NVIDIA has introduced the Rubin CPX, a new type of GPU designed for long-context AI processing. The chip is built to handle workloads that require models to process millions of tokens at once – whether that’s generating code in entire software projects or working with video content an hour in length.

Rubin CPX works alongside NVIDIA’s Vera CPUs and Rubin GPUs inside the Vera Rubin NVL144 CPX platform. A single rack delivers 8 exaflops of AI compute, with 7.5 times the performance of NVIDIA’s GB300 NVL72 system, along with 100 terabytes of fast memory and bandwidth of 1.7 petabytes per second. Customers can also adopt Rubin CPX in other system configurations, including options that reuse existing infrastructure.

Jensen Huang, NVIDIA’s founder and CEO, described the launch as a turning point: “The Vera Rubin platform will mark another leap in the frontier of AI computing – introducing both the next-generation Rubin GPU and a new category of processors called CPX. Just as RTX revolutionised graphics and physical AI, Rubin CPX is the first CUDA GPU purpose-built for massive-context AI, where models reason in millions of tokens of knowledge at once.”

From coding to video generation

The demand for long-context processing is growing fast. Today’s AI coding assistants are limited to smaller code blocks, but Rubin CPX is designed to manage far larger projects, with the ability to scan and optimise entire repositories. In video, where an hour of content can take up to one million tokens, Rubin CPX combines video encoding, decoding, and inference processing in a single chip, making tasks like video search and generative editing more practical.

The GPU uses a monolithic die design based on the Rubin architecture, packed with NVFP4 compute resources for high efficiency. It delivers up to 30 petaflops of compute with NVFP4 precision and comes with 128GB of GDDR7 memory. Compared to NVIDIA’s GB300 NVL72 system, it provides triple the speed for attention mechanisms, helping models handle longer sequences without slowing down.

Rubin CPX can be deployed with NVIDIA’s InfiniBand fabric or Spectrum-X Ethernet networking for scale-out computing. In the flagship NVL144 CPX system, NVIDIA says customers could generate $5 billion in token revenue for every $100 million invested.

Beyond Rubin CPX itself, NVIDIA also highlighted results from the latest MLPerf Inference benchmarks. Blackwell Ultra set records on new reasoning benchmarks like DeepSeek R1 and Llama 3.1 405B, with NVIDIA being the only platform to submit results for the most demanding interactive scenarios.

The company also set records on the new Llama 3.8B tests, Whisper speech-to-text, and graph neural networks. NVIDIA credited these wins partly to a technique called disaggregated serving, which separates the compute-heavy context phase of inference from the bandwidth-heavy generation phase. By optimising them independently, throughput per GPU rose by nearly 50%.

NVIDIA tied these results directly to AI economics. For example, a free GPU with a quarter of Blackwell’s performance could generate about $8 million in token revenue over three years, while a $3 million investment in GV200 infrastructure could generate around $30 million. The company framed performance as the key lever for AI factory ROI.

Rubin CPX is positioned as the first of a new class of “context GPUs” in NVIDIA’s roadmap. The company is also advancing the Rubin Ultra GPU, Vera CPUs, NVLink Switch, Spectrum-X Ethernet, and CX9 SuperNICs – all designed to work together in a one-year upgrade cycle. The roadmap emphasises NVIDIA’s push for full-stack solutions, not just standalone chips.

Industry adoption

Several AI companies are preparing to use Rubin CPX. The company behind Cursor, the AI-powered code editor, expects the GPU to support faster code generation and developer collaboration. “With NVIDIA Rubin CPX, Cursor will be able to deliver lightning-fast code generation and developer insights, transforming software creation,” said Michael Truell, CEO of Cursor.

Runway, a generative AI company focused on video, sees Rubin CPX as central to its work on creative workflows. “Video generation is rapidly advancing toward longer context and more flexible, agent-driven creative workflows,” said Cristóbal Valenzuela, CEO of Runway. “We see Rubin CPX as a major leap in performance, supporting these demanding workloads to build more general, intelligent creative tools.”

Magic, an AI company building foundation models for software agents, highlighted the ability to process larger contexts. “With a 100-million-token context window, our models can see a codebase, years of interaction history, documentation and libraries in context without fine-tuning,” said Eric Steinberger, CEO of Magic.

Supported by NVIDIA’s AI stack

Rubin CPX is integrated into NVIDIA’s broader AI ecosystem, including its software stack and developer tools. The GPU will support the company’s Nemotron multimodal models, delivered through NVIDIA AI Enterprise. It also connects with the Dynamo platform, which helps improve inference efficiency and reduce costs for production AI.

Backed by CUDA-X libraries and NVIDIA’s community of millions of developers, Rubin CPX extends the company’s push into large-scale AI, offering new tools for industries where context size and performance define what’s possible.

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information.

AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.

The post NVIDIA introduces Rubin CPX GPU for long-context AI appeared first on TechWire Asia.

China’s gaming and animation industries are no longer just local success stories. With titles like Black Myth: Wukong and Ne Zha 2 drawing attention from global audiences, the country has shown its ability to produce content that travels well beyond its borders. At the same time, rapid adoption of artificial intelligence is reshaping how stories are made, blending machine efficiency with human creativity.

