Feed aggregator

After a US law temporarily forced Americans to find a new home for their short-video habit, TikTok clone Likee saw a surge in usage.

The World Economic Forum in Davos, Switzerland, is underway this week — and there are calls for taxing the extremely rich to address global inequality.

(Image credit: Fabrice Coffrini/AFP via Getty Images)

• Cloudflare says it blocked a 5.6Tbps DDoS attack in October 2024
• The attack came from a Mirai botnet
• It included 13,000 IPs, and lasted 80 seconds

Cloudflare has claimed it recently blocked the largest Distributed Denial of Service (DDoS) attack ever recorded.

In a blog post, the company said that in late October 2024, its defense mechanisms blocked a 5.6Tbps UDP (User Datagram Protocol) DDoS attack. To put things into perspective, the (now) second-largest DDoS attack ever was 3.8Tbps, also blocked by Cloudflare, also in October 2024.

The company said the attack was launched by a Mirai-variant botnet, and targeted an Internet Service Provider (ISP) from Eastern Asia.

Shorter but more violent

The attack lasted just over a minute (80 seconds), and involved more than 13,000 Internet-of-things (IoT) devices, it was said.

As attackers change their strategies to better adapt to an evolving DDoS threat landscape, the attacks have generally been getting shorter in duration but more intense and frequent.

Despite its destructive potential, the attack did no damage, Cloudflare said, since both detection and mitigation were fully autonomous.

“It required no human intervention, didn’t trigger any alerts, and didn’t cause any performance degradation,” Cloudflare said. “The systems worked as intended.”

The researchers also stressed that while the total number of unique source IP addresses was around 13,000, the average unique source IP addresses per second was 5,500. Each of the 13,000 IP addresses contributed less than 8 Gbps per second, while the average contribution of each IP address per second was around 1 Gbps (~0.012% of 5.6 Tbps).

Mirai is one of the most infamous botnets out there. Its source code leaked in 2017, after which different threat actors started building their own variants. Today, Mirai and its variants often make headlines, targeting different organizations with large-scale DDoS attacks. Just this week, security researchers observed two variants, ‘gayfemboy’, and ‘Murdoc Botnet’.

You might also like

• Standing strong against hyper-volumetric DDoS attacks
• Here's a list of the best antivirus tools on offer
• These are the best endpoint protection tools right now

A new AI-powered search engine called Pearl is launching today, with an unusual pitch: It promises to connect you with an actual human expert if the AI answer sucks. WIRED gave it a spin.

There's a slight catch.

The Nutmeg State has fast and affordable internet available from top providers. These are CNET’s recommendations for the best ISPs in Connecticut.

The Gunners will be aiming to seal their place in the last 16 with a win at the Emirates.

Founded by crypto guru Jed McCaleb, Vast Space will run two missions to the International Space Station, and aims to launch its first space station, Haven-1, by the end of 2025.

• Nothing has posted a Pokémon image on its X account
• This particular character is Arcanine
• New phones are expected soon

Nothing is up to something. The firm is well know for its use of weird teaser images to generate excitement ahead of new product announcements, and the latest one has just dropped.

For the benefit of anyone who isn't familiar with the Pokémon codex, it's a pixelated image of Arcanine. It's surprising to see an image of a Pokémon, given the aggressively protective nature of The Pokémon Company and its partner, Nintendo – so perhaps this indicates an official tie-up between the two companies. The excellent Nothing Ear (a) are already available in a Pikachu-like yellow, so perhaps an Arcanine-like orange is coming.

Or perhaps what's more significant is either the number contained in Arcanine's name, or the number of its place in the Pokédex.

What do the internet detectives think of Nothing's teaser?

Over at GSM Arena, they're suggesting that the nine in Arcanine's name could suggest a product announcement on the 9th of February. Or perhaps its Pokédex number, 58, is the clue. The 58th day of the year would also be in February, albeit a bit later on February 27.

That's the possible whens. As for the what, the smart money appears to be on the Nothing Phone (3) or one of the other two Nothing phones expected to launch by mid-2025 – with or without Pokémon collaboration.

The (3) seems to be the most likely one. A leaked internal memo said that the firm was planning to launch a "landmark" AI-powered smartphone in the first quarter of 2025, which fits with a February announcement – although previously we'd expected a launch at MWC 2025 in March.

