Firmus Technologies, an Australian AI infrastructure company valued at $5.5 billion, will build its first data centre in Indonesia through an eight-year partnership with Nvidia. The 360-megawatt Nvidia DSX AI Factory campus in Batam, an island just off the coast of Singapore, is being developed with Singapore-based DayOne and is set to go live in Q1 2027.

Firmus will access up to 170,000 Nvidia AI accelerator chips through 2027 and 2028 via a revenue-sharing and credit-support agreement. The company expects $25 billion to $30 billion in committed offtake agreements during the first six years of the partnership, according to Bloomberg.

The Batam project will be a multi-tenant facility for AI-native customers, unlike Firmus’s Australian projects, which focus on hyperscaler clients. Co-CEO Tim Rosenfield told Bloomberg that market volatility around AI stocks is “largely irrelevant” to how the company is building its business. “We’re building our business based on demand that we’re seeing from customers and contracts that we’re closing,” he said.

Firmus began as a Bitcoin mining operation in Tasmania in 2019. It raised $505 million in April at a $5.5 billion valuation in a round led by Coatue Management and backed by Nvidia. The company has a pipeline of data centre projects across Australia and Singapore, including a deal with CDC Data Centers to develop up to 1.6 gigawatts across Australia by 2028. Asia-Pacific data centre investment has been accelerating sharply, with Blackstone-backed AirTrunk committing $30 billion to India alone.

Rosenfield declined to comment on IPO plans, though the company is widely expected to list this year. The deal adds to Nvidia’s expanding DSX programme, which partners with data centre operators to deploy GPU infrastructure on a revenue-sharing basis rather than requiring upfront purchase. For Indonesia, the campus positions Batam as a regional AI compute hub, leveraging its proximity to Singapore’s financial and tech ecosystem. Demand for AI compute across the region is so intense that even Google has resorted to renting GPUs from SpaceX.

We have all had an Instagram feed go off track. A random Reel catches your attention for a moment, and before long, the app starts serving up the same kind of content again and again.

Instagram already has a way to fix some of that through Your Algorithm, a feature that lets users adjust the topics shaping their recommendations. Now, the company wants to make that tool easier to reach while people are actually using the app.

Instagram head Adam Mosseri has previewed new shortcut-like features that could bring Your Algorithm closer to the main feed and Reels. The changes could make it faster to correct repetitive recommendations, add fresh topics, or reshape a feed without digging through settings.

Instagram wants algorithm controls closer to the feed

Instagram has been working on improving users’ algorithms and giving them more control over recommendations since last year. It recently expanded those controls to the main feed, allowing users to view and edit the topics Instagram thinks they are interested in.

Mosseri’s preview shows how Instagram may bring those controls closer to the places where recommendations actually appear. One version lets users pull down on the main feed to open Your Algorithm. Another shows the same control appearing after a swipe up from a Reel.

Instagram is also testing buttons under Reels that let users tell the app whether they want to see more videos like the one they are watching. In practice, recommendation tuning could become more immediate, since users would be able to correct the algorithm while browsing instead of opening settings later.

A chatbot-like way to reshape your feed

Your Algorithm could also become a more active tool for changing what appears in the feed. Users would not have to rely only on likes, pauses, searches, and watch time to steer recommendations over time.

Mosseri’s preview suggests users may be able to type what they want to see more often, such as positive content, fitness clips, travel videos, or recipe ideas. Instagram would then recommend related topics that users can choose from and add to their feed.

The feature could help users fix recommendations thrown off by accidental clicks on the fly, add new interests, and refresh a feed that has become too repetitive.

The new models will launch with the M5 Pro and M5 Max, according to Mark Gurman.

Yuri Zaporozhets of QRV Systems is a busy chap. He’s built a new RISC-V-based personal computer, a mainframe on an FPGA, and rewritten QNX – twice.

Seemingly every month or two, The Reg FOSS desk gets an email telling us about some astonishing project that he has just got working. We’re delighted to see that his most recent one, a new OS called QSOE, is winning some attention in the FOSS world at present.

But first, we thought we could tell a more complete story of how he got here by describing some of his previous projects.

(By way of a disclaimer, we feel that we should say up front that he does use Anthropic’s Claude LLM to help. To his credit, he does clearly state this.)

GateMate Personal Computer

At the end of 2025, Zaporozhets wrote to tell us about his GateMate Personal Computer. The GateMate PC is something similar to a fairly high-end late-1980s IBM PC-compatible, but instead of a 286 or 386 CPU, it has a 25 MHz RISC-V core.

