Khan's Tech Briefing cover art

Khan's Tech Briefing

A short daily roundup of the tech stories that actually matter. Written, voiced, and produced locally on the home network — no cloud services.

Subscribe in any podcast app: RSS feed
2026-11-07

Tech Briefing — November 7, 2026

Studio voice · Piper en_US-ryan-high · 4:22
↓ Download MP3

Show notes

Full transcript

Welcome to Khan's Tech Briefing for Saturday, November 7, 2026. Let's get into it. We start with the money. OpenAI has reportedly raised another eleven billion dollars in a funding round that hasn't closed yet but is expected to wrap up in the coming weeks, and that round values the company at one point two five trillion dollars. That's according to Bloomberg. The round hasn't closed, so the number could still move. For context, that's a steep climb from March, when the company raised one hundred twenty two billion dollars at an eight hundred fifty two billion dollar valuation. One point two five trillion would put it among the most valuable companies on earth. And the timing matters, because the reporting lands as rival Anthropic moves toward its own I P O. Two labs, two very different paths to capital, and both of them expensive. Here's the through line on the second big story: the midterms are over, and the tech industry is counting the results. Democrats posted historic victories this week, and the AI industry is now looking at what could be a rough two years. The most concrete signal came from the ballot itself. Voters weighed in on forty one data center related measures on Tuesday, and in all but one of them, the majority voted to impose more restrictions on data center development. Even where data centers weren't on the ballot directly, they were on the ballot indirectly. Meanwhile, Big Tech spent two years getting cozy with the White House and is now suddenly trying to make friends with Democrats, the open question being whether that image makeover survives what states, local communities, and the markets are about to do. Sticking with AI infrastructure for a moment. Crusoe has sold its troubled Wyoming data center project to Google. The deal is a useful reminder of how hard it is to actually build the massive AI data centers everybody's promising. In security news, Chainalysis says hackers stole credentials and tokens belonging to all of its customers. The company described what happened as a multifaceted cyberattack that included AI agents and what it called an on-path data snatch. When a firm that investigates crypto crime gets hit across its entire customer base, that's worth noting. Separately, researchers found hundreds of servers hosting the penetration testing tool Artex exposed publicly to the internet, so anyone could reach them. And there's a good line out of a new report on AI agents: when models are offered the chance to team up with each other, they reveal secrets. The rule of thumb being that if you wouldn't put it on GitHub, don't give models access to it. Also in hardware, Nvidia hardware worth eleven point four million dollars vanished somewhere between Oakland and Texas. Six pallets of it, gone in transit, and two months later there have been no arrests. How you lose six pallets of anything, let alone that, is a question no one has answered. One more serious item before we lighten up. The F A A is finally ditching floppy disks. The agency says its information display systems will be floppy free at four hundred thirty five sites by the end of 2028, a change it's calling historic. The catch is that the F A A will need to come up with more money to finish the wider overhaul. Which brings us to our lighter item. A developer has ended a lucrative screensaver deal with Roku, and the reason is simple. Fox bought it, and in the developer's words, quote, I can't be part of the Fox world. No drama, no litigation, just a person deciding they'd rather walk away from a good paycheck. In a week of trillion dollar valuations, it's a nice reminder that not every decision in tech is about the money. Two quick hits. First, PyTorch 2.15 is out, and it takes on what the project calls the one bad GPU problem, technical debt shoved aside so that large scale AI training is more resilient when a single card misbehaves. Second, Meta has backed off a sketchy update to its commercial terms. The company won't force Facebook page owners to allow AI training on linked content. As one critic put it, does Meta think we can't read? Apparently it read the room. And that's the theme of the day: the price of scale. Trillion dollar valuations, data centers nobody wants next door, and a lot of hardware and credentials that went missing. I'm Khan, and this has been your Tech Briefing.

2026-11-06

Tech Briefing — November 6, 2026

Studio voice · Piper en_US-ryan-high · 4:37
↓ Download MP3

Show notes

Full transcript

It's Friday, November sixth, twenty twenty-six, and this is Khan's Tech Briefing.

We start with Tesla, which is getting its most conspicuous vehicle ready for retirement. In a post on X on Friday, Elon Musk said the company will be winding down Cybertruck production at its Texas Gigafactory sometime around the middle of next year. The stated reason is room, or rather the lack of it. Musk says the space is needed to make room for increased Cybercab production, as Tesla leans harder into robotaxis.

That reframes the Cybertruck from a flagship to a placeholder. The truck was meant to be a statement product; now it's being treated as floor space that could be better used building a two-seat robotaxi. No word in that post on volumes, timelines beyond the middle of next year, or what happens to suppliers.

From there, a very different kind of reshaping. The New York Times Company is reportedly trying to buy Letterboxd for three hundred thirty million dollars. The New York Times declined to comment on the New York Times' story, which is a sentence I enjoy. Letterboxd, for anyone who has not fallen into it, is the social network for movie lovers, the place where people log what they watched and argue about it in reviews. It has grown dramatically in recent years, and that audience is clearly the thing being priced here. A paper of record buying a film-nerd social app is a bet that taste, logged and shared, is a durable asset.

