AMD’s answer to Nvidia’s DGX Spark AI workstations, codenamed the Ryzen AI Halo, will be available for pre-order later next month for anyone with $3,999 burning a hole in their pocket.

That might sound like a lot for an AI mini PC, but don’t worry. Compared to cloud APIs, it practically pays for itself. Or, well, that’s AMD’s sales pitch. The House of Zen argues that if you spend eight hours a day vibe coding, the system could save you $750 a month.

Whether this helps you justify paying for hardware that less than a year ago could be found for between $2,200 and $2,999 or not, it’s (probably) not AMD being greedy here; the RAMpocalypse has been hard on everyone.

Much like the DGX Spark, which now retails for $4,699, up from $3,999 when we reviewed it last fall, AMD’s rendition aims to provide a curated developer environment for running local models and agentic AI frameworks.

This is really the core value proposition behind both of these devices. They aren’t the most powerful or the fastest AI systems, but they’re able to run models that a few years ago would have cost $20K or more.

A little box of TOPS

The diminutive system measures in at 5.9 x 5.9 x 1.7 inches (150 x 150 x 43 mm) and is powered by a 120 watt Ryzen AI Max+ 395 APU, better known by its codename Strix Halo. 

The chip is backed by 128 GB of LPDDR5x 8000 MT/s memory, which feeds both its 16 Zen 5 cores and 40 RDNA 3.5 GPU compute units, providing up to 256 GB/s of bandwidth, more than a Ryzen 9000 Threadripper (non-Pro) system.

For local AI enthusiasts, that’s enough to run models up to 200 billion parameters in size at 4-bit precision — just like the more expensive Spark.

The bulk of the Ryzen AI Halo’s compute comes from its integrated graphics, which are capable of delivering roughly 56 teraFLOPS at 16-bit precision.

While impressive for onboard graphics, that’s still between 55 and 88 percent slower than what the DGX Spark advertises.

Unlike the Spark’s Blackwell-based GB10 APU, Strix Halo doesn’t support FP8 or FP4 data types in hardware. At BF16, the Spark delivers 125, at FP8 250, and FP4 500 teraFLOPS. Double those figures if you happen to find a workload that can leverage Nvidia’s 4:2 sparsity.

That performance discrepancy won’t necessarily be obvious in every workload. In fact, in LLM inference, AMD claims the AI Halo generates tokens 4-14 percent faster than the Spark.

The lower end of that roughly matches what we saw when we pitted the Spark against a similarly equipped HP Z2 Mini G1a back in December. The G1a packs the same silicon as AI Halo, and in Llama.cpp with the Vulkan backend, eked out a small but meaningful lead over the Spark in tokens per second generated.

However, the speed any GPU can generate tokens at is largely dictated by effective memory bandwidth, not floating point performance. GPU compute has a much bigger impact on things like prompt processing time.

In our testing, the Spark’s more capable tensor cores gave it a 2x to 3x lead in prompt processing. For shorter prompts, this isn’t all that noticeable, usually the difference between waiting 100 ms versus 200 ms or 300 ms, but for longer prompts, it did become more pronounced.

We saw the Spark take similar leads in our image generation and fine tuning benchmarks, but it’s worth noting that AMD’s software stack has matured greatly since our initial review and the performance gap has likely closed somewhat since then.

AMD’s AI Halo does have two things going for it that can’t be said of the Spark. Alongside the GPU is an XDNA 2-based neural processing unit (NPU) that AMD rates for 50 TOPS. What good that’ll do you depends heavily on the application in question. Many content creation apps have now been updated to take advantage of it, but the number of generative AI inference engines that could properly harness it was quite limited the last time we looked.

The second thing AMD’s Ryzen AI Halo has going for it is that it’s a standard x86 box at its heart, and you can run Windows or your preferred flavor of Linux on it if that’s more your style. On the Spark, you’re stuck with a lightly customized version of Ubuntu 24.04. Beyond that, you’re coloring outside the lines.

Particularly for developers building for Microsoft’s NPU-accelerated AI PC ecosystem, this is an obvious advantage.

In terms of networking, AMD’s Spark-clone falls a bit flat. One of the hallmark features of Nvidia’s AI workstation is a 200 Gbps ConnectX-7 NIC, which allows for clustering of up to two and eventually four systems.

AMD’s AI Halo has a single 10 Gbps NIC, which should help with downloading large model files in a timely manner. In theory, the system should be able to achieve high-speed networking over USB-4, but it’s not clear whether this is actually a supported use case.

That said, Apple has already demonstrated just this using RDMA over Thunderbolt, so it should work so long as AMD has a playbook for configuring RDMA on its systems.