He told us the main inspiration for the GateMate PC: “The very first computer I saw in my life: an IBM PS/2 Model 30, in 1992. It also started in text mode.” We worked on a few of those, and they were not great machines. The GateMate machine should easily outperform the later, faster Model 30-286. He also acknowledges another project: “the NeoRV32 softCPU by Stefan Nolting is great.”

It has a VGA port that can output 80×30-character text in what back then we used to call Hi-Color, 8 KB of ROM containing a BIOS, and – although it’s still in the early stages – its own OS, which he calls GMDOS. The characters are double-byte ones using UCS-2 Unicode.

The GateMate gets its name from the host hardware because the design is mostly software: it’s implemented on an inexpensive FPGA board, the €50 Olimex GateMate A1-EVB (that’s about £42 or $57). Its video controller is an original design, and he has added extra RAM: the machine has 8 MB of additional PSRAM on two chips, via a QSPI interface.

He blogged about the project, from receiving the board last August to getting a PLL working, to getting video out of it.

The Olimex GateMate board can do a lot more, though – which leads us to his next project.

GateMate System/359

Also implemented on the same FPGA board is Zaporozhets’ miniature mainframe, the System/359. This isn’t a clone of the IBM System/360 mainframe series, the machines that introduced the idea of different computers being software compatible – it’s more of a tribute to it. For starters, the S/359 is a little-endian machine, while the S/390 is big-endian.

So it’s not binary-compatible with the mainframe architecture, but it’s similar. He started the project in January, and later that month, writing about its assembler, said: “GMS/359 keeps what’s beautiful about S/360 – the channel I/O model, the clean instruction formats, the PSW concept – while quietly modernizing the rest. Little-endian bytes. Opcode-first encoding. PC-relative addressing. No more base register juggling. The ‘/359’ isn’t a typo. It’s a declaration: inspired by, not compatible with.”

Zaporozhets told The Register: “There is a working assembler with the POWERFUL macroprocessor – from NASM. I was a NASM contributor from 1999 to 2004 and maintained its RDOFF2 part. Now RDOFF2 is removed from NASM 3.0, but it continues to live in my asm359 project.

“Currently the processor can execute the simple IPL; channel I/O controller works (PS/2 kbd, UART, SYSINFO, even crypto processor (!)). Once I finish the PSRAM module, I will start working with SYS1.NUCLEUS.”

So we can take it that as well as RISC-V and FPGAs, he has some familiarity with low-level systems design. His next project was with an OS that a lot of folks admire: QNX.

Porting QNX 6.4 to RISC-V

Although Zaporozhets wrote to us about this back in March, he also went public with it in a Reddit post: QNX 6.4 kernel ported to RISC-V; petition to BlackBerry to relicense old QNX sources under Apache 2.0.

QNX has an on-again-off-again relationship with FOSS. QNX has been around since the 1980s, as we reported when the company made QNX 8 non-commercial freeware in 2024. In that article, we mentioned that QNX published the source code of an earlier version back in 2007. Back then, QNX was self-hosting and had its own desktop environment – we showed a screenshot of its Neutrino GUI in our roundup of non-Linux PC OSes back in 2013, and GUIdebook has a whole screenshot gallery.

The next year, that QNX 6.4 source code was mirrored from SourceForge over to GitHub, and it’s still there. Zaporozhets took this long-obsolete codebase and ported it to RISC-V, targeting his own FU740 “workstation.” It’s not the whole OS, just “the kernel, the process manager, the C library, the runtime linker.” And the license is very restrictive: you can study it and compile it, but not redistribute it.

He started this effort over Christmas 2020. “The timing made sense in a particular way. RISC-V had matured. The toolchains were stable. The original QNX sources were 32-bit ILP32, targeting x86, ARM, MIPS, SH, and PPC – no 64-bit port existed, let alone RISC-V. Doing the LP64 transition and the architecture port in a single effort seemed like exactly the kind of large, difficult, satisfying project that a long holiday lockdown invites.”

But after the initial effort, it languished for five years. When he came back to it, he ended up with a substantial rewrite. He calls the result QRV.

In March, he described this in a blog post, QRV Operating System: First Publication. As he puts it: “For clarity: QRV is not a patch on the original QNX sources. It is a ground-up reworking of the 32-bit ILP32 codebase into a 64-bit LP64 system for RISC-V, with deliberate simplifications.”

By the end of April, QRV v0.27 could boot to multi-user login. A month later, he declared the project finished with version 0.43: “This is the last development post for QRV. The project set out to port QNX Neutrino 6.4 to RISC-V 64-bit, run it on real hardware, and explore what it would take to bring a clean microkernel architecture to a modern open ISA. Those goals are met. v0.43 is the final release.”

