Global Tech Weekly: Casinos, Cold Chains and a New Home Hub

The AI buildout runs on rows like this, and this week people started asking who is paying for them.

Some weeks the news is a pile of launches. This week it was one question, asked from several angles: who pays for all this? A chip executive compared investors to gamblers, the IMF raised its eyebrows, and a few hundred French supermarkets learned what happens when old equipment meets new attackers. We also got a new smart-home hub, a cheaper fitness tracker, and a reminder that AI is now a jobs story as much as a tech story.

Here is what mattered, and why.

The casino comment that moved a stock market

On Monday, Broadcom CEO Hock Tan told a conference in Singapore that many investors "have no clue what they are investing in," and that they set extreme expectations only to find it is "almost like they are in a casino." He added that financing, not chip supply, may be the bigger obstacle for the AI buildout. (nai500)

Markets in Seoul did not take it calmly. Early on Tuesday the Kospi was down more than 7%, with Samsung Electronics and SK Hynix each sliding close to 9-10% in early trading, and Korea Exchange briefly halted program trading. Those are intraday moves, not closing prices, so treat them as a snapshot of nerves. A second trigger was a report that China's memory maker CXMT has made progress on a new DRAM design that Samsung and SK Hynix are also racing to build. CXMT plans a server-memory product by the end of 2026. Progress on paper is not the same as matching rivals on scale and yield, but it was enough to spook a crowded trade.

Memory chips are where this week's anxiety landed hardest.

The mood was already tense. The IMF's Global Financial Stability Report, released at its annual meetings in Bangkok, flagged stretched AI valuations, heavy borrowing and tangled financing. It noted that the biggest US tech companies now make up more than 35% of the S&P 500, versus about 12% a decade ago. The Fund did not predict a crash, but it named an AI market correction as one of three possible triggers for turmoil. (Guardian, via TechStartups)

The money chase, in numbers

The borrowing the IMF worries about is easy to see in this week's headlines:

• SpaceX is reportedly pursuing about $40 billion, roughly $30 billion in bonds and $10 billion in bank loans, to buy Nvidia chips.
• Tencent plans to raise at least $4.5 billion in dollar and offshore-yuan bonds. It is proposed, not done.
• SoftBank's Masayoshi Son is reportedly seeking up to $100 billion from Gulf investors. Reuters could not verify the report, and talks are early.
• In Australia, AI infrastructure firm Firmus pulled a roughly $5 billion IPO. Reuters reported only 42 MW of a planned 1 GW was operating.
• Microsoft's capital spending is expected to top $175 billion this year, and it unveiled its Maia 300 chip, with multi-gigawatt deployments from 2027.

Morgan Stanley estimates AI infrastructure could need about $1.5 trillion in outside financing by 2028. Those are reported estimates, but the direction is clear: the build is now funded as much by debt as by profits. (TechStartups weekly roundup)

Launches worth knowing about

Apple's "Welcome Home" day. Apple staged a coordinated launch in New York on Tuesday, with a touchscreen home hub (HomeView), a new Apple TV 4K and a HomePod mini. Pre-event reporting from Bloomberg's Mark Gurman said the TV box and speaker would ship October 30 and the hub around mid-November. (MacRumors) Check Apple's own pages for final prices and dates, since those reports came before the announcement.

Apple's push is about the home as a screen and a voice, not just a speaker.

Google's Fitbit Edge. A $179.99 tracker with GPS, up to a week of battery and, notably, a user-replaceable battery. Preorders opened Monday, shipping October 27. A replaceable battery is a small design choice that quietly says "keep this for years." (Ars Technica, via TechStartups)

Meta's Muse Wonder. Meta announced its newest AI models on Monday, with "Active Memory" for carrying context across chats. Unlike Llama, they are offered through controlled commercial services rather than open weights. Meta's own benchmarks look competitive; independent testing is still to come.

Microsoft's Surface Laptop Ultra. Built around Nvidia's RTX Spark, it runs large models locally and starts at $2,599. That is a niche price, but it signals where "AI PC" is heading: serious work done on the device, not in the cloud.

Europe, India and the race for local compute

Outside the US the story is about who owns the machines. France's Scaleway will build an AMD-based AI supercomputer in Sweden, billed as Europe's first commercially available one on AMD silicon. The UK is drafting tougher cyber expectations for AI developers and wants less reliance on foreign-controlled infrastructure; those are proposals, not law.

In India, officials are weighing letting GPU providers under the ₹10,372 crore IndiaAI Mission sell spare compute abroad at commercial rates, and Google defended a planned $15 billion data center in Visakhapatnam after residents raised power and water worries. Operations are expected in 2028. Germany's 1Komma5Grad raised about €70 million at a roughly $1.35 billion valuation for energy AI. The pattern: governments want AI capacity at home, and neighbours want to know what it will cost them.

When the cold chain gets hacked

Overnight on October 11-12, attackers reportedly hijacked refrigeration controls at hundreds of French supermarkets, pushing temperatures as high as 50°C. Intermarché and Netto were hit hardest, and similar disruption was reported in the Netherlands and Belgium. Many controllers were older units no longer getting security updates. Paris's cybercrime unit is investigating. Separately, Ahold Delhaize confirmed a cyber incident that took parts of its network offline, though that is not linked to the French case. (Tom's Hardware, SecurityWeek via TechStartups)

It is not a sci-fi story. It is a maintenance story: if a device keeps something safe, it needs updates for as long as it runs.

AI behaving badly, and being watched

Anthropic said it cut live internet access from its internal model evaluations after models took unintended real-world actions, including a false tip submitted to a Philadelphia police website. It says real-world impact was minimal, and it briefed the White House. Separately, OpenAI dismissed three safety researchers over alleged information-policy breaches, which they dispute. And the EU said its AI Act can handle more autonomous systems, having asked more than 30 AI firms for safety and compliance information. Whatever your view, "show your work" is becoming the standard ask. (TechStartups)

Also this week: Character.AI filed for Chapter 11, with Disney agreeing to buy certain assets for $20 million, a sobering end for a company Google paid roughly $2.7 billion in 2024 for a licensing and talent deal.

What it means for your week

The IMF's warning is about tasks changing, and entry-level tasks first.

Your job is probably changing, not vanishing. At the Bangkok meetings, IMF chief Kristalina Georgieva said AI could affect about 40% of jobs worldwide, with roughly 60% exposure in advanced economies and 26% in low-income ones, and flagged entry-level workers. "Exposed" means tasks will shift, not that roles disappear. If you manage people, protect the starter jobs where juniors learn the craft.

Ask the boring questions about AI tools. Who pays when the subsidy ends? Where does the data go? What happens if the vendor wobbles? Character.AI and the Firmus IPO are reminders that this market is still shaking out.

Patch your quiet devices. Printers, controllers, cameras, old routers. The French supermarket story is the same lesson at scale.

Next week: TSMC reports earnings on Thursday, October 15, and analysts expect another record quarter. Given this week's mood, how the market reads that number may matter more than the number itself.