That's the thing about teasers. You can make them mean almost anything, but by their very nature they reveal nothing. Or in this case, nothing about Nothing.

You might also like

• Our review of the Nothing Phone 2a Plus
• Leaked Nothing Phone 3 memo teases a flagship phone packed with AI
• The classic Nokia Snake game returns – as a Nothing home screen widget

• MSI's new Claw 8 AI+ has limited availability with only some customers receiving orders
• The release date was set for January 15 in the US, and February 11 in the UK
• It's currently only available at Currys in the UK 

CES 2025 saw a range of upcoming handheld gaming PCs revealed, including Lenovo's Legion Go 2 prototype. While the Nintendo Switch 2's announcement came shortly after the tech event, it's hard to believe that it stands a chance against the new slate of handheld gaming PCs, notably MSI's Claw 8 AI+ – at least, if you can manage to find one in your region.

As reported by VideoCardz, the MSI Claw 8 AI+ is suffering from limited availability at launch, with only a few users in the US getting their hands on the device despite its original 15 January launch date from MSI.

The same issue seems to apply to users in the UK. The release date was set for 11 February, but at time of writing there's only one preorder listing of the Claw 8 AI+ – at Currys, priced at £899. The fact that it's only available on one retailer site in the UK, and only a small number of users in the US have managed to get their hands on it, certainly suggests issues with availability.

The Claw 8 AI+ utilizes Intel's Lunar Lake Ultra 7 258V processor, for improved gaming performance over the previous Claw A1M model, thanks to the Intel Arc 140V integrated GPU. Based on multiple previews including my colleague John Loeffler's hands-on coverage, MSI's latest model looks set to outperform the Asus ROG Ally X and potentially even the currently-in-prototyping Legion Go 2 (powered by AMD's new Z2 Extreme APU).

While we're still awaiting official Nintendo Switch 2 specifications, the rumored internal specs are underwhelming, to say the least: it's supposedly set to use the Cortex-A78AE processor (reportedly using Nvidia's DLSS upscaling tech), and only 8GB of RAM. That doesn't stack up well against the AMD Z1 Extreme, with 16GB of RAM used in the original ROG Ally.

(Image credit: MSI) Will limited availability and price impact the MSI Claw 8 AI+'s success?

While we're only a month into 2025, we'll be seeing handheld gaming PC additions from both MSI and Lenovo with the Claw 8 AI+ and the upcoming Legion Go S, which we also demoed at CES. Considering the pricing of the Claw 8 ($899 / £899 / around AU$1,400), combined the limited availability, it may lose out to other handhelds that are already available, despite appearing to be the stronger performer on paper.

When it comes to handheld gaming PCs, affordability is what matters most. I'm sure the Core Ultra 7 258V will prove to be incredible for the Claw 8 AI+, alongside the hefty 80Whr battery, but its rather steep pricing could end up being a dealbreaker (especially if this limited availability continues).

I've stated it before with handheld PCs – just look at the likes of Acer's Nitro Blaze 11, with pricing of $1,099 (around £1000 / AU$1,740) that could well be its downfall. The same could apply here, even if it's slightly cheaper than Acer’s monster handheld. At those prices, if you don't already own a desktop gaming PC, it makes little sense to opt for a handheld. However, I'm hoping its Claw 7 AI+ counterpart could be a compromise, using the same processor but with a 54.5WH battery and 7-inch screen instead, at $799 / around £700 / around AU$1,245.

You may also like...

• The Lenovo Legion Go S is the true Steam Deck alternative we've all been waiting for - but should you buy it or wait for a better handheld?
• I'm patiently waiting for Apple to turn the Mac mini into a games console - could this Nvidia RTX 5080-powered Asus ROG NUC mini PC be the answer?
• Windows 24H2 is a nightmare for handheld PC gamers right now, and that's why Bazzite is an absolute must for the Asus ROG Ally

• The new Polestar 7 SUV will launch in late 2026 or early 2027
• It won’t replace the Polestar 2, but will be an affordable entry to the brand
• Compact electric SUV will compete with Tesla, Audi and Volvo

Polestar has revealed that it will be adding a new member to its gang of stylish-yet-understated electric vehicles with the upcoming Polestar 7 SUV, which the company says will come before the much-hyped Polestar 6 sports car.

