ARM Making Chips for Meta – Big Industry Shift?

Welcome back to Tech News of the Week, where Chris and I break down the biggest and weirdest stories in tech. We're a week behind because Chris decided to lose power—how selfish! But we’re back, and we’ve got four spicy news stories to dive into. Let’s go!

🧠 **Meta Wants ARM-Made Chips**
ARM might start making its own chips, and Meta is reportedly first in line to buy them. This is a big shift for ARM, which has historically just designed and licensed chip architectures rather than manufacturing its own. If true, this could shake up the chip industry and make ARM a competitor to companies it currently licenses to. The first chips are rumored to launch this summer, so we won’t have to wait long to see what happens. Will this push companies toward RISC-V? Fingers crossed!
🔗 https://techcrunch.com/2025/02/13/arm-is-launching-its-own-chip-this-year-with-meta-as-a-customer

📖 **Facebook’s AI Reads Minds (Kind Of)**
FAIR (Facebook’s Fundamental AI Research unit) teamed up with scientists in Spain to create a machine that can read your mind—well, sort of. By analyzing brain activity using M-E-G and E-E-G, they achieved an 80% accuracy rate in predicting what subjects were typing or saying. Right now, the tech is clunky and requires a controlled environment, but smaller, scarier versions are likely on the way. What could go wrong?
🔗https://www.techspot.com/news/106721-meta-researchers-unveil-ai-models-convert-brain-activity.html

💰 **SolarWinds Goes Private for $4.4B**
Remember SolarWinds? The company that got hit with a massive supply chain attack in 2020? Well, private equity firms have decided it’s still worth squeezing for cash. Silver Lake and Tomo Bravo bought up a majority stake, and now TurnRiver is taking the whole thing private for $4.4 billion. Expect less innovation, more “cost optimization,” and an eternal cycle of rent-seeking. Somewhere in Middle-earth, Sauron is proud.
🔗 https://www.darkreading.com/cybersecurity-operations/solarwinds-private-billions

🖥️ **AI is Just Fancy Copy-Paste, Confirms Study**
A new report shows that AI-assisted coding is leading to lower code quality. GitClear analyzed 200 million lines of code and found that, surprise surprise, AI-generated code is often just old code copied and pasted with minimal thought. Google’s own research backs this up, showing rising defect rates in published code. Microsoft even warns that overreliance on AI is killing critical thinking skills. So, uh… we’re definitely headed toward a bright, bug-free future, right?
🔗 https://www.gitclear.com/ai_assistant_code_quality_2025_research

Transcript

[00:00:00.00]
Announcer: Welcome to Tech News of the Week with your host, the Tupperware container you still haven't returned to your mother.

[00:00:07.08]
Ned: Welcome to Tech News of the Week. This is our weekly Tech News podcast, where Chris and I take a look at four interesting news articles that caught our attention were a week behind, and that's because Chris selfishly lost power. How dare you, sir?

[00:00:24.12]
Chris: I keep trying to have a nuclear reactor in the living room, but apparently that's problematic.

[00:00:29.08]
Ned: It's not up to Code, whatever. All right, let's talk about some tech news things. Arm might make chips for Meta. According to a report from the Financial Times, they have received news that Arm Semiconductor is planning to create its own chips for use by Meta, that Facebook company thing. Arm has historically designed the instruction set and the cores for Arm architecture and then licensed its designs to companies like Qualcomm and Apple to produce the actual chips. In this regard, Arm has been a largely neutral party when it comes to selling chips, licensing their IP to whomever had adequate funds. If this report turns out to be true, Arm would become a direct competitor to those companies it currently licenses to. What type of impact would such a move have? In the short term, it's unlikely to change much. There's very little chance Qualcomm, Apple, Apple or Samsung would attempt to move away from Arm to something like Risc-V. However, in the longer term, Arm would have the relative financial advantage of not having to pay its own inflated licensing fees. Is this a one-off at the behest of Meta, who is desperate to spend money on AI hardware or part of a longer term strategy.

[00:01:50.28]
Ned: The report says that the chip should be released as early as this summer, so we won't have long to wait. If it does anything to accelerate risk five adoption, then you know I'm all about it.

