Group was using it to deliver a kernel rootkit in an espionage campaign called Operation Dream Job. And a kernel rootkit is a nightmare scenario. It really is. It means the attackers have burrowed so deeply into the core of the operating system, the kernel, that they essentially have God mode access. They can

hide their tracks from anti - So if a patch is rated a moderate 7 .0 on Tuesday, but on Wednesday we find out it's actively being used to deploy a North Korean kernel rootkit, are these CVSS severity scores? Basically useless as a prioritization metric? Well, it proves that a severity score is merely an input. It cannot be your only decision metric. The context of active exploitation is what matters. And the window of time you have to apply these patches is shrinking drastically.

Just how fast are we talking? The data shows a VMware vCenter vulnerability that was exploited in just five days. An SAP Commerce cloud flaw was exploited in three days. Three days from the moment the patch was released to active in the wild attacks. And threat actors aren't just going after the obvious, heavily fortified corporate servers anymore. They are attacking the blind spots. Look at the breach at Siva Logistics, a global freight forwarder. The attackers breached Siva, but the fallout cascaded down the supply chain to six downstream brands, exposing customer

data for companies like Valve's Steam Gaming Platform and ING Bank. Because supply chains are deeply interconnected data webs, your security is only as strong as your weakest vendor. I take it a step further because it goes beyond data. Look at the ransomware attack on the hospital in Winnipeg. The attackers didn't take down the clinical IT systems or the patient record databases. They took out the heating, ventilation, and air conditioning, the HVAC systems, and the physical door access controls. This is a critical vulnerability for so many organizations. If you ask a CSO to

list their critical assets, they pull up a list of servers, databases, and employee laptops. They aren't thinking about the internet -connected thermostat in the lobby. Threat actors are aggressively targeting building management systems because they often aren't even on the IT department's asset register. If you aren't patching the HVAC system, a threat actor will gladly use it as a backdoor into your network. The ecosystem is incredibly fragile right now. We even saw a security researcher going by the handle Nightmare Eclipse

publish a fully working zero -day exploit for Windows Defender. They called it Shield Break, and they released it simply because Microsoft threatened legal action over their previous vulnerability submissions. And there's no patch for it yet. Exactly. Speaking of immediate vulnerabilities, there is one highly practical, critical takeaway for our listeners today regarding Apple screen sharing. Yes. There is a critical flaw exposing Mac computers via port 5900. Attackers are scanning the internet, finding that open port, gaining

root access to the machine, and silently installing neuro -cryptocurrency miners. Wow. If you have Macs in your fleet with port 5900 exposed to the open internet, closing that port needs to be your top priority the moment this deep dive ends. Let's take a breath and recap the sheer scale of the terrain we've covered today. We started by unpacking how half a trillion dollar AI infrastructure deals are now entirely reliant on debt financing, creating a massive supply surge that is plummeting the cost of AI models. Huge shift. We saw how that incredibly cheap

compute is directly empowering threat actors to run near autonomous hacking swarms. We navigated the volatile hardware market, where the infrastructure lag and impossible expectations are punishing companies despite record profits. We looked at how global regulations like the EU AI Act are exporting themselves into your local systems via cryptographic watermarks. And we ended with the stark reality of three -day exploit windows, compromised hospital HVAC systems, and North Korean rootkits. It is a phenomenal amount of structural change occurring all at once. It really

is. And it leaves me with one final thought tomorrow. over building on everything we've discussed today yeah we are watching ai rapidly drive down the cost and complexity of offensive cyber operations you know those open source agent swarms running autonomous learning cycles but simultaneously we are seeing ai deployed for defense like open ai's daybreak models actively hunting for zero -day vulnerabilities right it begs a serious question are we rapidly approaching a point where human security teams are no longer the primary combatants are we simply going to be spectators

watching offensive AI, battle machine speed, defensive AI in real time, completely unable to intervene because the operational tempo is entirely beyond human cognition. Wow. That is a heavy but incredibly necessary question to sit with as we watch these tectonic plates shift in real time. The map has changed, and it is entirely up to us to learn how to navigate the new terrain. Thank you so much for joining us for this deep dive. As a casual reminder before you go, check those vendor contracts to see if you can get a better rate on your AI compute, update your acceptable use policies to account

for live cameras, and for goodness sakes, close port 5900 on your Macs. We'll catch you on the next one.

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