Last Week in Tech - Nvidia Doubles, Meta Pays $18 Billion, and the Musk-Altman Feud Reaches Your Code Editor
21 minAudio
Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! Last Week in Tech - Nvidia Doubles, Meta Pays $18 Billion, and the Musk-Altman Feud Reaches Your Code Editor Your Monday roundup of the technology and security stories that mattered to Australian businesses in the week of 24-30 August 2026. In this episode: Nvidia's record quarter (26 Aug). Revenue US$96.2bn, up 106% year-on-year; data centre US$89bn (~93% of the company); Q3 guided to ~US$108bn and CFO Colette Kress forecasting ~70% growth for fiscal 2028. What an accelerating AI capex cycle means for your cloud pricing and FY27 budget. - Meta's teen-harm settlement (26 Aug). Up to ~US$17bn - Meta itself puts it at ~US$18bn over ten years, ~US$12.7bn guaranteed - settling the multistate lawsuit filed by 33 state attorneys-general, without a finding of liability, but with enforceable product changes: two-hour daily limits and a midnight-6am curfew for under-18s, likes hidden by default on children's accounts, cosmetic filters banned for minors, most school-hours notifications off. A template for how Australia's under-16 social media ban could be enforced, and a warning on design-liability risk. - OpenAI to cut Cursor off (29 Aug). After SpaceX's US$60bn acquisition of Cursor-maker Anysphere closed on 14 Aug, OpenAI proposed ending model access on 12 November, citing its litigation history with Elon Musk's companies. Cursor says OpenAI models are ~5% of traffic. Concentration risk in AI dev tooling: know which models your tools depend on. - Alibaba's Wan 3.0 (24 Aug). Document-to-video AI at ~US$6 per 30-second clip - and the data-sovereignty question to ask before marketing uploads anything. - Security round. Citrix NetScaler CVE-2026-8452 (8.8 CVSS v4.0) - disclosed in June as denial-of-service, reclassified to unauthenticated RCE, added to CISA's KEV on 26 Aug with web shells found on compromised appliances; Zimbra CVE-2026-73570 (8.9 CVSS v3.1) - single-email command injection, 270+ servers confirmed compromised, ~8,200 still unpatched; and Boston Scientific's still-unattributed cyberattack halting order shipping and manufacturing worldwide. Coming up: Wednesday, the Quest Apartment Hotels breach. Friday, Manchester Airports Group in full. Visit www.kinsoft.com.au to talk through your security and IT needs. Sources: CNBC, Fortune and Kiplinger on Nvidia Q2 FY2027 results (26 Aug 2026); CNN Business, Bloomberg and Al Jazeera on the Meta settlement (26-27 Aug 2026); CNBC and the OpenAI blog on Cursor model access (29 Aug 2026); Seeking Alpha on the SpaceX-Anysphere close (14 Aug 2026); Dataconomy and Winbuzzer on Alibaba Wan 3.0 (24-25 Aug 2026); Help Net Security, SecurityWeek, BleepingComputer and CISA KEV on Citrix CVE-2026-8452; The Hacker News, runZero, Shadowserver and CCB Belgium on Zimbra CVE-2026-73570; TechCrunch, The Register and BleepingComputer on Boston Scientific (26 Aug 2026).
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 writing an $18 billion check just because the push notifications on your app were a little too effective. Yeah. I mean, that is a terrifying thought for literally any product team out there right now. It really is. So today we have our hands on a late August 2026 tech and security briefing from Kinsoft. And the mission of this deep dive is to basically cut through the noise of these massive billion dollar headlines, you know, to figure out how they actually change
your daily operations. Right. Because it's easy to just read the headline and move on. But the stuff trickles down fast. Exactly. I mean, if you think about the current tech industry, it operates a lot like this massive planetary system. You have these gigantic gas giants companies like NVIDIA. meta SpaceX. And when one of these giants shifts its orbit even slightly, every single smaller moon in the system immediately feels that gravitational pull. That's a great analogy. Thanks. And by smaller moons, I mean your budget, your development pipeline, and of course, your security posture. Yeah, that gravitational
pull is really the perfect way to look at this briefing because we are looking at a fundamental restructuring of how software is, you know, built, paid for and regulated. The shifts happening at the very top are dictating the reality on the ground. Doesn't matter if you're trying to just ship a new feature or keep a physical manufacturing plant online. So let's start with the biggest center of gravity in tech right now, like the literal sun at the center of our planetary system, NVIDIA. Oh, yeah. The numbers are just staggering.