AMD’s own AI lab

As we mentioned earlier, much of the Ryzen AI Halo’s value proposition comes from being validated hardware with well documented playbooks for common use cases and known good software.

Finding the right combination of device drivers, ROCm, HIP, SYCL, CUDA, PyTorch, TensorFlow, and JAX has long plagued the AI/ML devs, regardless of which ecosystem you opt for.

Having validated environments for workloads, whether it be vLLM, Llama.cpp, Ollama, ComfyUI, or something else ensures users spend more time doing something productive than debugging mismatched dependencies.

At launch, AMD says the Ryzen AI Halo will ship with five preinstalled playbooks, with another 10 available online and additional playbooks to be added monthly.

Additionally, customers will gain access to AMD’s developer program, cloud credits, and exclusive playbooks. 

More memory on the way

The 128 GB Ryzen AI Halo will be available for pre-order next month starting at $3,999, but if that isn’t enough for you, AMD is already prepping a higher capacity version of the system with 192 GB of memory on board.

That system will feature a refreshed Ryzen APU in the AI Max+ 495, which just like the rest of AMD’s 400-series lineup gets a modest clock bump to the CPU, GPU, and NPU, and not a whole lot else.

Still, 192 GB of unified memory opens the door to even larger, more capable models, if you can stomach the presumably higher asking price. ®

Qivalis plans for a market launch in the second half of 2026.

Bank of Ireland and AIB have joined a consortium of European banks working towards issuing a euro-denominated stablecoin.

The two Irish banks joined the Qivalis consortium alongside 23 other banks – including Cecabank, Erste Group, Groupe BPCE and the National Bank of Greece – taking the group’s total strength to 37 banks from across 15 countries in the region.

The group wants to introduce euro-denominated stablecoins to expand Europe’s financial infrastructure and compete with US-backed versions, which make up the overwhelming majority of stablecoins in circulation. Big names such as Danske Bank, ING and KBC joined the consortium last September.

Stablecoins are a digital asset generally linked 1:1 to a chosen fiat currency. These digital assets allow for round-the-clock liquidity and near instant settlements.

The combined market capitalisation of all stablecoins exceeds $300bn, with around 90pc of it held by Tether and Circle; euro-denominated stablecoins only total around €395m, according to EU figures from November – or around 0.2pc of the total in circulation, according to Qivalis.

“The euro is Europe’s currency, and on-chain financial infrastructure should carry it – built by European institutions and governed by European rules,” said Jan-Oliver Sell, the CEO of Qivalis.

Formed in 2025, Qivalis is currently in pursuit of European regulatory approval to become an authorised electronic money institution. The consortium is targeting a market launch in the second half of 2026.

“We are investing in this consortium because we believe Europe needs trusted, regulated innovation in payments and settlement,” said Geraldine Casey, the managing director of retail banking at AIB.

“Qivalis will provide access to a euro-denominated stablecoin that is being developed to operate within the EU regulatory framework.”

As part of the consortium, AIB has said that it will collaborate with other European banks to innovate payments systems.

“This is a practical step for AIB to learn, innovate, test and collaborate with other leading European banks, and to help shape how new forms of digital money can be used safely, responsibly and within the regulated banking system,” Casey added.

The European Central Bank, meanwhile, has moved to the next phase of the digital euro project, with plans for a pilot digital euro project in mid-2027 and issuance in 2029.

Howard Davies, Qivalis’s chairperson, added: “This infrastructure is essential if Europe is to compete in the global digital economy whilst preserving its strategic autonomy.

“The euro’s role in the eurozone’s monetary system will increasingly depend on whether it is present – as the primary settlement currency – on the rails where global value moves.”

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.

Disclaimer: Unless otherwise stated, any opinions expressed below belong solely to the author. Data sourced from the Economic Development Board and Singapore Department of Statistics.

After a turbulent 2025, following the return of Donald Trump to the White House and the wave of trade tariffs the US unleashed on the world, 2026 offered a promise of greater stability. Singapore weathered the storm exceptionally well, bouncing up to a 5% GDP growth, despite fears of a global recession, raising expectations for the new year as well.

Unfortunately, another geopolitical event shook the world on Feb 28, when the US and Israel decided to deal a final blow to the Iranian regime, leading to massive disruptions in oil and gas trade from the Persian Gulf and sending global energy prices soaring.

Since all economies depend on fuel and electricity, virtually every area of our activity is affected. It should be no surprise, then, that the city-state’s business outlook for the coming months has taken a tumble as a result.