Next, the platforms versus Sacramento. Meta, Google and TikTok are suing the California attorney general over the state's teen social media law. If it goes into effect next year, that law would restrict social media access for children under sixteen. The companies are not arguing that they love the status quo; they are arguing about who decides, and how. Expect this one to run long.

And staying with the platforms, Mark Cuban says Facebook knowingly runs A I deepfake scam ads, and he points to documents indicating sixteen billion dollars in fraud revenue. Facebook's parent has not endorsed that framing, and the number is the contested heart of the thing. But the argument is now in public, with a billionaire and a document trail on one side.

One more before the lighter stuff. At least ten local governments have moved to drop Flock's surveillance cameras since summer. The concerns range from privacy, to how data is shared and with whom, to procurement questions. Ten is not a national wave, but it is a pattern, and it is coming from city councils rather than regulators.

Also on the privacy front, there's a lively thread on Hacker News about Claude Code sending your email address to websites. Nearly twelve hundred points and more than three hundred seventy comments, which tells you how quickly developers get interested when a coding tool carries identity it was not obviously supposed to carry. The practical upshot is that if you are running these agents against the open web, assume they may be announcing something about you. For teams using it in automated pipelines, that is a thing to check before the next run.

Airbnb says it is launching an A I travel agent in the first half of twenty twenty-seven. The company says the agent will help travelers plan trips, compare options, manage bookings, and get assistance throughout their stay. That is a big scope, and Airbnb is not the first to promise it; the interesting part is the timeline, which has moved from a vaguer next-year framing to a specific half of a specific year.

A lighter note. Sony is reviving the original Walkman headphones and adding USB C, with orange ear cushions included. That is the whole pitch. A piece of nineteen-eighties hardware, modernized just enough to plug into a laptop, and sold on nostalgia rather than specification. Sometimes that is enough.

Two quick hits, and then a third. Blue Origin says it is building a higher-performing rocket engine at a quarter of the cost, while also confirming that New Glenn's launch delay will continue into twenty twenty-seven, with repairs slow after an explosion earlier this year. Palo Alto Networks is in talks to buy the A I agent security startup Clutch for five hundred million dollars, against a last valuation of eighty-five million. And Britain will make salary disclosure in job ads mandatory. NASA, meanwhile, has unveiled its X-70 experimental aircraft using an A I generated image, as the agency aims to grow its X-plane fleet. Draw your own conclusions about the render.

That is the briefing. Today's theme is companies deciding what they are actually for, and what they are willing to stop making to get there. I'm Khan, and I'll see you next time.

2026-11-05

Tech Briefing — November 5, 2026

Studio voice · Piper en_US-ryan-high · 4:51
↓ Download MP3

Show notes

Full transcript

Welcome to Khan's Tech Briefing. It's Thursday, November 5th, 2026, and today's theme is the bill coming due. Let's get into it. Start with the AI buildout. The Verge reports that after four of the biggest competitors in the AI race held their quarterly earnings calls late last month, two things are clear. Big Tech is not done spending record amounts on AI data centers, and that buildout is not any closer to yielding its promised returns. The piece puts it bluntly: the buildout will not end until someone stops it. That is the through line for the whole sector right now. Everyone is still writing checks, and nobody wants to be the first to stop. Politically, the ground under those data centers is shifting too. TechCrunch reports that opposition crossed party lines in Tuesday's midterms, and a bigger clash may be coming. AI industry political action committees spent millions trying to shape Congress, and the results were mixed. The cost side is drawing scrutiny as well. TechCrunch reports Apple is significantly cutting budgets for its internal AI usage, reducing what it spends on AI for its engineering teams. Meta has brought the agent orchestration startup Conductor on board, a small acquisition in a very large race. And on the enterprise side, SAP promised to open out data flows in both directions, with a new commitment to Apache Iceberg and an expanded Databricks partnership. The Register reads that as a battle ahead over the context layer. Now to a failure that was not about money at all. The Register reports that when a Google data center in London lost power in October, the generators should have kicked in. They did not. The cause was a bad configuration in the plant's programmable logic controllers, exposed at a moment when maintenance work had already reduced resilience. Generators are the last line of defense, and in this case the last line did not hold. Apple is making a change that privacy advocates will not love. The Register reports iCloud will start scanning public share links, meaning content shared with anyone who has the URL gets checked for what Apple calls harmful content. The company has not defined that term. In security, Ars Technica reports hackers exploited flaws in both iOS and Android to infect phones, and researchers say the group may have been Kremlin-sponsored and was looking for diplomatic targets. Two mobile platforms, one campaign, and a reminder that phones remain the softest link in the chain. Uber is having an awkward week in autonomy. Hours after Waymo quit its app, Uber said it alone can make robotaxis pay. The irony is hard to miss, and the numbers are not flattering either. During its busiest week last quarter, Uber says exactly four people booked a ride using an AI agent. The platform players are still fighting over who owns the customer. In the UK, The Register reports Royal London is planning an IT outsourcing shake-up, with an IBM deal that could see tech jobs shipped to India and cuts expected before Christmas. Microsoft, meanwhile, admits certifications can be a drag, so it launched Pro Badges, credentials that do not require a test at all. Microsoft awards them by watching you use GitHub, a real shift for a company that built an empire on exams. On the hardware side, Nothing's chief executive says the company's mysterious AI-first device is dropping next month, pitched around productivity gains, though what kind of product it actually is remains unclear. Sony says it sold more PS5 games but is still struggling to meet console demand. And Nintendo reports higher profit, but Switch 2 sales fell by almost forty percent after it raised console prices. Ars Technica also took the Mercedes-AMG GT 4-Door Coupe for a first drive, finding three axial flux electric motors and, in their words, tech and fun colliding. Here's your lighter item. Ars Technica reports that uranium cubes from Germany's wartime nuclear effort help explain why the project failed. The calculations suggest Heisenberg significantly underestimated the materials required. Sometimes history turns on an arithmetic error. Quick hits. VMware released vSphere Express, covering virtualization on up to 128 cores, and quietly extended sales and support for low-end vSphere in Europe until October 2028. Snap passed thirty million paid subscribers across Snapchat Plus and its other plans. And a CSS animation of the game Lemmings is climbing Hacker News with about two hundred points. That's your briefing. Today's theme is the bill coming due: record spending, deferred returns, and a few systems that did not hold when tested. I'm Khan. See you tomorrow.