Although the Swedish manufacturer has been a single-model brand for a long time (offering just the Polestar 2 since 2019, which is also due a refresh), it has recently released the larger Polestar 3 and Polestar 4 crossover vehicles, while the upcoming Polestar 5, which will take on the Porsche Taycan and Audi e-tron GT, is slated for release later this year.

The recently-appointed CEO, Michael Lohscheller, has been talking to various media outlets about his plans to take the brand in a new direction, with the Polestar 7 embodying what he deems to be the perfect car for entering a hugely popular and contested market – one that many premium manufacturers have traditionally avoided.

Although Polestar 7 details are fairly scant, Lohscheller has promised that it will ride on an all-new platform (although he won’t say which of the Geely-owned companies will provide it) and boast “the latest technology”, which can be construed as 800V architecture and 350kW fast charging capabilities.

The new CEO also said that it would sit in-between the Volvo EX30 and EX40, which gives a good indication of size, while pricing will likely also sit between those two models, so between £40,000 and £45,000 in the UK. That will work out as $45,000 and $55,000 for the US market (or around AU$72,000-AU$87,000).

This is prime Tesla Model Y territory, which itself has just receive a long-awaited ‘Juniper’ refresh, but CNBC reports that the brand has recently shed $15 billion in value due, in part, to an aging line-up of vehicles.

During an interview with Car Magazine, the Polestar CEO revealed that the he wants his design team to be “more confident” in the next generation of vehicles. With the Polestar 5 and Polestar 6 already way beyond concept stage, it is up to the Polestar 7 to embody these changes.

“Polestar is sometimes a bit too modest, a bit too humble. In several markets, like in Scandinavia, this works but I would like a little bit more confidence. In the US the Americans don’t like too much understatement,’’ he said.

Analysis: Trump's America will be a tough nut to crack

Polestar CEO Michael Lohscheller (Image credit: Polestar)

The Polestar chief believes that 2025 will be the strongest year in Polestar history, but the company faces numerous question marks over its ability to crack the US market.

Currently, the Polestar 2 and recently released Polestar 3 are available to buy in the US, with the latter built in the company’s Georgia plant. But the outgoing Biden administration has set the wheels in motion for an outright ban on any vehicle that features connected tech hailing from China.

This will adversely affect the Swedish marque, seeing as it is owned by Chinese company Geely and borrows much of its technology from the group. But Polestar’s new CEO says it will seek new suppliers to get around the ban, according to The Financial Times.

To compound matters, Trump’s recent signing of executive orders that could end numerous electric vehicle subsidies will no doubt force Polestar prices in the US up, leaving Tesla to rely on its vast EV economies of scale to offer the cheaper product.

You might also like

• I drove the new screen-obsessed Polestar 4 – and its lack of a rear windscreen isn't the only thing it should be remembered for
• Five ways Trump's new policies will impact electric vehicles in the US
• Tesla reveals its long-awaited Model Y 'Juniper' refresh, but the EV isn’t the trailblazer it once was

A China-linked cyberespionage group has reportedly exploited a legitimate VPN service to spread malware and spy on victims' activities. The ESET security research team found the malicious code – alongside the legitimate software – in the Windows installer of IPany, a South Korean VPN provider.

The so-called PlushDaemon APT group is also known to have hijacked legitimate updates of Chinese applications, but this technical-advanced supply-chain attack against a trustworthy Korean VPN firm makes the hacking group "a significant threat to watch for," said ESET experts.

The SlowStepper backdoor

ESET's new report shed light on a previously undisclosed China-aligned APT group so-called PlushDaemon which experts believe to have been active since at least 2019 – and one of its malicious operations aims to spy on the target's activities.

To do so, hackers have hijacked legitimated updates of Chinese apps and launched a supply-chain attack against South Korean VPN developer IPany. Both involve injecting a malicious backdoor into the device while the victims install the software.

Named SlowStepper, the backdoor is built on an advanced infrastructure that enables extensive data collection and spying through the recording of audio and videos.

"We found no suspicious code on the download page to produce targeted downloads, for example by geofencing to specific targeted regions or IP ranges," experts explain. "Therefore, we believe that anyone using the IPany VPN might have been a valid target."

You can read the full technical analyses in the ESET blog post here.