Services reeling from fear, manufacturing still supported by AI

Expectations of Singapore-based businesses for the next six months are overwhelmingly negative. Only four industries expect an uptick in activity, with the rest firmly in the red (the figures show the net weighted percentage of positive/ negative answers collected by EDB and Singapore Department of Statistics in their quarterly surveys).

Local manufacturing is, fortunately, still on life support, thanks to the ongoing tech boom in Artificial Intelligence, which is keeping demand for semiconductor products and related sectors high.

Interestingly, the IT rally has been more than enough to offset the suffering of other manufacturing sectors, boosting the overall average:

You can see it reflected in employment forecasts as well, as manufacturing is expected to look for more employees than before. The caveat, of course, is that the demand is highly uneven and driven by a potential bubble which may suddenly burst, leaving thousands of people out of a job overnight.

Employment plans collapse

While this dire prediction about AI frenzy is still a ‘maybe,’ job prospects in services—where a large majority of Singaporeans are employed—are already very grim.

For comparison’s sake, let’s start with what the figures showed in the first quarter of the year:

All but two industries expected increased hiring, having just had a really strong 2025, against the odds and fears.

Now, however, let’s look at what Singapore employers are saying today:

As you can see, the mood has shifted entirely. Only those in Recreation & Personal services expect to bring more people on board (although it’s hard to say how many of these jobs are directed to Singaporeans).

Not even the lucrative Finance industry is being spared this time, and may see job cuts instead of hiring this quarter.

To understand just how big a shift it is, just look at the collapse in sentiments between the two quarters:

With the exception of Real Estate, which is somewhat insulated from the global turmoil (cooling measures for foreign buyers have been in place for a while already, while the local market remains stable and predictable), every other industry has seen a decline in expected hiring activity.

Retail trade appears to be worst hit, having dropped sharply, by 35 percentage points, from a net positive 12 to negative 23.

That said, virtually everybody is less optimistic than they were at the start of 2026. And unless the war with Iran comes to a conclusion soon, companies may be reluctant to increase hiring, given higher operational costs and uncertainty about the consequences of the conflict.

The only consolation is that we already lived through a similar turmoil around the same time last year. All’s well that ends well.

• Read other articles we’ve written on Singapore’s job landscape here.

Featured Image Credit: Shadow_of_light/ depositphotos

Production at Gardenia’s Pandan Loop manufacturing facility will end on Jun 30

Home-grown bread manufacturer Gardenia is shifting its bakery production from Singapore to Johor Bahru, Malaysia. As a result, it will lay off 141 employees at its Pandan Loop facility, Channel News Asia reported on Wednesday (May 20).

The company cited ongoing efforts to improve operational efficiency and remain competitive amid an increasingly challenging global environment as the reason for the move.

Production at the Pandan Loop manufacturing facility will end on Jun 30.

“Gardenia informed employees of the decision at an internal meeting this morning and said affected staff will receive the appropriate notice period and support in line with local regulations and guidelines,” the company said in a media release.

“The company is also considering eligible employees for suitable roles within the group’s network of operations where possible.”

Singapore will remain Gardenia’s central hub for brand management, product development, quality and regulatory oversight, distribution, and supply chain—retaining about 250 employees post-transition.

Its Singapore team will also continue to oversee quality governance and ensure compliance with requirements set by the Singapore Food Agency and the Health Promotion Board.

Gardenia said that the Food, Drinks and Allied Workers Union (FDAWU) was informed in advance about the layoffs.

“This enabled the union to quickly mobilise support such as training, job placement assistance, and discussions on fair retrenchment terms. FDAWU was also able to quickly tap on its network of unionised partners to identify suitable vacancies for affected workers,” it added.

“The union worked closely with Gardenia to ensure fair compensation and transition support for affected employees.”

Nearly five decades ago, Gardenia started as one small bakery in Bukit Timah Plaza in 1978. Today, it operates across Malaysia, the Philippines, and Australia.

Gardenia’s decision comes after similar moves by other food and beverage manufacturers.

In Mar, beverage company Yeo Hiap Seng (Yeo’s) announced it would lay off 25 employees at its Senoko facility as it shifts can manufacturing operations to Malaysia.

Asia Pacific Breweries Singapore, which brews Tiger Beer, also said recently that it would cut about 130 roles as it shifts production to other regional markets such as Malaysia and Vietnam.

• Read more stories we’ve written on the latest job trends here.

Featured Image Credit: National Trades Union Congress

On a balmy, 86-degree day in Mountain View, just outside Google’s sprawling campus, I sought refuge from the glaring California sun in a Volvo EX60 — or so I thought. The air conditioning wasn’t working. 