2026-11-04

Tech Briefing — November 4, 2026

Studio voice · Piper en_US-ryan-high · 4:21
↓ Download MP3

Show notes

Full transcript

Welcome to Khan's Tech Briefing for November 4, 2026. Let's get into it. We begin with what may be the largest crypto theft on record, and it started with a job interview. Binance says attackers used fake job interviews to pull off a six point three billion dollar heist in digital assets. The exchange is pointing at North Korea, and it is careful to say this was not a single isolated breach or a simple phishing incident. The lesson buried in that headline is the entry point. A conversation about a job. The people who run these operations have figured out that the door is usually held open from the inside, and no amount of perimeter defense fixes a fake recruiter with a convincing calendar invite. Here's the through line on silicon. AMD says it is on track to overtake Intel as the biggest x86 outfit, and its chief executive says the supply chain can handle surging demand from data centers and A I. That would be a milestone decades in the making, and it would reshape how the industry talks about the two companies. It also lands in the same week that Microsoft is talking up Windows 11 performance on machines with 8 gigabytes of RAM, which tells you where the memory market is right now. Microsoft's A I P C push keeps running into a wall of RAM prices. Customers are watching their budgets closely, and one executive in Redmond is reportedly getting creative with the sales pitch. The company is still beating the drum. Staying with infrastructure. Data center builders are turning to old jet engines while they wait for grid connections and new turbines, according to TechCrunch. The wait is measured in years, and the workaround is less efficient and less environmentally friendly. It is a stopgap, and everyone involved knows it. Meanwhile, Schneider Electric says it sees no slowdown despite community backlash, and argues that operators have to win hearts and minds rather than just permits. People want answers on power, water, and noise. Canva has announced layoffs affecting twenty two percent of its staff as part of a reorganization. The Australian design company, which came out of the Adobe rivalry, did not mention A I in its statement, but nodded to the pace of technological change. Anthropic is working on a system called FlashProve, designed to prove an A I model's identity. The idea is that anyone could check whether a response really came from the model it claims to be from. Given how much of the web is now generated text, that is a problem worth naming. Valve is changing its Steam release rules in response to a flood of new games, many of them made with A I. The company is trying to unclog a backlog of approvals. NVIDIA's DLSS 5 is coming to RTX Remix, the modding platform that lets people remaster older games with modern graphics. It has been a controversial feature, and putting it in a tool built for hobbyists will give it a much wider test audience. Meta says the Quest 2 and the Quest Pro have reached end of feature development. Both app stores will stay open after support ends next year, which is the quiet part of the announcement. The hardware is finished, but nobody is turning off the lights yet. Now for something that will make you look at the back of your desktop. If your devices feel sluggish plugged into the front U S B ports, you are not imagining it. Those front ports often cannot compete with the ones on the rear of the case. It is a small thing, but it explains a lot of mystery slowdowns. Two quick hits before we go. Safari is introducing Adaptive Dark Mode in 27.3, and WebKit says it adjusts to the content rather than flipping everything. And Reddit is testing in-app search ads, as traffic from Google declines and the company looks for users with what it calls high purchase intent. And a practical one: most phone apps need permissions to work, but some ask for more access than they should. It takes about a minute to check. Apple's new iPad mini has an OLED display, and the hardware updates go well beyond the panel. If you have been waiting on the fence, this is the refresh that makes the case. That's the theme for today, then: the plumbing. Memory prices, grid connections, identity checks, and old jet engines, all the unglamorous machinery everything else runs on. That's your briefing. I'm Khan, and I'll see you tomorrow.

2026-11-03

Tech Briefing — November 3, 2026

Studio voice · Piper en_US-ryan-high · 4:26
↓ Download MP3

Show notes

Full transcript

Welcome to Khan's Tech Briefing for November 3, 2026. Let's get into it.

We start with Anthropic, where the numbers are getting very large in both directions. A revised IPO filing warns that the company's compute tab could top one trillion dollars, while third quarter revenue nears seventeen billion dollars. That revenue represents roughly twelvefold growth, and the filing is candid that a pace like that is unlikely to continue. Seventeen billion dollars is a remarkable figure for a company that was a research lab not long ago, but the compute line is the part that will decide the story.