When the malicious IPanyVPNsetup.exe installer is executed, it creates several directories and deploys both legitimate and malicious files. (Image credit: ESET )

Experts contacted the VPN software developer to inform them of the compromise. The company then removed the malicious installer from its website.

Nonetheless, ESET findings raise concerns for internet users' security, especially considering that the hacking group managed to fly under the radar for so long.

Experts wrote: "The numerous components in the PlushDaemon toolset, and its rich version history, show that, while previously unknown, this China-aligned APT group has been operating diligently to develop a wide array of tools, making it a significant threat to watch for."

Worse still, this is far from the only instance in which VPN users – so, someone actively looking to protect their online data – are the main target. Google reported a similar threat at the beginning of January 2025 warning against how Playfulghost attackers used VPN apps to infect devices with malware.

I recommend being extra careful when downloading new software from the web. If you notice your device acting oddly, you should run a malware removal service, whenever possible, and consider a system reboot to eradicate the potential threat.

• Security researchers warned about a vulnerability in older versions of 7-Zip
• The vulnerability allowed threat actors to bypass the Mark of the Web security feature
• The bug was fixed in late November 2024

A high-severity vulnerability was recently discovered, and patched, in the popular open source file archiver solution 7-Zip. Since the product does not have an automatic update feature, users are advised to upgrade to the newest version manually, as soon as possible.

The vulnerability in question is tracked as CVE-2025-0411. It is described as a Mark of the Web (MotW) bypass, that allows threat actors to execute malicious code on target endpoints that are extracting files from nested archives. It was given a severity score of 7/10 - high.

Mark of the Web is a security feature in Windows that flags files downloaded from the internet as potentially unsafe by adding metadata indicating their origin. This helps prevent malicious scripts or executables from running automatically, prompting users to confirm before opening such files.

Patching the flaw

7-Zip added support for MotW in June 2022, in version 22.00. However, the feature was improperly implemented, and could be bypassed. In a recently released advisory, cybersecurity researchers Trend Micro explain:

"This vulnerability allows remote attackers to bypass the Mark-of-the-Web protection mechanism on affected installations of 7-Zip. User interaction is required to exploit this vulnerability in that the target must visit a malicious page or open a malicious file," the researchers said.

"The specific flaw exists within the handling of archived files. When extracting files from a crafted archive that bears the Mark-of-the-Web, 7-Zip does not propagate the Mark-of-the-Web to the extracted files. An attacker can leverage this vulnerability to execute arbitrary code in the context of the current user."

The bug has since been mitigated, with a version 24.09 being released in late November 2024.

"7-Zip File Manager didn't propagate Zone.Identifier stream for extracted files from nested archives (if there is open archive inside another open archive)," the project’s developer, Igor Pavlov, explained.

Via BleepingComputer

You might also like

• This new malware is able to bypass all of Microsoft's security warnings
• Here's a list of the best antivirus tools on offer
• These are the best endpoint protection tools right now

• AMD has been telling us about the gaming performance of the Ryzen 9950X3D
• We’re told to expect frame rates that are similar to the 9800X3D
• This is despite some nice spec bumps for the as-yet-unreleased flagship for 3D V-Cache

AMD’s flagship 3D V-Cache processor, the Ryzen 9 9950X3D, will offer a ‘comparable’ performance for gaming compared to the existing mainstream Ryzen 7 9800X3D CPU.

In an interview with VideoGamer, AMD’s Product and Business Development Manager, Martijn Boonstra, said that Team Red expects the Ryzen 9950X3D and its lesser sibling, the 9900X3D, to deliver a “similar overall gaming performance to the 9800X3D.”

In short, there will be some give or take here, with the Ryzen 9 chips being slightly faster in certain scenarios, and the Ryzen 7 9800X3D edging a victory in others.

Boonstra clarified that with those incoming Ryzen 9 CPUs: “There will be some games that perform a bit better (if the game engine utilizes more cores and threads), and some games will perform a little worse (if the game engine favors a 1CCD configuration), but on the whole the experience is comparable.”

The Ryzen 9950X3D and 9900X3D were unveiled at CES 2025 earlier in January, and are expected to be available soon – but pricing is yet to be confirmed. They will, inevitably, be considerably more expensive than the Ryzen 7 9800X3D. If you recall, the original MSRP for the 7950X3D was 55% more than the 7800X3D (in the US).