They really are. Their Q2 earnings just dropped, and the numbers are almost hard to comprehend. We're talking about revenue hitting $96 .2 billion. Wow. Right. That is up 106 % year over year. The market was expecting around $92 billion, which was already a sky -high expectation, and they just blew past it. But you have to look at the composition of that revenue, though. Okay, what do you mean? Well, out of that $96 .2 billion, data center revenue alone accounted for $89 billion. Wait, just the data centers? Just the data centers.
That is 93 % of the entire company's revenue coming strictly from that one area. I mean, it literally gives you a physical map of where global capital is flowing right now. And their Q3 guidance is predicting around $108 billion. Plus, Colette Kress, the CFO, she's forecasting roughly 70 % revenue growth for the next fiscal year. Jensen Huang even told analysts that demand is, you know, accelerating. It's just not slowing down at all. But I want to push back on what this actually means for everyday IT budgets. Because
for the last, like, two decades, the fundamental assumption in tech has been that compute power always gets cheaper. Right, the whole Moore's Law mentality. Exactly. It gets cheaper and more plentiful over time. We just assume AWS or Azure will eventually lower their prices as servers get more efficient. But looking at these numbers... I mean, if compute is the new oil, are we assuming we'll just keep finding endless new reserves? Yeah, that's the trap. Doesn't NVIDIA's order book suggest that companies are essentially sleepwalking into massive permanent budget hikes for cloud
pricing instead of it just being a temporary spike? It absolutely points directly to a permanent structural shift. I mean, building out these AI data centers requires vast amounts of capital, physical infrastructure and grid level energy. Yeah, the power requirements are insane. Right. There is a hard physical limit to how fast you can actually build a nuclear power data center. So the cloud providers buying these chips at a massive premium, they have to pass those costs down to you. So the days of cheap cloud are just over. Basically, yeah. If you are an organization
budgeting for cloud compute over the next five years, assuming a return to the cheap, abundant compute of the 2010s is a highly, highly dangerous game. And, you know, the scale of this infrastructure spend is also exactly what's driving these astronomical startup valuations. When compute is this expensive, the companies that can marshal the resources to actually train foundational models become insanely valuable targets for acquisition. Which brings up a really interesting ripple. effect. Because when companies are spending billions to lock down AI talent and models, those corporate
turf wars bleed directly into the tools your developers are relying on every day. Oh, the cursor situation. Yes. Let's talk about cursor. For anyone who doesn't know, it's a wildly popular AI coding assistant used by engineering teams pretty much everywhere right now. On August 14th, SpaceX completed a $60 billion... All stock acquisition of AnySphere, the company that builds Cursor. Which is, I believe, the largest startup exit on record. It is. But then, just over two weeks
later, the concentration risk of that deal became very, very real for anyone using the tool. Because of OpenAR. Right. On August 29th, OpenAR announced it will just shut off access to its models inside Cursor, effective November 12th. Just a total rug pull for the user. Exactly. OpenAI stated that given its long litigation history with Elon Musk's companies, it simply cannot be confident that SpaceX will use its technology within its terms of service. So they just cut them off. Yep. Now, Cursor's CEO, Michael Truel, responded
by saying, well, OpenAI models only account for about 5 % of Cursor's traffic anyway. And they pointed out that developers can just use a bring -your -own -API -key option as a workaround. Which sounds easy on paper. Right, but to me, this sounds like renting an apartment where two landlords start fighting in the hallway and one of them suddenly decides to, like, take away the plumbing. And you're just trying to live there. That is exactly what it feels like. So let me ask you, this isn't bring your own API key, just a bandaid. Doesn't this mean the dreaded vendor lock -in has basically moved from the
cloud provider directly to the AI model itself? It absolutely means lock -in has moved to the model layer. Bring your own key shifts the burden entirely to the individual developer or the IT department to manage billing and access. But it doesn't solve the underlying fragility of the tool chain. Yeah, you still have to manage it all. Right. 10 years ago, a CTO worried about being locked into AWS or Azure. Today, you have to map out exactly which neural net is powering your IDE, your summarization tools, your customer service bots. That sounds like a nightmare. It
is. If your entire engineering pipeline is optimized around the nuances of GPT -4 and that access gets revoked because of a billionaire feud, your sprint cycle is completely derailed. It's a harsh lesson that you must have fallback models ready to deploy on incredibly short notice. Man, that is a massive headache for whoever is managing a company's tech stack. But, you know, while those enterprise AI tools are facing intense corporate feuds. Consumer platforms are facing a totally different kind of pressure. The legal