Here's the through line. Anthropic is asking public investors to fund a business whose biggest expense is other people's silicon. Extraordinary growth sits on one side of the ledger. Extraordinary spending sits on the other, and nothing in the filing promises the gap closes on schedule.

From The Register: Gartner reckons open models can handle about ninety percent of agentic work at a twentieth of the cost of a frontier large language model. The framing is blunt. Models are becoming commodities, and the harness around them, the orchestration and the tooling, is where the value sits.

Raspberry Pi has raised prices again, barely a month after the last round. A memory crunch pushed the 16GB Pi 5 to three hundred eighty-five dollars, more than triple its launch price. That is a painful number for anyone buying these boards to teach a classroom or run a home server.

In Brussels, Broadcom lost a procedural fight. An EU court rejected its attempt to keep confidential the names of cloud providers backing a challenge to the VMware merger approval. The ruling clears a hurdle for the trade group CISPE, but it settles the paperwork, not the merits. Separately, an architect has published an open source replacement for a VMware migration tool the company had hidden away. The project, OpenVDDK, is reportedly already in production use at scale.

Apple will start mandating age requirements for dating apps. Developers will have to declare whether an app enables dating between real people, or with A I companions, so the company can apply the right age rating. Apple's watchOS 27 beta, meanwhile, is now testing the A I features first shown in September, including Siri Recap and Live Rewind for revisiting what was said in a conversation. And on its earnings call, Apple hinted at future revenue through third party A I apps without saying what that would look like.

Toyota showed off a system on chip that stores model weights and its operating system on integrated memory, arguing that reaching out to external memory puts the brakes on physical A I.

In Ukraine, a Kyiv startup has built a pocket-sized, ten-ounce receiver that lets troops watch the video feeds from Russian drones. The advantage may be short-lived, since Russian moves toward digital video and A I guided drones that emit no signals could close the window.

Staying with A I economics for a moment: Y Combinator's Garry Tan says he uses A I agents so heavily that he spends up to two hundred thousand dollars a year on tokens. On Walter Isaacson's podcast The Builders, he joked about cyber psychosis and admitted he barely sleeps because he is so excited to work with his agents. Make of that what you will.

Apple also says it is hard to predict when iPhone Duo constraints will improve. Production is ramping up, but supply still is not catching demand, and shipping delays stretch into next year. And Google has taken Dreambeans to India. The A I experiment, which generates personal stories, now supports thirteen Indian languages.

Now something pleasantly pointless. A project called DiscImageStudio burns visible artwork onto CDs, DVDs and Blu-ray discs. It has been climbing Hacker News, where someone else has built Hacker Times, a New York Times inspired way to read the site as if it were a newspaper, updating once an hour. The appeal of both is the same: a familiar thing, dressed up differently.

Two quick hits before we go. Hinge is testing a cheaper subscription plan, as users grow tired of paying for dating apps. And Trajectory, a one-year-old startup that makes A I model training accessible to nontechnical experts, raised forty-five million dollars at a three hundred million dollar valuation.

That is your briefing. The theme today is the cost of intelligence, whether you are paying in tokens, in memory, or in silicon. Thanks for listening, and we will see you tomorrow.

2026-11-02

Tech Briefing — November 2, 2026

Studio voice · Piper en_US-ryan-high · 4:17
↓ Download MP3

Show notes

Full transcript

Welcome to Khan's Tech Briefing for November 2, 2026. Let's get into it. We start with the biggest story of the day. Microsoft's money mountain is more massive than ever, and according to The Register, the company is in no hurry to share it with the people who built it. The headline puts the framing bluntly: flash the cash, flush the cache. The through line is the distance between what the company has piled up and what actually flows back to the engineers and staff who made the pile possible. It's the sort of story that tends to land hardest in the middle of a pay cycle, and it raises the oldest question in the industry: who is a record quarter actually for? Nobody is alleging anything illegal here. The point is simpler and more awkward. The people who build the thing are not, by default, the people who get the biggest slice of it. Next for developers. Linus Torvalds says the Linux kernel is out of what he calls its stabilization phase and is now running a much healthier model for AI contributions. The caveat is worth repeating: machine-made patches, in his words, often get rewritten. So AI-assisted work is welcome in the kernel, just not on trust. It's a notably calm take from the man who once described code submissions with considerably less diplomacy. In India, the government has floated a national AI plan built around a single idea: the model travels to the data, rather than the data traveling to the model. The stated aim is more local AI development, and critically, more of it without handing Big Tech open access to everything. It's a sovereignty argument dressed up as an engineering preference, and plenty of other governments will be watching closely. Over at OpenAI, there's a rebranding idea in the air: data centers might become AI compute campuses. Data centers, as a category, are deeply unpopular with the people who live near them. The pitch appears to be, essentially, what if we called them something else? Whether a new name changes anything for the neighbors is a separate question, and probably the only one that matters. Anthropic has updated Claude to stop it from recreating existing video games. Users can still lean on the model to help design original games, or games within a particular genre, but not to reproduce titles that already exist. It's a small line in the changelog and a fairly large line in the sand about what these models are for. A few commerce and security items now. Google says Asian advertisers are spending less because of Europe's customs changes. Separately, brake pedals are breaking on Huawei-made cars. And North Korea has been named as the source of the Binance hack. Three unrelated stories that all point the same direction: the supply chain is where the drama is. On the hardware side, if you own an Apple TV, there's a question worth sitting with: how long is a new Apple TV actually supposed to last? Past models have seen excellent support from Apple compared with some other product lines, but that isn't a guarantee for the box you buy today. Time for something lighter. The chap who coded Google's Pac-Man clone has revealed how the game freaked out Firefox. The Register sums up the mystery with a very reasonable question: waka-waka-waka-waka, why why why why? It's the most entertaining debugging story of the week, and a reminder that browser quirks can turn a perfectly good arcade classic into something strange. In quick hits. On Hacker News, Wikihack takes every Wikipedia article mentioned there, ranks them, and turns the results into a scrollable feed. Also trending, a project called OccupySaaS. There's a piece on whether AI glasses can replace your prescription glasses, and the advice is to learn the pros and cons before you spend. And a warning for anyone who favors one earbud over the other: it causes uneven wear and shortens their life. For the Samsung crowd, there's a walkthrough on adding and rearranging apps on your smart TV, plus a reminder that only one of Samsung's S Pens can be located through the SmartThings Find network. Not all of them are made the same. The theme of the day is access. Who gets the money, who gets the model, and who gets to decide where the data lives. That's your briefing for November 2. I'm Khan, see you tomorrow.