To mitigate the heat even slightly, we opted to darken the transparent sunroof. And all we had to do was ask Gemini.

“Can you make the sunroof opaque?” Vivek Radhakrishnan, a technical program manager at Google, asked the car via voice command. Like magic, the electrochromic window stretching over us blocked the light beaming in from above. We didn’t have to rummage through the car’s settings to find the right button. 

At Google’s I/O developer conference on Tuesday, I got an early look at a handful of new features coming to cars supporting Google Built-in and Android Auto. The upcoming capabilities, rolling out later this year, are designed to help you keep your eyes on the road while offering useful information. You can lean on Gemini AI to handle tasks like sharing your ETA, describing that landmark that caught your eye and even helping you order dinner.

Watch this: Google’s Car Update Helps You Keep Your Eyes on the Road

02:10

The Volvo EX60 comes equipped with Google Built-in, a native operating system for car infotainment systems that lets you tap directly into Google’s services. We could ask Gemini to identify a dashboard warning light, for example, or have it gauge whether a 65-inch TV we just bought would fit in the back. 

Although we were technically parked outside of Shoreline Amphitheater, a giant TV in front of us simulated driving along a San Francisco road toward one of the city’s signature skyscrapers. We asked Gemini, “What’s that tall building in front of me, and can you tell me something interesting about it?”

Using the car’s front-facing camera, Gemini identified the Transamerica Pyramid, “which stood as the tallest building in San Francisco for 45 years.” We got some bonus information, too, as Gemini said, “Nearby, on your right, you’ll find the historic copper-clad Sentinel Building, a landmark that miraculously survived the 1906 earthquake and later became home to Francis Ford Coppola’s film studio.” It was neat to get that much detail and learn something new. 

Afterward, we hopped over to the Kia EV9 (which thankfully had functioning A/C) to get a look at upcoming Android Auto features, which are available by connecting your phone to the vehicle. They include a more personalized dashboard design built on Google’s Material 3 Expressive, so you can display a picture of your cat alongside custom widgets, for instance. 

While still parked, we then opened up YouTube and watched videos in 4K resolution at 60 frames per second, which can make the time you spend charging your vehicle a little more entertaining. Once you start driving, those videos automatically shift to audio-only so they’re not distracting. 

Magic Cue can pull information from your email, calendar or other apps to quickly respond to texts asking for your ETA or other event details. In our demo, an incoming text asked, “What time is the pottery class?” and Gemini found the date and time automatically: June 1 at 2 p.m. Google Product Manager Alanna Veiga just had to tap to send it as a reply.

If you want to make sure your dinner arrives home when you do, you can also ask Gemini to place an order through a food delivery app. Using voice commands, we asked it to order two fish tacos from Pacific Catch on DoorDash. Gemini then pulled up the DoorDash app on the connected phone and added those items to the cart, tapping through the steps as if it were a human. Once it was time to check out, all Veiga had to do was tap to confirm the order. 

On vehicles with both Google Built-in and Android Auto compatibility, you’ll be able to use the new Immersive Navigation for Google Maps, which shows a 3D view of buildings, overpasses and surrounding terrain, as well as details like lanes, traffic lights and crosswalks. That can help you get a better understanding of your surroundings and make navigation clearer and easier. 

The updates are part of a wider expansion of Gemini into cars, phones, wearables and smart glasses. And while the supercharged AI features might not be able to fix your broken A/C, they can simplify menial tasks and help you stay focused on the road.

Here’s an uncomfortable thought for every academic institution currently using AI detectors to police student and researcher submissions: the tools don’t work as reliably as institutions assume. 

A paper presented at this week’s 2026 IEEE Symposium on Security and Privacy by researchers at the University of Florida concludes that commercially available AI-generated text detectors are “poorly suited for deployment in academic or high-stakes contexts.”

That’s a polite way of saying universities are making career-altering decisions based on results from tools that are essentially unreliable.

What did the research actually find?

Patrick Traynor, Ph.D., professor and interim chair of UF’s Department of Computer & Information Science & Engineering, led a team that tested the five most popular commercially-available AI text detectors. 

Using roughly 6,000 research papers submitted to top-tier security conferences before ChatGPT even arrived, they had LLMs create clones of those same papers, and then ran both sets through the AI detectors. 

The results showed false positive rates ranging from 0.05% to 68.6%, and, even more surprising, false negative rates between 0.3% and 99.6%. That upper figure is close to 100%, meaning the worst-performing detector missed virtually all AI-generated text.