Tech Wire Asia spoke with Cheng Lu, President and CEO of CreateAI, about how Chinese studios are redefining entertainment, the role of AI in production, and what the future holds for immersive storytelling.

China’s growing influence in global gaming and animation

Asked about China’s role in shaping global entertainment over the next decade, Cheng points to the momentum already visible in the market. “China’s gaming and animation industries are gaining global dominance. Blockbusters like Black Myth: Wukong and Ne Zha 2 are showcasing our ability to captivate audiences worldwide,” he said.

That influence is backed by strong numbers. According to the “2025 H1 China Game Industry Report” from the Game Publishing Committee, the industry generated RMB 168 billion (USD 23 billion) in sales revenue in the first half of 2025 – a 14% year-on-year increase. Chinese self-developed games earned USD 9.5 billion overseas, underscoring their reach in a global market worth around USD 250 billion.

Cheng sees the next decade as one where Chinese studios can lead in immersive, AI-enhanced storytelling. “Fueled by rapid AI adoption and investment in creative production, Chinese studios are redefining entertainment with innovative technology and cultural storytelling,” he explains. “Over the next decade, China is likely to lead in immersive, AI-enhanced storytelling and cross-platform experiences, shaping global trends by blending technology with cultural heritage.”

Balancing AI efficiency and human artistry

AI has become more visible in creative production pipelines, raising the question of how to balance machine efficiency with human artistry. Cheng notes that animation, especially 2D, has long been labour-intensive. Scriptwriting, keyframing, and colouring often slow production, particularly when there’s a shortage of skilled animators.

Here, AI can make a difference without displacing artists. “AI can help studios produce higher-quality content more efficiently without sidelining human creators,” Cheng says. Tools like Animon.ai can turn simple images into anime videos, reducing repetitive tasks and giving artists more time to focus on storytelling.

The idea of synergy – AI handling technical tasks while humans guide the creative heart – runs through Cheng’s perspective. He points to the company’s recent release for example: “Our Animon.ai Studio Version, launched in July, exemplifies this approach by providing creators with tools like high-quality 2K visual generation and consistent keyframe editing, enabling both professionals and beginners to streamline workflows while retaining full creative control.”

Making local stories travel

Chinese titles often draw from local myths and traditions but still manage to resonate with audiences around the world. Cheng sees lessons here for other markets.

He highlights CreateAI’s game Heroes of Jin Yong, which reflects the role of chivalry in Chinese wuxia culture. “Chivalry is something that resonates in the world, but manifests uniquely in China with wuxia culture,” he explains. The broader lesson, he says, is to take cultural dynamics rooted in one place and show how they reflect universal human experience.

“All markets can consider cultural dynamics inherent to themselves that are shared in the human experience, and show how those come to fruition in their unique culture,” Cheng says. Done well, this approach lets people outside the culture connect to familiar values while sparking curiosity about a new one.

The future of immersive entertainment

Advances in motion capture, real-time rendering, and AI animation are already changing the entertainment industry. Cheng expects those shifts to accelerate. He outlines two big trends:

1. The gamification of everything. “Top content is becoming more immersive and interactive,” Cheng says.
2. Cross-media synergies. “Blending between what is a video game and TV show, and video game IP are being made into anime shows, and vice versa.”

At CreateAI, the team is exploring both directions. With the Three-Body Problem franchise, they are working on an anime feature film and a AAA video game, based on the second book of the series. The goal is to launch them side by side. “The generates maximum consumer exposure and greatly enhances fan experience,” Cheng says.

As AI gains the ability to generate characters, voices, and entire worlds, questions about authenticity and ethics are unavoidable. Cheng is clear on this point: “We believe in ‘safe AI’ and will do our part to promote the generative AI industry to grow according to high ethical values and local regulations.”

Asked to compare how AI and gaming innovation differ in regions, Cheng avoids making sweeping claims. “AI and gaming are truly global industries facing global competition,” he says. Still, he notes that companies succeed by excelling in three areas: creating compelling intellectual property, applying new technology for efficiency or storytelling, and using effective distribution channels.

Skills the next generation of China’s gaming creators will need

Looking ahead, Cheng sees a need for a new mix of skills among animators, developers, and storytellers. “The next generation of animators, game developers, and storytellers will need to blend technical proficiency with creative adaptability to thrive in an AI-driven industry,” he says.

A key mindset is to see AI as a tool for expansion rather than replacement. “Viewing AI as an opportunity to create more content, rather than a replacement, is key,” Cheng stresses. He also underlines the importance of cultural sensitivity and narrative innovation – qualities that make stories resonate beyond their home market.

Opportunities in China’s gaming and animation with AI

When asked what excites him most about the future, Cheng points to AI’s potential to open creation to more people. “At CreateAI, what excites us most is AI’s potential to democratise creation and deliver deeply personalised, immersive experiences,” he says.

“Our tools allow any fan to become a creator, freeing new opportunities in the creative economy, while adaptive narratives create immersive experiences,” Cheng explains. He sees breakthroughs coming from cross-platform projects, using global IPs like Heroes of Jin Yong and The Three-Body Problem to build both anime and AAA games.