pressure. Exactly. Which brings us back to that $18 billion check I mentioned at the start of the deep dive. Yeah, this is a truly landmark settlement. Meta just agreed to a deal worth up to roughly $18 billion. It's paid over 10 years with $12 .7 billion guaranteed. And this resolves a massive multi -state lawsuit filed back in 2023, right? Yeah. By 33 state attorneys general. That's right. And the core allegation was that Meta deliberately designed addictive products that harmed young people's mental health. The briefing notes this actually follows a March
jury verdict in California that found Meta and Google liable in a related trial. Now, Meta settled this specific lawsuit without a formal finding of liability. But the real story here. isn't the size of the payout, is it? No, it's the enforceable product changes that the court is mandating. Right. Meta has to implement a two -hour daily use limit for users under 18. They are forced to implement a curfew from midnight to 6 a .m. Which is huge. Huge. Likes and reactions must
be hidden by default on children's accounts. Cosmetic procedure filters are banned for minors. And most push notifications have to be disabled during school hours. That last one is really the kicker. Yeah, I want to unpack the mechanics of this, actually. Why are these specific features, like a simple push notification, being targeted so aggressively by the courts? Because the state attorneys general successfully argued that these are not just, you know, neutral design choices. They are active mechanical engagement loops. OK, what does that mean in practice? Well, think about how a push notification works psychologically.
It leverages variable reward schedules, which is literally the exact same psychological mechanism used in slot machines. Oh, wow. Yeah. You feel a buzz in your pocket, and you don't know if it's a critical message from a family member or just an alert that someone liked your photo. Right, so you have to check it. Exactly. That uncertainty triggers a dopamine response, compelling you to open the app. The legal argument was that deploying these specific mechanisms against minors, knowing how their developing brains process dopamine and variable rewards, constitutes a defective
and harmful product design. It is wild to think about a... Software feature being legally classified as a defective product, like a faulty airbag or something. Exactly. The Kinsoft briefing points out that this settlement is essentially a commercial template. It's a blueprint for online safety regimes globally, like Australia's incoming under -16 social media ban. It's going to change everything for consumer apps. But this feels like it requires a massive shift in corporate structure. It highlights the irony that UI and UX designers are now potentially
the biggest liability for a tech company. Absolutely. If a simple push notification strategy contributes to an $18 billion settlement, do we start seeing compliance officers sitting directly on design teams, aggressively vetting every single button color and engagement ping? We will absolutely see that integration. I mean, in highly regulated spaces like finance or healthcare, legal teams already review product features. This settlement drags consumer software into that exact same reality. So move fast and break things is over.
The era of move fast and break things in consumer UI is completely dead. Designers will now have to document the psychological intent behind an engagement loop just as rigorously as developers document code for security audits. If you are building a product today, your design decisions are carrying multi -billion dollar legal exposure. That is. Man, designing products safely is clearly becoming incredibly bureaucratic and perilous. But what happens when the risk isn't coming from what you design, but from what your employees quietly adopt behind your back? Ah. The shadow
IT problem. Exactly. Because as strict as these new legal regimes are, they collide directly with how cheap and tempting new AI tools have become. Let's look at Alibaba One 3 .0, which launched on August 24th. It's an incredible tool, technically speaking. It really is. This is an AI video model that takes your internal documents, your spreadsheets, your slide decks, and just turns them into high -quality 30 -second videos. And it costs as little as $6 a clip. The output quality is highly impressive, but the immediate enterprise risk is just massive. Right. Because
the moment a user uploads an internal strategy deck or a client spreadsheet to generate that video, you're introducing a severe data sovereignty problem. Because of where it's processed. Yes. You are taking sensitive corporate IP and uploading it to a platform that operates under Chinese jurisdiction, which means it's subject to entirely different data access laws than a server in Ohio or Frankfurt. Let me paint a picture here, because anyone who is working in a corporate office knows exactly how this plays out. I'm sure they do. Right. Imagine a stressed marketing coordinator.