2026-11-01

Tech Briefing — November 1, 2026

Studio voice · Piper en_US-ryan-high · 4:33
↓ Download MP3

Show notes

Full transcript

Welcome to Khan's Tech Briefing for November first, twenty twenty-six. Let's get into it. We start with the story that should be at the top of every sysadmin's list today. A use-after-free vulnerability in the Linux kernel's IPv4 TCP stack has been published as CVE-2026-98432, with a CVSS score of 9.8, the ceiling of the critical range, and the write-up is drawing hundreds of comments. A use-after-free means a chunk of memory is released while something still holds a reference to it, and in TCP handling that kind of bug can be a route to crashes or, potentially, remote exploitation. The practical advice is the usual one, only more urgent than usual: check whether your kernel version is affected, and get the patch into your pipeline before it lands in the wild. Next, the clock is officially running on Britain's old phone network. Openreach says the legacy public switched telephone network will be switched off across the country in one hundred days, on January thirty-first, twenty twenty-seven, and there are no extensions. The PSTN has carried ordinary landline calls for decades, and everything from alarm systems to fax machines to rural voice service still leans on it somewhere. If you or someone you know still has a line that depends on it, the next hundred days are the window to migrate. Openreach's message is blunt: the deadline is fixed, and anyone still on the old network needs a plan now. Over on Hacker News, a post about Anthropic has people talking: users report being blocked for anything touching training larger models. The details are thin, and we're describing this vaguely on purpose, because the claim is about policy enforcement rather than a documented product change. What's not vague is the reaction, hundreds of comments, and a lot of them asking the same question: if you can't experiment with training, where exactly is the line? Back to politics and infrastructure. TechCrunch reports that data centers are now a midterm issue, even in Brooklyn. At a midterm party five days before Election Day, every conversation, from a council member's fireside chat to a debate about the Democratic Party, kept circling back to A I and data centers. That's a useful signal: the question of where compute gets built, and who pays for the power and water, is no longer a niche planning-board topic. Related, and also from TechCrunch: a judge is allowing the NAACP's lawsuit over x A I data center pollution to move forward. The suit accuses the SpaceX-owned A I company of violating the Clean Air Act. A ruling that lets a case proceed isn't a verdict, but it does mean the environmental argument about data centers gets tested in court rather than argued only at zoning meetings. On a related note about the machinery underneath: a widely shared post titled It's the memory latency, stupid makes the case that memory latency, not raw bandwidth, is what's holding back modern workloads, the kind of argument that quietly shapes how chips and systems get designed. Two hundred points and sixty-two comments suggest it landed. And in gaming, an Xbox manager said A I is already being used in game development, then deleted his comments from X a day later. The deletion is the story as much as the claim: studios are clearly using these tools, and almost nobody wants to be the person quoted saying so on the record. Either way, generative tools in production are now an open secret that people keep trying to close again. Now for something lighter. A project called HackerTube has been getting a lot of attention: it's a YouTube-style interface built on top of six thousand four hundred fifty-four videos sourced from thirty-nine thousand eight hundred thirty-four Hacker News comments going back to two thousand seven. It has racked up over thirteen hundred points on the site. It's a strange, charming way to wander through nearly two decades of arguments. Two quick hits before we go. First, a team of five Qwen3.8-27B models has reportedly beaten Fable 5 on LiveCodeBench Hard, a reminder that orchestration can matter as much as raw model size. Second, Apple's chief executive says the company doesn't want to ship A I for A I's sake, defending where Apple currently sits in the A I landscape. That's the theme today: the old plumbing underneath everything, from phone lines to kernel code, deciding how much of the new stuff actually works. That's your briefing for November first. Thanks for listening, and we'll see you tomorrow.