That approach, he says, creates communities of fans and creators who help redefine how stories are consumed. “We have a full pipeline of projects currently, but we are always opportunistic to work with strong IP holders, using our technology and development know-how to innovate and bring immersive content to a global audience.”

Closing thoughts

From China’s expanding influence in the global gaming market to the role of AI in reshaping production, Cheng Lu’s perspective underscores how technology and cultural storytelling are increasingly intertwined. The future of entertainment, he suggests, won’t be about choosing between AI or human creativity but about finding new ways for them to work together.

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information.

AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.

The post CreateAI’s CEO on how AI is changing China’s animation and gaming appeared first on TechWire Asia.

Microsoft has introduced its first in-house AI models, a move that could reshape its position in the AI race. The company rolled out MAI-Voice-1, a speech model, and MAI-1-preview, a text-based model it calls a glimpse of what’s coming next inside Copilot.

The MAI-Voice-1 model is built for speed. According to Microsoft, it can generate a full minute of audio in less than a second using just a single GPU. The model is already in use inside some of the company’s tools. For example, Copilot Daily uses it to deliver short news summaries through an AI voice host. It also helps produce podcast-style conversations that break down complex topics into easier explanations.

The second release, MAI-1-preview, is designed for text tasks. Microsoft trained the model on roughly 15,000 Nvidia H100 GPUs, giving it the scale to handle instruction-following and natural Q&A. Users can already try it on Copilot Labs, where they can test its ability to respond to everyday queries. Microsoft says the model will soon support text-based use cases inside its Copilot assistant.

Competition with OpenAI

These launches come while Microsoft is still heavily tied to OpenAI, the maker of ChatGPT. Microsoft has invested more than $13 billion into the startup, which now has a valuation of about $500 billion. OpenAI continues to rely on Microsoft’s cloud infrastructure to run its own models, while Microsoft uses OpenAI’s systems inside Bing, Windows, and other products.

At the same time, the two companies are drifting into competition. Last year, Microsoft added OpenAI to the list of rivals it names in its annual report, alongside Amazon, Apple, Google, and Meta. OpenAI has also been spreading its infrastructure needs across other providers such as CoreWeave, Google, and Oracle, as demand for ChatGPT climbs. The chatbot now draws about 700 million weekly users.

Early results and rankings

Performance comparisons show Microsoft’s work still trails some of its peers. On Thursday, the new MAI-1-preview ranked 13th for text workloads on LMArena, behind models from Anthropic, DeepSeek, Google, Mistral, OpenAI, and Elon Musk’s xAI. While not at the top, Microsoft has positioned MAI-1-preview as its first foundation model built entirely in-house.

“MAI-1-preview represents our first foundation model trained end to end in house,” Mustafa Suleyman, Microsoft AI chief, wrote on X.

A consumer focus

Suleyman has been clear about the group’s direction. In an interview last year, he explained that Microsoft’s AI models are aimed at consumer use rather than the enterprise market. “My logic is that we have to create something that works extremely well for the consumer and really optimise for our use case,” he said. He pointed to Microsoft’s access to large amounts of consumer data—such as ad performance and telemetry—as a strength in training models for everyday companions.

The company has also said that it does not plan to rely on one general-purpose model. Instead, it sees potential in offering multiple specialised models designed for different types of requests. “We believe that orchestrating a range of specialised models serving different user intents and use cases will unlock immense value,” Microsoft AI wrote in a blog post.

Building an AI division

MAI-1-preview builds on earlier small-scale models released under the Phi name. But this marks the first time Microsoft has trained a foundation model of this size from start to finish. The effort reflects how the company has been building out its AI group since hiring Suleyman and many of his former colleagues from the startup Inflection.

Suleyman previously co-founded DeepMind, the research lab Google bought in 2014. In the past year, Microsoft has brought on about two dozen former DeepMind researchers to expand its internal team. The hires show how the company is drawing on talent with long experience in AI development to accelerate its own projects.

For now, Microsoft is positioning its new models as additions to its Copilot ecosystem while it continues to rely on OpenAI for many core features. But the release of MAI-Voice-1 and MAI-1-preview signals a step toward more independence in model development. Analysts say it could also set up a new phase of competition between Microsoft and the company it helped make into an AI giant.

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information.

AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.

The post Microsoft debuts its first in-house AI models appeared first on TechWire Asia.

Instagram’s new location-sharing feature is sparking alarm among some users, who say it could put people at risk by revealing where they are without their knowledge.

The Meta-owned platform quietly added the option on August 6, introducing a map that lets people share their location with friends – a tool similar to one Snapchat has offered since 2017.

It didn’t take long for worried posts to appear online. As reported by AFP, in one widely shared exchange, Instagram user Lindsey Bell said she was shocked to learn her location was visible to her followers.

“Mine was turned on and my home address was showing for all of my followers to see,” she wrote in response to a TikTok warning from reality TV personality Kelley Flanagan. “Turned it off immediately once I knew but had me feeling absolutely sick about it.”

In a TikTok video, Flanagan called the feature “dangerous” and walked viewers through how to disable it.