It is 4 p .m. on a Friday. They desperately need a promo video for a campaign launching on Monday. We have all been there. Exactly. And the internal creative team tells them, well, it'll take two weeks and cost $5 ,000. But one 3 .0 says it will take two minutes and cost the price of a fancy coffee. Well, it's a no -brainer for the employee. Is corporate policy even strong enough to stop a $6 shortcut? I mean, how do companies actually enforce data sovereignty when the bleeding -edge tools are this fast and this accessible? The truth is they often don't. And that phenomenon
is exactly what we call shadow IT. It's any software or device used by employees without the explicit approval of the IT department. And AI is making it worse. AI is supercharging it. You can have the most robust compliance policies in the world, written by the most expensive lawyers, but they frequently fail at the hands of a stressed employee on a Friday afternoon with a corporate credit card. Because convenience wins. Convenience almost always bypasses policy. It shows how user behavior is the ultimate wildcard in cybersecurity. Assuming
your written policies are enough to keep your data safe is a total illusion when the barrier to entry for these powerful tools drops to $6. That illusion of safety is actually the perfect bridge into the security reality check portion of this briefing. Because if you think a marketing coordinator bypassing policy is bad, wait until you see what happens when the actual... IT department thinks they are secure. Yeah, the Citrix Net Stealer situation is a perfect example of this. It's a terrifying case study in how fast the threat landscape moves. So we are talking about a vulnerability tracked as CVE -2026 -8452. On
June 30th, Citrix originally disclosed this as a denial of service bug. And just to clarify, a denial of service or DOS bug typically just means an attacker can crash the system. It's a nuisance, sure. It stops traffic, but it doesn't actually give the attacker access to your underlying data. Okay, right. But then, in mid -August, researchers at Watchtower proved that this wasn't just a crash bug. They pushed it all the way to an unauthenticated remote code execution. No credentials needed, no user interaction required. A massive escalation. Massive. The severity was
upgraded to a CVSS score of 8 .8, and the briefing notes that defenders are now actively finding web shells on compromised appliances in the wild. But I need to stop and ask for some translation here. First, what does the CVSS score of 8 .8 actually mean in practical terms? Okay, so CVSS stands for Common Vulnerability Scoring System. It's a 1 to 10 scale used by the security industry to rate the severity of a bug. So 10 is the worst. Exactly. Anything above an 8 is critical, meaning it's highly exploitable and highly damaging. Okay, and my second question, I don't get the
leap. How does a bug that just crashes a system suddenly evolve into a master key for hackers two months later? Explain that to me. So, when a system crashes from a denial of service attack, it's often because the attacker fed the system unexpected data that corrupted its memory. In June, That just crashed the server. It just panicked and shut down. Right. But over the next two months, hackers studied exactly how that memory corrupted. They figured out how to carefully craft that unexpected data so that instead of crashing the
system, it actually writes malicious instructions directly into the server's memory. Oh, wow. Yeah. That is remote code execution. They went from just kicking the door so it gets stuck to actually picking the lock. And what about these web shells they are finding? A web shell is essentially a permanent backdoor. Once the attacker uses that remote code execution to get in, they drop a small piece of code, the web shell, onto the server. So they can get back in later. Exactly. Now, even if you patch the original vulnerability, the attacker can still log in through that hidden
backdoor anytime they want. So if I'm an IT admin, my blood is boiling. I applied the patch back in June, thinking it was just a nuisance bug. I checked the box. Now you were telling me I have to go back and forensically hunt for hidden back doors on a machine I thought was completely safe for the last two months. I know, it's brutal. How was a normal IT team supposed to keep up when the goalposts moved that drastically? The harsh reality is that they often don't. If you are an IT admin, a downgraded severity rating
in June does not mean you can relax in August. You have to operate under the assumption of a breach. So assume you're already hacked. Basically, you can't just apply the new patch today. You have to assume that in the window between June and August, an adversary realized the true potential of that bug and left a web shell behind. It sounds exhausting and unfortunately is not an isolated incident in this briefing. Let's talk about Zimbra mail servers. There is a vulnerability here, an unauthenticated command injection flaw scoring an 8 .9 on the CVSS scale. Another critical one.