(Image credit: AMD) Analysis: An easy choice for gamers, it seems

So, you might be thinking – you only get the same performance from a pricier CPU – what gives with that? Well, remember, this is gaming frame rates we are talking about specifically.

The Ryzen 9 9950X3D and 9900X3D are designed as dual-use processors, which are excellent for creatives (and can easily cut through other heavyweight tasks), as well as providing great gaming performance. Although given the 3D V-Cache flagship’s spec this time around, some folks might have expected better performance in PC games – though AMD is clearly priming us not to, and this appears to be what this bit of marketing info is all about.

Regarding that spec, the Ryzen 9 9950X3D not only provides 16-cores (and 32-threads), versus the 8-cores (16-threads) of the 9800X3D, but it also has a 5.7GHz boost clock, plus more of that crucial 3D V-Cache (144MB versus 104MB).

However, it seems despite this extra pep, the situation will remain the same as it was with the previous-gen X3D chips, where the 7950X3D and 7800X3D were pretty much the same in terms of gaming performance in the main, give or take a bit. At least, after any CPU configuration gremlins were ironed out, because it’s true that in the past, the 7950X3D could be held back in some situations because it has two CCDs. (This means two separate Core Complex Dies, or chiplets, only one of which has 3D V-Cache – and sometimes a thread for a running game might be shoved over to the second chiplet, causing some latency. This didn’t happen with the 7800X3D as it only has one CCD).

You don’t really need to get into the technical weeds here, though – all you really need to know is what AMD is serving up as a headline here. If you’re gaming, the Ryzen 9800X3D is the CPU for you, purely because it’ll be much more affordable, and gives you about the same performance with PC games as the Ryzen 9950X3D.

However, if you need the best all-round performance, for heavyweight apps as well as gaming, then the Ryzen 9 X3D models are for you. Yes, you’ll pay the price for them no doubt, but it’s likely those initially towering MSRPs will come down considerably as time passes (as happened with the Ryzen 7950X3D).

The other point to consider is right now, the Ryzen 9800X3D remains difficult to find (at or near its MSRP), as it’s very popular. So some PC gamers might look at the new Ryzen 9 3D V-Cache processors when they arrive just so they can pull the trigger on an X3D purchase for their new gaming PC (or upgrade). Depending on where pricing shakes out – as noted, the last-gen flagship’s 55% premium over the Ryzen 7 model represented a pretty steep ask.

Via Tom’s Hardware

You might also like

• These are the best AMD processors
• These are all the best cheap graphics card deals
• What is a processor: Your CPU explained in plain terms

Looking for an ultralight ultraportable? These are our favorite lightweight laptops, tested and reviewed by CNET's computer experts.

Samsung is launching an all-new Galaxy Watch for Kids 'experience' on the eve of today's (Jan 22) Samsung Galaxy Unpacked presentation – a few hours before its big keynote kicks off.

Rather than making a dedicated version of the Galaxy Watch, like the Fitbit Ace LTE, Samsung has created the Galaxy Watch for Kids experience, a new mode you can select on the Samsung Galaxy Watch 7 LTE – one of the best Samsung watches. This mode prevents the child from accessing certain features, allows parents to stay connected, offers safety features like SOS messages, and provides 'self-expression and learning experiences.'

These experiences are launching with over 20 'teacher-approved' apps, including some from a variety of recognizable IPs, such as Barbie, Crayola, PBS Kids, and Unicorn Academy.

However, it's Marvel that takes front and center in the marketing. The Marvel HQ app offers watch faces inspired by characters like Captain America and Groot but kicks it into high gear by allowing kids to mix tunes on the app's DJ interface and dance in real life to move with Groot.

(Image credit: Samsung/Google)

By pairing the watch with a parent's smartphone, the parent can select "set up for a child" on the Watch 7 LTE, using Google Family Link to create multiple accounts. After activating the eSim on the watch, a phone is no longer required for the child to use the watch and access the core functionality.

Safety tools for parents include location sharing, managing their child's contacts, a 'Do-Not-Disturb-like' mode, and SOS Messages, which can be sent by pressing the watch's side button five times.

Kids bands will be available to purchase for the watch, which acts as a protective silicone covering, although users buying a Galaxy Watch 7 LTE from Samsung.com will receive a free kids band.

Like the Fitbit Ace LTE, Galaxy Watch for Kids is only available in the US at present.