Instagram chief Adam Mosseri addressed the concerns in a post on Threads, saying the feature is off by default. “Quick Friend Map clarification, your location will only be shared if you decide to share it, and if you do, it can only be shared with a limited group of people you choose,” he wrote. “To start, location sharing is completely off.”

According to Instagram, the map is meant to help people share places they’ve visited and connect with friends. The company says users can choose who sees their location and can turn the feature off at any time.

A privacy concern on the heels of a legal case

The unease over Instagram’s map comes just a week after Meta faced scrutiny in court. A federal jury in San Francisco sided with women who said the company misused health data collected by Flo, a third-party app that tracks menstrual cycles and fertility.

The jury found that Meta knowingly received sensitive health information from the app and used it to target ads. Evidence presented in court included internal communications suggesting that some employees made light of the nature of the data.

“The case was about more than just data – it was about dignity, trust, and accountability,” said Carol Villegas, lead attorney for the plaintiffs.

How the new map works

Instagram’s map sits at the top of the messages inbox. It lets people share their live location while they’re using the app and see the locations of others who have chosen to share.

The company says no one can see a user’s location unless the person opts in. People can also limit the visibility to certain followers or turn it off entirely.

Meta describes the tool as “a new, lightweight way to connect with each other.” Similar functions exist on other platforms – Snapchat offers a personalised map, Apple’s iOS allows users to share locations with contacts, and Meta’s other apps like Facebook and WhatsApp have live location features.

A rocky reception

Despite these reassurances, the rollout has been met with scepticism. Many users, including professional creators, have raised safety concerns, warning that the feature could be misused for stalking or harassment.

The backlash has reached US lawmakers. According to NBC News, Senators Marsha Blackburn and Richard Blumenthal have reportedly written to Meta CEO Mark Zuckerberg, urging him to drop the feature.

Some confusion stems from how the map displays past posts. Users have reported seeing their older posts – ones with location tags – appear on the new map, even live location sharing enabled.

Mosseri explained that this is because the map includes both real-time locations and past posts with location tags. Those tags were already public to followers but weren’t previously collected in one place.

When location tags become personal

Allie Taylor, an educator who shares disability-related content on Instagram, said they learned about the map when followers messaged to say their location was visible. Taylor had posted a video tagged with the city of Cincinnati while at work. The map appeared to pinpoint the street they were on.

“It was terrifying,” Taylor said. “Why was this even a feature?”

Turning it off

There are several ways to disable location sharing. In Instagram, users can go to the messages inbox, open the map, tap the settings icon, and select “no one” for location sharing.

On a phone, location services for Instagram can be turned off entirely in the device’s settings.

Instagram promises changes

Mosseri has acknowledged the confusion and said the company will make the feature clearer. “We’re never going to share your location without someone actually actively asking to do so,” he said in a post last Friday.

In a statement, Meta said: “Instagram Map is off by default, and your live location is never shared unless you choose to turn it on. If you do, only people you follow back – or a private, custom list you select – can see your location.”

Mosseri also admitted that Instagram could “do a better job” explaining what appears on the map. “We can, and will, make it easier to understand exactly what’s happening,” he wrote, adding that improvements are planned for early next week.

Balancing connection and safety

The feature’s intent – to make it easier for friends to meet up and share experiences – is not new in social media. Apps have long offered ways to share location, from Snapchat’s Snap Map to Apple’s “Find My” function. The tools have drawn both praise for making coordination easier and criticism for the risks they pose when misused.

For Instagram, the challenge lies in ensuring that users understand exactly what they are sharing, with whom, and how it appears on the map. The backlash suggests that many people either missed or misunderstood the opt-in nature of the feature, especially when older posts appeared without warning.

Privacy advocates often caution that location data is especially sensitive. While a post tagged at a restaurant may seem harmless, patterns over time can reveal personal routines, places of work, or home addresses. The makes clear controls – and clear explanations – vital.

Instagram says it will continue to refine the map and its settings. Whether that will be enough to restore user confidence remains to be seen. In the meantime, those concerned about privacy have the option to disable location sharing entirely, either in the app or through phone settings.

Find out more about the Digital Marketing World Forum series and register here.

The post Instagram’s new map feature sparks privacy and safety concerns appeared first on TechWire Asia.

Huawei is expected to unveil a technology that could lessen China’s dependence on high-bandwidth memory (HBM) chips for running artificial intelligence reasoning models, according to the state-run Securities Times.

As reported by the South China Morning Post, the announcement will be made at the 2025 Financial AI Reasoning Application Landing and Development Forum in Shanghai today. The event focuses on AI in the financial sector.

Huawei did not respond to a request for comment on Monday. If confirmed, the development would mark another step by the US-sanctioned company in strengthening China’s AI hardware capabilities and reducing reliance on foreign technology.

HBM chips are a key component in advanced AI systems, particularly for running reasoning models. The models take an already-trained AI system and apply it to real-world data, making decisions based on patterns the AI has learned. HBM is important for these workloads because it can move large amounts of data quickly between the processor and memory.

The current market for HBM is dominated by US companies Micron and AMD, as well as South Korean firms Samsung Electronics and SK Hynix. The chips are often integrated directly into AI processors used in data centres.