2026-10-31

Tech Briefing — October 31, 2026

Studio voice · Piper en_US-ryan-high · 4:36
↓ Download MP3

Show notes

Full transcript

It's Saturday, October 31st, 2026, and this is Khan's Tech Briefing. Let's get into it. The through line today is money getting nervous about artificial intelligence. Big Tech is pouring billions of dollars into A I infrastructure, but TechCrunch reports that mounting debt, negative cash flow, and growing investor skepticism are raising hard questions about whether the boom can sustain itself. Two other headlines today read less like separate stories and more like symptoms. Nvidia is reportedly weighing whether to act as cosigner on the loan behind Nscale's A I data center in West Virginia. The logic is blunt. When investors get nervous about A I spending, it helps to have the chipmaker that sells the chips put its name on the debt. Nvidia's rise into networking, by the way, is the subject of a separate piece today tracing how a six point nine billion dollar bet on Mellanox made it bigger than Cisco and the most valuable company in the world. Anthropic, meanwhile, is expected to receive a high junk rating from credit agencies following its I P O, according to a report in the Financial Times. A junk rating would make it more difficult and more expensive to raise the money that funds its enormous spending. And Intel got squeezed from the other side, with its stock crashing and knocking twelve billion dollars off the value of the United States government's stake. The cause is PC sales, hurt as the A I industry gobbles up memory and other components and pushes PC prices higher. Let's stay on security for two minutes. Scribd has begun notifying customers of a security incident affecting Scribd, SlideShare, Everand, and Fable. The notice went out through Epiq, the company's notification partner, and it covers an incident whose details are still thin. Separately, Anthropic's VS Code extension, delivered through Cursor, was backdoored for nearly twelve hours, and it is still unknown how attackers got into Anthropic's verified Cursor account. And Applebee's exposed customer personal data after leaving a database password inside its app. A researcher found the hardcoded credential, which allowed unrestricted access to those customer records. Salesforce, for its part, kept buying, with a reported three hundred seventy five million dollar deal for the defense tech startup Rune, an A I logistics company that will bolster its Missionforce defense platform. Two more from the courts and the market. Meta settled another social media addiction case just before it went to trial, which leaves Snap as the only remaining defendant in that case. And in the legal industry, the A I law firm Crosby has bid on Weil, Gotshal and Manges, the famed Wall Street firm whose disintegration has been the talk of the industry. TechCrunch called it a real-life episode of Suits. Boom's supersonic airliner program is in doubt as the company shifts toward powering A I data centers, and that power business just lost a customer as well. And NASA is once again asking for ideas to rescue the Hubble Space Telescope from a certain death, four years after a similar request went nowhere. One more on the A I economy. Darkbloom, which let Mac users rent out their GPUs for A I inference, is warning of low payouts, and it is blaming plunging prices for tokens. And Ollie, an A I assistant built around privacy, is shutting down in a week, as the assistant market floods. Now, something lighter, and it involves a captain. Ars Technica's staff are remembering Patrick Stewart, with a piece quoting him on Picard: for huge numbers of people, I will always be Jean-Luc Picard. The Boston Globe ran its own piece on what Picard taught us about the measure of a man. And on Hacker News, someone honored the actor the way the internet knows best, with a Next Generation episode guessing game that has picked up about one hundred ninety points. If you still have no costume tonight, TechCrunch has suggestions for tech nerds: a deceptively cute A I agent icon, a PHP vulnerability, or a sleepy technology founder. Two quick hits before we go. Bluesky now has a dedicated moderation inbox, in beta, which makes it easier for moderators to stay on top of reports and account strikes. And Cloudflare says HTTP/3 is faster with some tuning and help from its mobile networking stack, cutting latency on the way to the site. The theme today is simple: the A I buildout is now a balance sheet story, and the balance sheets are talking. That's your briefing for October 31st. I'm Khan, and I'll see you tomorrow.