While two of the five detectors performed well initially, they were rendered largely useless after the researchers asked the LLM to rewrite its outputs using more complex vocabulary (the paper calls this a lexical complexity attack).

Why does this matter beyond academic integrity?

Traynor put it plainly: “We really can’t use them to adjudicate these decisions. People’s careers are on the line here.” An accusation of AI-generated writing in a submission can permanently damage a researcher’s reputation, but we can’t put blind trust on tools making those accusations.

The argument is that the evidence about widespread AI use in academic writing is itself unreliable. “For as many studies as we see claiming that a certain percentage of academic work is AI-generated, we actually don’t have tools to measure any of that,” Traynor added. 

His research doesn’t just critique the tools; it exposes a systemic failure of due diligence by every institution that adopted these tools without demanding evidence whether they are accurate.

US lawmakers plan to introduce an amendment Thursday at a House committee markup hearing that would prohibit any recipient of federal highway funding from using automated license plate readers for any purpose other than tolling—a sweeping restriction that, if adopted, would bring an immediate end to state and local ALPR programs across the United States.

The amendment, obtained first by WIRED, is sponsored by Representative Scott Perry, a Pennsylvania Republican and Freedom Caucus member, and Representative Jesús “Chuy” García, an Illinois progressive whose state has become a flash point in the national fight over ALPR misuse.

The House Transportation and Infrastructure Committee will mark up the underlying bill—a $580 billion, five-year reauthorization of federal surface transportation programs—at 10 am ET on Thursday.

Neither Perry nor García’s offices immediately responded to WIRED’s request for comment.

The amendment runs a single sentence: “A recipient of assistance under Title 23, United States Code, may not use automated license plate readers for any purpose other than tolling.”

The amendment is brief, but its reach would be vast. Title 23 funds roughly a quarter of all public road mileage in the US, including most state and county arteries and many city streets where ALPR cameras are becoming ubiquitous. Conditioning that funding on a ban of the technology would, in practical effect, force any state, county, or municipality that takes federal highway money (essentially all of them) to either remove the cameras or restructure their use around tolling alone.

The amendment’s cosponsors, Perry and García, represent opposite ends of the House’s ideological spectrum but converge on a surveillance concern that has gathered momentum in legislatures and city halls across the US as ALPR networks have quietly become a pervasive layer of American road infrastructure.

ALPR cameras—mounted on poles, overpasses, traffic signals, and police cruisers—photograph every passing license plate, log times and locations, and feed data into searchable databases shared across agencies and jurisdictions.

In Illinois, where García’s district sits, Secretary of State Alexi Giannoulias announced last August that an audit by his office had found Flock Group—the Atlanta-based company that operates the country’s largest ALPR network—in violation of state law for giving US Customs and Border Protection access to Illinois ALPR data. Giannoulias ordered the company to cut off federal access.

Flock said at the time that it would pause federal pilots nationwide, arrangements the company had previously denied existed in what Flock CEO and founder Garrett Langley said were public statements that “inadvertently provided inaccurate information.”

Flock did not immediately respond to WIRED’s request for comment.

Privacy advocates have long warned that the aggregation of license plate data amounts to a de facto warrantless tracking system. New York University School of Law’s Brennan Center for Justice has documented the integration of ALPR feeds into police data-fusion systems that combine plate data with surveillance and social media monitoring. And the Electronic Frontier Foundation, a digital rights nonprofit, has documented a range of police misuse, including the past targeting of mosques and the disproportionate deployment of the technology in low-income neighborhoods.

Court records obtained by the EFF and reported by 404 Media last year revealed that a Texas sheriff’s deputy has queried Flock’s nationwide network—roughly 88,000 cameras at the time—to track a woman because, he wrote, she “had an abortion.”

“Flock cameras are easily abused and have already been banned in many municipalities across the nation for their failure to keep our data safe,” says Hajar Hammado, senior policy adviser at Demand Progress, who believes the Perry-García amendment is “commonsense” and says that the country has become a “mass surveillance dystopia”

• The ex-chief of Samsung’s semiconductor business has made a more optimistic prediction about the RAM crisis
• The memory situation will improve thanks to a surge in RAM production from Chinese companies, and some deflation in the AI bubble
• Due to those factors combined, we’re told, “There is a possibility that the market will change starting from the second half of next year or the first half of 2028.”

Could the RAM crisis be over sooner than you thought — and maybe even in not much more than a year? An ex-Samsung exec has stated that this could be a possibility.

Wccftech flagged a report from Seoul Economic Daily (via PC Gamer), which quoted Kye-hyun Kyung, who was head of Samsung’s semiconductor business until a couple of years ago.