China’s two main memory chip producers, Yangtze Memory Technologies and Changxin Memory Technologies, have expanded their capabilities, but analysts say they are still behind their US and Korean competitors in technical performance. That gap has left China dependent on imports for the most advanced HBM products, an issue made more pressing by US export controls on advanced chipmaking tools and technologies.

While China works to strengthen its domestic supply chain, demand for HBM worldwide is rising sharply. Orders have surged as major tech companies build more AI data centres.

Micron, one of the top HBM producers, raised its forecast for fourth-quarter revenue and profit on Monday, citing strong demand for AI infrastructure. The company now predicts $11.2 billion revenue, plus or minus $100 million, up on its earlier estimate of $10.7 billion. Adjusted earnings per share are forecast at $2.85, plus or minus 7 cents, up from a prior estimate of $2.50.

Micron also increased its adjusted gross margin outlook to 44.5%, from 42%, plus or minus 1%, pointing to stronger pricing notably in DRAM product lines.

“We look at all of our different end markets around the world, the pricing trends have been robust, and we have had great success in being able to push that pricing up,” said Sumit Sadana, Micron’s chief business officer, during an industry event on Monday.

Analysts say the combination of limited HBM supply and surging AI demand has allowed producers to raise prices – a reversal from past years when memory chipmakers faced shrinking margins.

SK Hynix, another leading HBM supplier, expects the market for AI-focused memory chips to grow by about 30% per year until 2030.

Trade measures could still affect the sector. The US recently imposed 100% tariffs on certain imported chips, although the duties will not apply to companies that manufacture in the US or have committed to doing so.

In June, Micron said it would increase its planned US investment by $30 billion, bringing its total commitment to $200 billion in the country.

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is co-located with other leading events including Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

The post Huawei to unveil tech to cut China’s reliance on foreign AI memory chips appeared first on TechWire Asia.

Chinese AI start-up Z.ai, formerly known as Zhipu AI, has made its GLM models compatible with Huawei Technologies’ processors, adding momentum to China’s efforts to build up its own technology supply chain, according to the South China Morning Post.

The Beijing-based company said its models now work with Huawei’s Ascend chips, used in AI servers, and Kirin processors, which power smartphones and laptops.

“The tie-up marks a major breakthrough in cloud-device collaboration between home-grown large [language] models and computational architecture, highlighting the deeper integration of a domestic AI ecosystem,” Z.ai said.

The move comes shortly after Huawei announced it would open-source its Compute Architecture for Neural Networks (CANN) – the software toolkit for its Ascend processors. Opening the code lets developers build, adapt, and scale applications for domestic chips without relying on foreign platforms.

CANN competes with Nvidia’s proprietary CUDA toolkit, long used by Chinese AI developers who depend on the US company’s GPUs in many data centres. Working with Z.ai helps Huawei push wider use of its own processors in AI projects.

Last month, Huawei’s Ascend division became a founding member of the Model-Chips Ecosystem Innovation Alliance, alongside Chinese AI companies like StepFun, Infinegence AI, SiliconFlow, MetaX, Biren Technology, Enflame, Iluvatar Corex, Cambricon Technologies, and Moore Threads.

Z.ai said it will use CANN to fine-tune its GLM models on Huawei’s Ascend-powered cloud, showing the open-source toolkit in action.

In June, OpenAI described Zhipu – before its rebranding – as making “notable progress” in delivering AI infrastructure to governments and state-owned firms in non-Western markets. Backed by more than 10 billion yuan (US$1.4 billion) in funding, Z.ai filed pre-IPO documents in April, with plans to go public as early as 2026.

China’s response to GPT-5

OpenAI’s latest flagship model, GPT-5, debuted last week with claims of being “smarter, faster, and more useful,” offering improved abilities in coding, maths, writing, health, and visual perception. It also includes a “thinking” function that switches between standard and deep reasoning modes depending on the task.

“It’s like a PhD-level expert in anything, any area,” said OpenAI CEO Sam Altman.

In China, where OpenAI services like ChatGPT are unavailable, experts were unconcerned about falling behind.

“GPT-5 is not significantly ahead of Chinese models, so it won’t put substantial pressure on Chinese researchers and developers,” said Zhang Linfeng, assistant professor at Shanghai Jiao Tong University. He added the model “doesn’t come with revolutionary breakthroughs; it lacks memorable characteristics.”

Zhang noted that the “thinking” feature is already in some Chinese systems, including Alibaba’s AI products. Still, he credited OpenAI with reducing hallucinations – incorrect AI outputs – and improving coding and general intelligence.

Despite the muted reaction, interest was high. A GPT-5 discussion on Zhihu drew more than 3.2 million views, with some users praising the upgrades. GPT-5 is now the default model on ChatGPT for both free and paid users, and Microsoft is adding it to products like GitHub Copilot and Visual Studio Code.

AI talent in high demand

Competition for AI experts in China is also heating up where Alibaba’s Tongyi Lab, creator of the Qwen open-source models, has lost senior staff to rivals.

Yan Zhijie, who joined Alibaba in 2015 and led Tongyi’s speech lab, left in February. He joined JD.com’s Explore Academy but later moved to Tencent, leaving soon after an internal restructuring. His role was filled by Li Xiangang, co-founder of 01.AI.