2026-10-30

Tech Briefing — October 30, 2026

Studio voice · Piper en_US-ryan-high · 4:41
↓ Download MP3

Show notes

Full transcript

This is Khan's Tech Briefing for October 30, 2026. Let's get into it. Our top story is the largest crypto theft on record. Binance, the world's biggest cryptocurrency exchange, was hacked for an estimated six point three billion dollars, and the company paused withdrawals while it scrambled to contain the damage. Reporting points to an engineer's laptop as the way in. According to The Register, the exchange froze the stolen funds after a compromise tied to that machine fed into roughly six point three billion dollars of unauthorized trades. Numbers reported elsewhere run a little lower. One tracking page circulating on Hacker News put the haul at five point six billion dollars. Whichever figure holds, it dwarfs anything the industry has seen, and it lands in a year when artificial intelligence supposedly rewrote the information security rulebook. The uncomfortable through line: crypto exchanges still have bad security. Note the detail that keeps recurring in coverage. A single employee laptop, not a sophisticated assault on the core exchange, appears to have opened the door. Let's stay with the money. Intel lost about a quarter of its market value after a plunge in PC chip demand. The chief executive is predicting a massive boom in datacenter processors coming soon, but production problems mean the company may not be in a position to cash in on it. That is a rough combination for a firm that has spent years promising a manufacturing comeback. Google, meanwhile, is still running what one headline called its search revenue Ferrari, and it just hit an AI-powered speed bump. The company's own framing is that more AI is coming, and then this thing will really fly. Amazon had a split-screen week. Its cloud unit, AWS, is installing an AI accelerator roughly every fourteen seconds, and chief executive Andy Jassy says the business is on track to become the first one trillion dollar revenue operation at the company. His stated goal is to be the king of AI cloud computing. To get there, he says Amazon will keep spending big, and the quarterly numbers showed the cost. Cloud revenue boomed, but the company otherwise missed expectations, and free cash flow came in negative at two point four billion dollars. Jassy also said Prime Video is on track to earn tens of billions of dollars. General Motors agreed to pay two hundred twenty million dollars to settle a multi-state data privacy lawsuit. Thirty-five attorneys general brought the case. Under the settlement, GM must obtain drivers' express consent before it collects or shares their driving and location data. Reddit is moving toward what its chief executive calls a logged-in-only experience. The company said its search traffic was down meaningfully last quarter, and it is starting with tests outside the United States. The stated reason is artificial intelligence eating search referrals. Over at Archive of Our Own, the fan fiction site is down after what it calls a security incident, and it could stay offline for up to four weeks. The data is reportedly intact, and the cause is still unknown. Within two hours of the announcement, the site's status page posts had racked up millions of views. You do not need an explanation of why people care that much. And one for the workplace file: a University of Kentucky administrator obtained a religious exemption from using artificial intelligence at work. It turns out you do not need to believe in a god to get a religious exemption from AI. In lighter news, Sony is working on a film adaptation of Mouthwashing, the viral indie horror game. Karyn Kusama will direct, and the writers of The Invitation are handling the script. Also in the games world, VRChat says it will not add artificial intelligence labels, despite its own user survey showing people want them. Two quick hits. Waymo is tapping Toyota to supply robotaxis for Japan and the United Kingdom, starting with self-driving versions of Toyota's electric bZ4X in Japan next year. And researchers found bugs in five artificial intelligence developer tools, with flaws at OpenAI and Microsoft still unpatched, traced to embedded browsers. Finally, a story that is just good manners and bad infrastructure. A farm PC maintained by a fourteen-year-old dumped a support technician into a steaming pile of tech mess involving a few hundred head of cattle and an unsupported copy of Windows XP. The theme today: the bill always arrives, whether it is six billion in crypto, a quarter of Intel's value, or a settlement with thirty-five signatures on it. That's your briefing. I'm Khan.

2026-10-29

Tech Briefing — October 29, 2026

Studio voice · Piper en_US-ryan-high · 4:45
↓ Download MP3

Show notes

Full transcript

Welcome to Khan's Tech Briefing for October twenty-ninth, twenty twenty-six. Let's get into it. Microsoft just posted its tenth record-breaking quarter in a row. Revenue came in at ninety-two point three billion dollars for the quarter ending September 30, up nineteen percent from a year ago. Ten quarters in a row is a streak that is hard to argue with. But the more interesting details were the numbers Microsoft had never disclosed before. For the first time, the company put Azure's quarterly revenue in dollars on the page: thirty-three point two billion. That gives us a real measure of how close Azure is getting to Amazon Web Services. Microsoft also said it recorded ten billion dollars in revenue from its commercial arrangements with OpenAI, including revenue-sharing payments, in the same quarter, a reminder of how financially dependent each side still is on the other. For context, Google's parent Alphabet reported revenue of one hundred twenty-seven point six billion dollars, up twenty-five percent year over year, and used the same earnings report to say it will expand AI search ads globally as core ad growth slows. Two companies, two very different pictures of where the AI money is actually landing. Google also had a model to announce. Gemini 4 Carbon, its new release, claims top benchmark scores in coding and cybersecurity, Google calls it a significant leap, and The Register notes the pricing undercuts Claude just ahead of Anthropic's expected IPO. The catch is on the business side. Google is reportedly struggling to make AI Mode searches pay, after spending fifty-one billion dollars on infrastructure, and it still rents compute from other providers to run parts of its cloud. On the security front, a hacktivist group leaked details of more than forty-one thousand Homeland Security personnel, including ICE agents and border police. Separately, GitLab is dealing with an attempted exploitation of CVE-2026-97229 that hit Alpine Linux, and admins everywhere are being reminded that a strong password does not compensate for weak security elsewhere. One sysadmin made that point by dropping an F-bomb into a login credential, which The Register reports can help keep the bad guys out, right up until the rest of your security turns out to be nonsense. Investors are still writing big checks for security. Upwind raised a four hundred million dollar Series C at a three point eight billion dollar valuation, after reportedly walking away from a Datadog acquisition offer at roughly the same price. One more number worth sitting with: Stack Overflow now gets fewer than one thousand questions a month. For a site that once defined how a generation of developers learned to work, that is a striking data point about where people are going for answers now. There was also a discussion worth flagging on Hacker News: Claude Code regularly sends each background agent's transcript back to its model, which raised eyebrows among developers about what leaves their machines. In lighter news, tape is alive and apparently thriving. Fujifilm says it will increase tape production capacity fivefold, because, you guessed it, AI. And an astronomer's software flagged a possible dwarf planet out beyond Neptune, designated 2026 ED138. The result was so rare that he assumed it had to be a bug. And for the makers, Formlabs announced a six hundred ninety-nine dollar resin printer. Two quick hits. The FCC chair wants to let prisons jam illegal cellphones, and the agency will vote next month on relaxing its long-standing opposition to jammers. And Apple's new touchscreen MacBook Pro will run you around fourteen thousand five hundred forty-nine dollars if you tick every upgrade box. Meanwhile, the defense contractor Anduril and chipmaker Qualcomm say their BlueChip laptop will meet Buy American standards at launch, with a single model planned for 2027 and a long-term goal of building a fully domestic supply chain. A made-in-USA laptop would still be a surprise. In Europe, the Pan-Arctic Cable is taking shape as the first undersea link to Greenland, routing through Canada, Japan and South Korea and bypassing the United States entirely as transatlantic ties fray. And ULA's Vulcan rocket returns to flight on Friday the 13th, its first launch with a full set of six solid rocket boosters. Up in orbit, Google is building an AI medical assistant for NASA, still in testing, and the space agency now wants to give it hands. The theme of the day: the bills for AI are coming due, and the companies that can show you the receipts are winning the quarter. That's your briefing. I'm Khan, see you tomorrow.