This mode is similar to Apple's 'Apple Watch For Your Kids' experience, which also offers emergency SOS messages, child-orientated movement tracking, kid-friendly watch faces and educational apps, alongside Apple Cash, a way to send money through Apple devices to act like an allowance used with Apple Pay. Like the Samsung experience, you need an LTE watch to use the 'Apple Watch for Kids' feature.

You might also like...

• Samsung Galaxy Unpacked live: the countdown to the S25
• Your Samsung Galaxy Ring might be the key to a smarter home thanks to SmartThings' connected device roadmap
• Samsung Galaxy S25 shown off hours before launch

Join CNET as we watch Samsung's next big reveal, starting at 9:30 a.m. PT (12:30 p.m. ET).

Never before have business leaders had so little time to make decisions. Between consumers expecting real-time updates on every purchase or service they pay for, to the competitors looking to beat them to every customer, slow and careful deliberation simply isn’t a luxury that many can afford.

Unfortunately, leaders are also judged — even defined — by the ability to make the right call under pressure. Napoleon Bonaparte once said that "nothing is more difficult, and therefore more precious, than to be able to decide."

But if you don’t have the time to think, how can you make good decisions? Without the full context to ground your judgement, or the time to process them, what are the alternatives?

The risky business of gut-feel

For many, the answer is to go with your gut.

New research — exploring how 200 CEOs, MDs, and C-level executives in the UK approach decision-making — finds that this is true for more than three in five (61%) in the UK. The majority make ‘snap decisions’.

This is because, for many, the pressure to make a fast decision — any decision — overcomes the need to make the right decision. Leaders are choosing to substitute the full picture for a subjective interpretation, based on their experience and opinions alone.

It also indicates a lack of technological maturity within a business. More than half (58%) say they rely on ‘gut feel’ because they’re unable to access insights in real time, while 52% say that data is out of date by the time it reaches them.

In other words, business-critical systems simply can’t provide the data business leaders need in time to factor it into the decision-making process.

In some cases, instinct and experience can be enough. But these qualities are, by definition, retrospective — they can’t accurately factor in new information, or unexpected changes. And in a business environment that evolves faster than ever before, there’s always something new to consider.

Ultimately, gut feel is only part of the picture. The other part is evidence; beliefs should not replace analytics but refer to them for context.

Demand for data-led decisions

Decision-makers make that sentiment clear. While gut feel is prominent, 84% said they’d like to make more decisions based on accurate data, while a similar 85% believe they’d make better decisions if they were able to base them on real-time data.

Business leaders need data and insights available at their fingertips. Rather than compromise and slow down the decision-making process, they need real-time data to inform and complement their gut feel — and they’re often not equipped to access this.

We can see this with 80% of leaders stating that they need to “completely overhaul” their existing data approach. There’s a desire and intent to make things better to meet the demands of fast-paced businesses.

In many organizations, data can’t move through their systems in sufficient time to deliver real-time insights. Batch processing data is a common obstacle here, with many businesses designed to process data in huge, inefficient cuts after it’s already been collected. The siloing of that data — between different apps, departments, and cloud providers — can make things even harder.

Swim with the stream

Consequently, many business leaders are looking to data streaming platforms. Rather than processing data in cumbersome blocks, hours or even days after it’s collected, they’re prioritizing systems that can make data accessible in real time — no matter where in the business it is — to allow analytics to match that pace.

The numbers here are striking. More than three quarters of business leaders (77%) are investing in data streaming to improve on their decision-making processes. And a huge 97% are investing, or have plans to invest in, real-time dashboards to make real-time analytics readily available.

It’s no surprise, with the early adopters who are already walking the walk seeing considerable benefits. According to recent data, 84% of IT leaders who had invested in data streaming were seeing 2x to 10x returns on their investment, with 41% citing an ROI of 5x or more.

Access to real-time data that can ground your business decisions in reality is already changing how businesses operate, for the better.

That’s because data isn’t just an IT problem. Data can speak to the forces that impact every part of the modern business — from financial performance to customer experience. The more accurate and in-the-moment that data is, the more value it can bring.

The same is true of our leaders. For all the value and wisdom in their instincts and experiences, real-time data streaming is becoming a critical part of business leadership. Data-powered decisions will come to define great leaders.

We've featured the best online collaboration tool.