Bo Liefeng, former head of Tongyi’s applied vision division, has also moved to Tencent’s Hunyuan AI team. Tencent and Alibaba did not comment.

Alibaba says its Qwen models have been downloaded over 400 million times worldwide, leading to the creation of 140,000 derivative models.

The flow of talent mirrors trends in the US, where firms like Meta have hired away AI experts from Google, OpenAI, and Apple. In China, ByteDance, Alibaba, and Tencent have all launched fresh recruitment drives, many aimed at AI research roles.

ByteDance’s “Top Seed Talent Programme” lists 65 AI-related openings, while Alibaba’s “Star Top Talent” campaign is targeting researchers in foundational models, infrastructure, and AI applications. Tencent recently opened internal applications for roles tied to its Yuanbao chatbot, Hunyuan model, and WeChat e-commerce.

As China pushes to build its AI ecosystem with home-grown chips and models, the fight for top talent is becoming as important as the technology itself.

The post AI race in China heats up with new partnerships and talent moves appeared first on TechWire Asia.

OpenAI’s newest model, GPT-5, is now available to all users – including those on the free tier. The company says GPT-5 works faster, answers better, and is more useful for tasks like writing, coding, and helping in areas like health care.

OpenAI CEO Sam Altman said he didn’t enjoy switching back to GPT-4 after using the latest model. “I tried going back to GPT-4, and it was quite miserable,” he told reporters during a recent briefing.

ChatGPT, OpenAI’s chatbot powered by its language models, is now used by around 700 million people weekly. The company is also in early talks with investors about a possible stock sale that could value it at $500 billion, according to CNBC.

One of the key changes in GPT-5 is how often it “hallucinates” – a term used when AI makes up information. OpenAI says this happens less often with the new model. The company also ran more than 5,000 hours of safety tests while building it.

GPT-5 also takes a different approach when faced with risky or sensitive questions. Rather than refusing to respond entirely, the model offers a general response that avoids giving harmful details. “GPT-5 has been trained to recognise when a task can’t be finished, avoid speculation and can explain limitations more clearly,” said Michelle Pokrass, who works on post-training at OpenAI.

During the demo, the company showed how GPT-5 could turn simple prompts into working software. They asked it to create a language learning app to help English speakers study French. The prompt called for features like flashcards, quizzes, a fun theme, and progress tracking. Two separate GPT-5 sessions produced different versions of the app in seconds.

The apps needed a bit of cleanup, an OpenAI staffer said, but users can make changes easily – like adjusting the layout or adding new tabs.

The release marks the first time free ChatGPT users will get access to a model with reasoning capabilities – meaning it runs an internal thought process before replying. If they hit their use limit, they’ll be shifted to GPT-5 mini, a smaller version of the same model.

Subscribers to the Plus plan get more use room, and Pro users can access GPT-5 without limits, as well as GPT-5 Pro, a higher-tier version. Enterprise and education customers will start getting access about a week after the public rollout.

Microsoft, OpenAI’s close partner and investor, is bringing GPT-5 into its own products. A company blog post said GPT-5 will power Microsoft 365 Copilot and Azure’s AI tools. On social media, Microsoft CEO Satya Nadella reflected on how far AI has come since Altman visited Microsoft’s headquarters in 2023 to present GPT-4.

Other companies are already testing the new model. Box, which makes file management software for businesses, tried GPT-5 on large sets of data. CEO Aaron Levie said older models had trouble understanding long documents that involved complex math or logic, but GPT-5 handled them better. “The model is able to retain way more of the information that it’s looking at, and then use a much higher level of reasoning and logic capabilities to be able to make decisions,” he said in an interview with CNBC.

OpenAI is also giving developers three different versions of GPT-5 through its API: the standard model, GPT-5 mini, and GPT-5 nano. The are designed to support different needs when it comes to speed and cost.

Earlier in the week, the company also released two open-source language models, the first time OpenAI has released models like this since GPT-2 launched in 2019. The new models are meant to be easier and cheaper to use, especially for developers and researchers.

Altman called the two open-source models – GPT-OSS-120b and GPT-OSS-20b – “a big deal,” saying on social media that they’re “the best and most usable open models in the world.”

The release into the open came as China’s AI firms continue to release more open-source models. Companies like Alibaba Cloud have built popular model families, like Qwen, that are gaining international attention. According to Bloomberg, OpenAI’s recent moves – including GPT-5 and the open models – are part of a broader strategy as it considers a new stock sale.

OpenAI said its two open models perform about as well as o4-mini and o3-mini, which are among its smaller reasoning models. The release of open models suggests the company is returning to its original goal of making AI more accessible to the public – an effort many saw as slowing down in recent years. However, China appears to have the upper hand in terms of the number of competitive open models available.

“In the open-source space overall, China still has an edge over the US in the number of highly competitive models available,” said Ray Wang, a research director at consultancy Futurum Group. He added that OpenAI’s move might put pressure on its Chinese competitors.

The open-source trend picked up speed after Hangzhou-based DeepSeek introduced two low-cost, high-performance models in late 2023 and early 2024. That moment reportedly led Altman to question whether OpenAI had made the right call by not pursuing open-source releases as much as its competitors.