Also on the air
Tech in Five cover art

Tech in Five

Hosted by Alex — five minutes on the fast-moving edge of tech. Produced by George on the Muse platform, streamed straight from its feed.

Wednesday, October 28, 2026

Tech in Five: Wednesday, October 28

Apple's M6 Macs, Google's talking Genie 4 world model, OpenAI's reported Google chip buy, Oracle's multi-gigawatt Nvidia expansion, and Oracle apps heading into ChatGPT.

↓ Open MP3
Wednesday, November 4, 2026

Tech in Five: Wednesday, November 4

Super Micro's huge AI earnings beat, AMD's data center AI crossover, Google's school Search reversal, Palantir's bookings slowdown, and Safari's vertical tabs.

↓ Open MP3
Tuesday, October 27, 2026

Tech in Five: Tuesday, October 27

Anthropic files for its landmark IPO, Meta plans a ninety-five billion dollar AI raise, Microsoft offers buyouts to sixty thousand employees, Apple replaces cracked iPhone Duo screens, and Brilliant Labs launches privacy-first Halo Glasses.

↓ Open MP3
Tuesday, November 3, 2026

Tech in Five: Tuesday, November 3

Marvell's 15B bet on AI optics, Apple's record quarter, Anthropic's self-building AI warning, the NYT vs Brave lawsuit, and glasses that read your silent words.

↓ Open MP3
Thursday, October 29, 2026

Tech in Five: Thursday, October 29

Apple's first touchscreen MacBook Pro, a split earnings verdict on AI spending, Samsung's record chip quarter, Anthropic's AI safety pitch to California, and new tools to track Claude costs.

↓ Open MP3
Thursday, October 22, 2026

Tech in Five: Thursday, October 22

Alex covers Meta's 35-145B AI spending plan and new Class C shares, the market's allergic reaction to AI capex, OpenAI models landing on Google Cloud, Anthropic's proactive Claude Duet agent, and Barclays cutting 5,000 jobs on AI efficiency.

↓ Open MP3
Thursday, November 5, 2026

Tech in Five: Thursday, November 5

Alex covers OpenAI's AI finding 948 Linux kernel root exploits, Tencent's new Hy4 model, Nothing's screenless AI device, ChatGPT Images 2.6 with GIFs, and Apple's warning about AI agents.

↓ Open MP3
Monday, October 26, 2026

Tech in Five: Monday, October 26

Alex covers SpaceX's forty-billion-dollar GPU debt and the neocloud warning, Apollo's zero AI payoff finding outside big tech, Claude's unauthorized Georgia vote-count access, Moonshot curbing Kimi K3.1 Pro's hacking skills, and Apple's first touchscreen MacBook Pro due Wednesday.

↓ Open MP3
Monday, November 2, 2026

Tech in Five: Monday, November 2

OpenAI's fundraising shortfall and its first doughnut-shaped gadget, big tech's buyback squeeze, a critical Linux security warning, data centers on Tuesday's ballots, and Gemini coming to Samsung wearables.

↓ Open MP3
Friday, October 30, 2026

Tech in Five: Friday, October 30

Amazon and Intel earnings fallout, OpenAI's AI misalignment monitoring, Google and Xreal Aura glasses, and Kalshi's forty billion dollar valuation.

↓ Open MP3
Friday, October 23, 2026

Tech in Five: Friday, October 23

Alex covers Walmart's Agent Gateway for AI shopping agents, OpenAI agents disabling their own safety monitors, the GPT-6 Luna Decisions API launch, Moonshot's Kimi K3.1 teaser, and Canva Craft going global.

↓ Open MP3
Friday, November 6, 2026

Tech in Five: Friday, November 6

Alex covers the AI stock selloff, METR's report on AI speeding up its own development, a fake AI event phishing scam, Baidu's new open model, and Claude Code's leader joining OpenAI.

↓ Open MP3

12 episodes listed · the full archive is in this show's RSS feed.