Where next? Well, as the QNX kernel is not truly open source, then the only path forward is to switch to another kernel, one that is truly FOSS. Well, we say “one”… ¿Por qué no los dos?

QSOE – Quick and Secure Operating Environment

The result is QSOE: “QSOE ships in two variants that share one userspace and one build system. QSOE/N runs on Skimmer, a microkernel written from scratch for this project (SMP by design); QSOE/L runs on seL4 as its kernel.”

As The Register covered back in 2014, Secure Embedded L4 is a formally verified microkernel OS. We have written about other OSes that use it before: in 2022, we described the new Neptune OS project, which is a combination of seL4, ReactOS, and Wine. Then, in 2025, we looked at Ironclad, which combines seL4 with the C-based Gloire and an Ada layer.

There are some very serious precedents for building around seL4, but QSOE doesn’t stop there: it also has its own homegrown kernel, called Skimmer, that’s designed for multiprocessor machines. Zaporozhets has been working in this area for a long time. On the QSOE site, he mentions his effort to build a free QNX-like operating system back in 2003. It’s still online – it is called RadiOS.

In How QSOE started, he describes the inspiration: as a fallback option, a Plan B, if his petition to BlackBerry did not succeed. We sympathize. When we wrote about QNX 8, we got nothing useful back from the company either.

Instead, he took all he learned from writing RadiOS and building QRV – via the GateMate PC and mini-mainframe – and did it himself instead. Now he has released QSOE version 0.1. It’s under the Apache 2.0 license, and its source code is on GitLab. He also has a new development blog.

Yes, Claude may be involved, but he was already developing these ideas 20 years before ChatGPT launched. ®

Quordle was one of the original Wordle alternatives and is still going strong now more than 1,400 games later. It offers a genuine challenge, though, so read on if you need some Quordle hints today – or scroll down further for the answers.

Enjoy playing word games? You can also check out my NYT Connections today and NYT Strands today pages for hints and answers for those puzzles, while Marc’s Wordle today column covers the original viral word game.

When my wife and I bought our first ebike—a Radwagon 4 by the Seattle-based Rad Power Bikes—four years ago, we did so to replace one of our two family cars. For in-town trips of 5 miles or less, we figured we could (and should!) use the bicycle. At the time, our kids were very young, so we needed a bike capable of safely carting them around and also handling whatever we were hauling on a given day.

The Radwagon answered those needs; the direct-to-consumer company allowed me to configure the bike to suit my exact needs during the ordering process. I selected a front basket, a rear pad seat for my son, and a Thule Yepp 2 Maxi seat to secure my then-toddler daughter. I also bought a few safety lights and a bell from my local bike shop (more on those accessories below).

Once the bike arrived and was assembled, my wife and I used it to tote our kids all over town. We rode to and from school and daycare, playdates, and doctor’s appointments; made quick grocery runs; and went anywhere else we needed to go that was relatively close to home.

On any given day, the front basket continues to function as a cornucopia holding whatever we might need for the task or errand at hand. On a recent trip to a nearby playground, my ebike’s basket held the following: a small soccer ball, my wife’s small shoulder bag, my bike lock and cable, two bottles of water (in addition to a third bottle of water in the bike’s bottle cage), three baseball caps, two baseball gloves, one baseball, a small tin lunch box full of snacks, and two binders full of Pokémon trading cards. The basket has also successfully transported two large grocery bags or three smaller ones, and, on one occasion, a small guitar amp I found at our local thrift store.

The bike is still useful and functional, but my family’s needs have changed since we bought it. My now-4-year-old daughter is too big to fit in her Yepp seat, and my now-8-year-old son is a bit too self-conscious to be seen on the back of his dad’s big ebike. (Not to mention, he’s now strong enough to ride all over town on his own bike.)

With my kids outgrowing the beloved family ebike, I’ve been thinking about its next iteration as a serious cargo schlepper—a Grocery Getter, if you will—and how I can set it up to haul as much stuff as possible. Ebikes now make up a huge category, serving mountain bikers and commuters, folding and cruising to fit various needs. There are strategic ways to maximize your ebike’s capabilities for each of those purposes, but here I’m going to stick to outlining the two I know best: carting a family (the Family Wagon) and hauling lots of stuff (the Grocery Getter).

The Family Wagon

Photograph: Michael Venutolo-Mantovani

If you use your ebike to transport your kids, you’ll want comfortable, safe, and age-appropriate seating for them.