In the weeks since, more Chinese AI companies have entered the spotlight. Alibaba upgraded its Wan video generation tool and released Qwen-Image, a model that improves on previous iterations’ ablities at reading and editing text inside images. Zhipu AI, based in Beijing, launched GLM-4.5, which it says ranked third globally and first in China in a series of benchmarks. Another Beijing firm, Moonshot AI, released the Kimi 2 model last month, which has 1 trillion parameters – far more than OpenAI’s open-source offerings.

Altogether, China now has 1,509 AI models, both open and proprietary, according to data from the 2025 World Artificial Intelligence Conference in Shanghai.

OpenAI’s open-source models are getting attention in China too. A discussion thread about the models on Zhihu, a Chinese Q&A site, received over 250,000 views. Some local companies are paying close attention. Zhipu referred to the new models as “game changers.”

While GPT-5 and the open-source models are getting most of the attention, Altman says there’s more to come. He believes GPT-5 represents a step forward not just in speed and quality, but in how natural it feels to talk to.

“I think having something like GPT-5 would be pretty much unimaginable at any previous time in human history,” he said.

Rival companies are also claiming breakthroughs. Last month, Elon Musk said the latest version of Grok was “better than PhD level in everything” and described it as the “smartest AI” to date.

OpenAI is also promoting GPT-5 as a helpful coding assistant – part of a wider trend among US firms aiming at software developers. Anthropic’s Claude Code is one example of a model focused on similar tasks.

Altman said one major shift is how AI is changing what people can do with simple ideas. “People are limited by ideas, but not really the ability to execute, in many new ways,” he said.

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is co-located with other leading events including Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

The post GPT-5 now available to all users as OpenAI revisits open-source roots appeared first on TechWire Asia.

Alibaba has rolled out an updated version of its Qwen3 large language model (LLM), showing improvements in math, coding, reasoning, and multilingual tasks. As reported by the South China Morning Post, one variant of the model, the Qwen3-235B-A22B-2507-Instruct, has been released on Hugging Face and Alibaba’s ModelScope platform, along with a more efficient FP8 version that’s easier to run on limited hardware.

The company claims that the new version performs better in a range of tasks like logic, tool use, and long-context understanding. Based on benchmark tests, it also edges ahead of some competitors.

One key update is the FP8 model – short for 8-bit floating point – which cuts down on the memory and computing power needed to run the model. This makes it more useful for smaller teams or companies without access to large-scale infrastructure. Users can run it on basic hardware or deploy it more efficiently in the cloud, which helps lower energy costs and speeds up response times. Although Alibaba didn’t share exact numbers, similar FP8 setups typically reduce costs and resource use by a wide margin.

In short, the FP8 format allows organisations to do more with less. Instead of relying on massive clusters, users can run the model on a single GPU or even a personal machine, making it more practical for private use or local development. It also gives teams a chance to fine-tune models without the usual infrastructure headaches.

The new version also posted strong scores on several public tests. In the 2025 American Invitational Mathematics Examination, Qwen3 scored 70.3. That’s well above DeepSeek’s 46.6 and OpenAI’s GPT-4o, which scored 26.7. In coding tests, it pulled in 87.9 points on the MultiPL-E benchmark – just ahead of DeepSeek and OpenAI but slightly behind Claude Opus 4 Non-thinking from Anthropic, which scored 88.5.

As the name suggests, the latest Qwen model only supports a non-thinking mode – meaning it gives direct answers without showing steps or reasoning chains. But it’s able to process longer input, with a token limit now stretched to 256,000, making it better at handling large documents or long conversations in a single run.

The shift is tied to another change: Alibaba is dropping its earlier “hybrid” reasoning model approach, VentureBeat wrote. The idea behind hybrid mode was to let users switch between thinking and non-thinking behaviour, depending on the task. Users could toggle it manually – for example, adding a “/think” command before a prompt – to have the model work through a chain of logic before giving an answer.

That flexibility gave users more control, but also introduced design issues. Sometimes the model behaved unpredictably depending on the prompt, and users had to decide when to turn the reasoning on or off. After reviewing feedback, the Qwen team said it will now train separate models for instruction and reasoning tasks instead of blending both modes in one model.

A company post on social media read, “After talking with the community and thinking it through, we decided to stop using hybrid thinking mode. Instead, we’ll train Instruct and Thinking models separately so we can get the best quality possible.”

That means the new 2507 release focuses solely on following instructions and generating direct responses. For now, the reasoning model will be a separate track.

Alibaba is also starting to put its Qwen models into real-world products. A 3-billion-parameter version will power Xiaowei Hui, a smart assistant from HP that runs on the company’s PCs in China. The assistant is expected to help with tasks like writing and meeting summaries.

The Qwen3 series, launched in April, spans from smaller 600 million parameter models up to the larger 235 billion ones. One variant – the Qwen3-235B-A22B-no-thinking – ranks among the top open-source models globally, coming in just behind Chinese models from Moonshot AI and DeepSeek, based on a recent LMArena report.

Hugging Face’s own rankings from last month also placed several Qwen models in the top ten among Chinese LLMs, further boosting the model family’s reputation in the open-source AI space.