[00:02:04.18]
Chris: Facebook design machine uses AI to read your mind, literally. File this one under, the actual thing I'm talking about is mostly only as terrifying as the way I just described it. Facebook, you may or may not know, does a fair bit of research on the edges of science. The Fundamental Artificial Intelligence Research or FAAR, Fair Unit, for example, located in Paris, focuses on... Well, I mean, it's right there in the name. I'm sure you can figure it out. This week, FAAR, along with the Basque Center on Cognition, Brain, and Language in San Sebastian, Spain, announced that they built a machine that could read a subject's mind and record what they were typing or saying with up to 80% accuracy. They did this by recording brain activity using magno... That's a word. Magnetoencephalography and electroencephalography, M-E-G and EEG, respectively.

[00:03:09.01]
Ned: Well done.

[00:03:10.08]
Chris: Then using the AI bit to interpret what it recorded. So, yeah, this is all super great and doesn't have any negative or malicious use cases, I'm sure you're thinking. On the one hand, I mean, yeah, fair point. We were all thinking that. At the moment, though, the tech required for the meg part of this machine, in particular, requires very specific conditions. It is a large machine that has to be isolated from external signals, and the subject whose mind is being read has to stay perfectly still the entire time. Not to worry, though, smaller, more terrifying, and probably portable versions of this technology are, of course, under active development.

[00:03:59.03]
Ned: Solarwinds selling private for $4. 4 billion from private equity. In case you were wondering if innovation was well and truly dead at SolarWinds, this puts the final nail in the coffin. After suffering a massive supply chain attack in 2020 that rocked the IT industry, trust in SolarWinds never fully recovered. Investment firms Silver Lake and Tomo Bravo proceeded to scoop up 65% of public shares, putting the writing on the wall for an eventual private equity sale. Solarwinds was already well known for selling bloated solutions to enterprises and MSPs who were looking to implement a single pane of glass. The result was more often a single glass of pain, as any solution that claims to do everything does so in a mediocre way. Following a spree of acquisitions in the 2010s, their main source of revenue came from customers who were locked into using their suite of products. A mature product that lacks innovation and has large customers deeply entrenched, that's the time for rent extraction. And PE firm, Turnriver, is ready to drill baby drill. The purchase price is 1850 a share or $4. 4 billion in total, and the purchase has already been approved by the board and shareholders.

[00:05:17.20]
Ned: It is unlikely that any governmental organization will stand in the way, so SolarWinds is ready to sail into Valignor, where they can live an eternal peace, charging their customers forevermore while delivering less and less, as Saaraman intended. No, I don't know why I lapsed in the Lord of the Rings references at the end, and you certainly don't either.

[00:05:41.20]
Chris: Tom Bombadil, of course, still doesn't care about anything that's happening. Yet another study shows that AI is basically just a real expensive copy paste machine. Copilot. We talked about copilot when it was released a long time ago. Eons, in fact, if you take time in podcast time. Back then, we were worried that people were going to just take everything the AI machine says as fact, throw it into their project, make it barely work, then pat themselves on the back for writing 3,000 lines of code every day. Well, fast forward to now, 2025, and yeah, that's basically exactly what happens. Research by the Software Engineering Intelligence platform, GitClear, looked at more than 200 million lines of code authored between 2020 and 2024, and assessed them on code quality metrics. Tldr, they basically found that code quality is eroding. This is not just their data or their subject matter, by the way. The Google team did a similar study and also shows that defect rates in published code are increasing. Cool. Both studies show that it lines up nicely with AI adoption trends across software development. Correlation, causation, blah, blah, blah. I think it's still interesting. One thing that is happening the massive reuse of old code that AI is want to provide you with.

[00:07:21.05]
Chris: So far from writing real code in modular form from scratch, your favorite AI helper is simply doing the fast version of find the solution on Stack Overflow so copy, paste, merge. One concern here is that the software packages are placed in doubt by the use of questionable code. Instead of building new features, the research suggests, future teams will spend more time tracking down bugs due to this poor programming practice. Additionally, it also indicates less and less developers are developing any skills at developing, making this a developing vicious cycle. I need to stop. This getClear report comes hot on the heels of a Microsoft report, asserting that overreliance on AI atrophies critical thinking skills in general. That report was based on self-reporting, basically implying that the subjects of the report knew, or at least suspected that this atrophy is happening, but they're still going to go ahead and overuse AI anyway. Boy, are we in for an interesting rest of the decade.

[00:08:29.09]
Ned: You assume we'll make it that far. All right, that's it. We're done. Go away now. Bye.