This article was produced as part of TechRadarPro's Expert Insights channel where we feature the best and brightest minds in the technology industry today. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/news/submit-your-story-to-techradar-pro

These are the most reliable filaments for your next 3D printing project, from basic PLA to specialty filaments.

The momentum of AI-driven applications is accelerating around the world and shows little sign of slowing. According to data from IBM, 42% of companies with more than 1000 employees are actively using AI in their business, and a further 40% are trialing and experimenting with it.

As AI adoption gains pace, with platforms such as OpenAI’s GPT-4o and Google’s Gemini setting new benchmarks of performance, organizations are discovering new applications for these technologies that can deliver better outcomes. Faced with new challenges of deploying the technology at scale. More and more enterprise workflows are embedding calls to these AI Models are dramatically increasing usage. Do the use cases justify the escalating spending on the latest models?

Embracing AI also means embracing the usage of AI models and paying for AI inference costs, at a time when many organisations are in cost-cutting mode. With continued economic uncertainty, rising operational costs, and increasing pressure from stakeholders to deliver ROI, businesses are looking for ways to optimize their budgets and reduce unnecessary expenditures. The escalating costs of AI infrastructure can be a cause of tension, as organizations want to remain competitive and leverage the power of AI, while also finding the balance between these investments and financial prudence.

To further complicate things, AI agents which, according to McKinsey are the next frontier of GenAI and are largely expected to form the next wave of applications, will dramatically increase usage of AI models as they rely on them for ongoing reflection and planning steps. Instead of singular API calls to underlying models like those from OpenAI, agentic architectures can make scores of calls, thus racking up those costs. How can businesses navigate the rising costs of data while powering the AI applications they need?

Understanding the cost of AI at scale

The rapid deployment of AI is leading to increased costs on multiple fronts. First and foremost, organizations are spending on AI inference which is the process of using a trained model to make predictions or decisions based on provided inputs. Often, they would rely on APIs from the leading providers such as OpenAI, Anthropic or cloud service providers like AWS or Google and would pay based on usage. Alternatively, some organizations run their own inference and buy or rent GPUs on which they deploy open source models such as Llama from Meta.

Secondly, in many cases organizations want to customize their AI models by ‘fine-tuning’ them. This can at times be an expensive process involving data preparation by creation of training datasets and require compute resources for training.

Finally, building AI applications requires additional components, such as vector databases, which help augment inference by helping retrieve relevant content from designated knowledge bases and thus improve accuracy and relevance of responses from AI models.

By examining the root causes and the drivers of their AI costs such as AI inference, training or fine-tuning and additional components such as databases, businesses can minimize storage costs and enhance their AI application performance.

Optimizing efficiency through semantic caching

Semantic caching is a highly effective technique that organizations are deploying to manage the cost of AI inference and to increase speed and responsiveness of their applications. It refers to storing and reusing the results of previous computations based on their semantic meaning.

In other words, instead of relying on new AI computations for new queries, a semantic cache can check a database for queries with similar meanings that have been asked before, thus saving costs. This approach helps reduce redundant computations and improves efficiency in applications like inference or search.

In one particular study, researchers showed that up to 31% of queries to AI applications can be repetitive. Every unnecessary AI inference call adds avoidable costs, but by implementing a semantic cache, organizations can significantly reduce these calls, cutting them by 30-80%. This method is crucial for building scalable and responsive generative AI applications or chatbots. This approach not only optimizes cost but also accelerates response times, helping businesses achieve more with less investment.

Balancing performance and cost

Organisations need to optimise their technology stack and operational strategies to be able to deploy cutting-edge AI applications without incurring unsustainable infrastructure costs. This can help them strike that crucial balance between performance and cost. Techniques such as semantic caching can play a vital role in this.

For companies that are struggling with scaling AI applications in an efficient and cost-effective way, learning how to effectively manage this would become a key market differentiator. The key to businesses navigating the spiraling cost of generative AI applications and maximizing their value could lie in the AI Inference strategy. As generative AI systems get more and more complex, every LLM call needs to be as efficient as possible. By doing so, customers can get to the information they need faster and businesses can minimize their cost footprint.

We've featured the best AI website builder.

This article was produced as part of TechRadarPro's Expert Insights channel where we feature the best and brightest minds in the technology industry today. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/news/submit-your-story-to-techradar-pro