Yeah. A single crafted email delivers shell commands directly to the server. Now, it was patched a month ago, but this briefing notes that over 270 servers are confirmed compromised, and roughly 8 ,200 unpatched instances are still exposed to the open internet. Mechanics, get out. How does a single email execute commands on a server? Right, so a command injection flaw happens when a... A piece of software takes input from a user, in this case an incoming email, and accidentally passes that text directly to the underlying operating
system. Wait, it just hands it over? Pretty much. The attacker crafts an email where the subject line or the header isn't just text, it contains actual operating system commands. Because the Zimbra software doesn't sanitize or filter that input properly, the server reads the email, thinks it's receiving a direct instruction from an administrator, and just executes the code. And because it's a mail server? It's inherently facing the public Internet, just sitting there waiting for an email. Exactly. It has to be exposed to work. And the fact that 8 ,200 instances are still unpatched a month later is a classic example of patching
fatigue. Yeah, people are just tired. Zimbra is heavily used by smaller businesses and universities who might not have massive security teams. But the automated scanning tools used by threat actors, they don't care about your team size. They sweep the entire internet in hours. If you are sitting on an unpatched Zimbra server today, you must assume someone has already sent that crafted email and compromised your network. Man. And if anyone listening thinks these vulnerabilities just result in stolen data or a defaced website, you need to look at the final story in today's
Kinsoft briefing. Boston Scientific. This one is truly alarming. It is. This is a $20 billion medical device maker. They supply stents and pacemakers to hospitals globally. And on August 25th, they suffered a cyber attack. But you have to look closely at the fallout from this. This wasn't just a data breach where patient records were stolen. Right. Their order processing and shipping systems went down. Manufacturing lines were actively affected. Workers at their plants in Ireland were literally sent home because the digital systems required to run the physical plant were paralyzed. Wow. Now, no one has claimed
responsibility yet. And this specific attack vector hasn't been confirmed. But the real world impact is undeniable. How do we get from a digital network flaw to factory workers in Ireland being sent home? It is the convergence of IT and OT. IT and OT. Yeah. IT is information technology, your email servers, your databases. OT is operational technology, the digital systems that control physical valves on a manufacturing floor or the robotics assembling a pacemaker. Oh, I see. Historically,
those networks were kept completely separate. But today, for efficiency, they are deeply connected. When ransomware or a malicious payload breaches the IT network, security teams often have to proactively shut down. down the OT network to stop the virus from spreading to the factory floor. So they essentially paralyze themselves to stop the spread. Exactly. The gap between a minor software bug and a global shipping freeze has effectively vanished. Cyber risk is no longer just a specialized IT problem. It is a physical supply chain risk. Cyber resilience is now completely synonymous with business survival. So let's step
back and connect the dots of everything we've explored today. We started with NVIDIA's massive economic engine driving the global AI build -out, pointing toward permanent infrastructure budget hikes. Right. Then we saw how those same billions fuel corporate feuds that can break your AI development tools, like Cursor, with just a few weeks' notice. We examined the mechanics of how UX design choices, like a simple push notification, now carry $18 billion in legal liability. Which still blows my mom's… Seriously. Then we highlighted how
the shadow IT adoption of cheap tools like Alibaba's WAN 3 .0 threatens your corporate data sovereignty. And finally, we look at how unpatched servers and shifting vulnerabilities can literally halt the manufacturing of life -saving medical supplies. The through line is inescapable, really. Whether you are budgeting for cloud compute, designing a product feature, or deciding when to patch an email server, these are no longer isolated departmental issues. The decisions being forced by these tech giants dictate your fundamental ability to just keep your doors open. I want
to leave you with a final thought to mull over as you navigate your week. We've talked about the sudden fragility of... of AI tool chains and the very real physical paralysis of cyber attacks at a company like Boston Scientific. What happens when the cheap, accessible AI tools we are increasingly rushing to rely on these models generating our code and summarizing our data become the actual vectors for the next major zero -day exploit? Imagine a scenario that combines the toolchain fragility of a massive corporate
feud with the real -world supply chain paralysis of a physical cyberattack. That is a nightmare scenario. It is not science fiction, though. The building blocks are sitting right in front of us. Thank you for joining us on this deep dive. Take a hard look at your toolchains, verify your patch logs, and remember that when the gas giants shift their orbit, it's the moons that have to brace for impact. We'll see you on the next deep dive.
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...
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...
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...
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...
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;...