Nvidia CEO Jensen Huang also weighed in during his trip to China last week, where he acknowledged the progress China has made with its open-source AI work. He described Alibaba’s Qwen, along with DeepSeek and Moonshot’s Kimi, as “very advanced” and “the best open reasoning models in the world today.”

The Qwen team’s decision to split instruction and reasoning into separate models marks a shift in strategy – one that favours predictability and performance over flexibility. While this means users won’t be able to toggle reasoning on and off in a single model anymore, it could lead to better results in each task area.

For teams looking to run powerful models on lower-cost infrastructure, the new FP8 version adds another reason to consider Qwen3. And for those tracking benchmarks, Alibaba’s latest model is now firmly in the race with some of the best-known names in AI.

The post Alibaba’s new Qwen3 scores higher than rivals in key AI tasks appeared first on TechWire Asia.

Oracle is in talks to sign a large software deal with Skydance Media, a move contingent on Skydance completing its planned purchase of Paramount Global.

According to Bloomberg, sources familiar with the matter say the deal would be worth about US$100 million a year. If it goes through, Paramount and its group of media companies would begin using Oracle’s cloud software to run parts of their business. The discussions are private, and the companies haven’t confirmed any details.

Skydance already uses Oracle’s products with the media company spending US$2.2 million on Oracle’s cloud services in the year ending May 2024, according to public filings. Skydance was founded by David Ellison, whose father, Larry Ellison, is Oracle’s chairman and a major shareholder. Larry Ellison is also backing Skydance’s US$8 billion bid to take over Paramount, providing cash and investment support.

None of Oracle, Skydance or Paramount responded to requests for comment.

If the merger is approved and the Oracle deal is finalised, the combined Skydance-Paramount group would become one of Oracle’s larger cloud clients. It would also deepen the existing relationship between the two companies. A person familiar with the discussions said moving Paramount’s operations to Oracle’s cloud could lead to major cost savings – possibly in the hundreds of millions – as outdated systems are replaced with newer tools.

Paramount owns a number of high-profile media brands, including CBS, MTV, and Nickelodeon. The networks generate huge amounts of video and audio data needing a lot of storage and computing power. By shifting to Oracle’s cloud, the company could centralise much of that work, making it easier to manage and potentially offering cost savings.

David Ellison has said he wants to run Paramount more efficiently if the deal goes through. A big part of that plan involves updating its technology systems. While talks with Oracle are still ongoing, the deal isn’t finalised and could still change, according to sources.

The merger between Paramount and Skydance was first announced last year but still needs approval from US regulators. The final major hurdle is the Federal Communications Commission (FCC). Last week, Ellison met with FCC chair Brendan Carr and other officials to discuss the proposal and try to move things forward.

Meanwhile, Paramount has been facing scrutiny. Earlier in July, the company reached a legal settlement in a case brought by US President Donald Trump, who accused CBS News of media bias. On Monday (Jul 21), a group of US senators sent a letter to Ellison asking whether the timing of that settlement was tied to efforts to close the merger.

Oracle has been expanding its cloud business beyond its traditional database software. The company has been picking up customers that need large-scale computing and storage, particularly those working with artificial intelligence. Oracle already counts TikTok, Uber, and Zoom among its major clients.

The cloud deal with Skydance-Paramount comes shortly after Oracle announced a new round of spending in Europe. The company said it plans to invest US$3 billion over the next five years to expand its data centre capacity in Germany and the Netherlands. Oracle has earmarked US$2 billion for Germany and US$1 billion for Dutch operations.

According to Oracle, this investment will help local companies, governments, and organisations shift their workloads to the cloud and make use of newer AI tools. The announcement didn’t tie the spending directly to the Skydance talks, but it highlights the growing demand for AI computing infrastructure.

Big names in tech – including Microsoft, Amazon, Google, Meta, OpenAI, and xAI – have also committed large amounts of money to data centres that can power AI models. Oracle is part of the US-based group, Stargate, that includes OpenAI and others. The project plans to pour hundreds of billions into new infrastructure to support AI.

Oracle’s stock rose about 1% in early trading on the day the Skydance talks were reported. Shares are up more than 37% this year and are just below their highest point to date.

The post Oracle eyes major cloud deal if Skydance seals Paramount buyout appeared first on TechWire Asia.

OpenAI has rolled out ChatGPT Agent, a tool that the company says can take over digital tasks for users, carrying out actions on a user’s computer. That includes booking appointments, checking the calendar, browsing files, placing online orders, and building presentations, all with minimal user input.

Unlike earlier versions of ChatGPT that mainly focused on writing and answering questions, this works more like a digital assistant. It connects directly to your computer and can access apps, files, and the web to complete tasks. In early demos, the agent helped plan a wedding trip, purchased clothes online, and wrote workplace documents using live data.

OpenAI CEO Sam Altman called it a big step in AI’s ability to act on its own. He said the tool can “think, use tools, take actions, and repeat this process.” But he also warned that the technology is still experimental, stressing that people should be cautious when using it, especially when it comes to sensitive data like emails or bank accounts.

To reduce risk, OpenAI has built in several safety features. The agent needs permission before doing anything permanent – like sending emails or booking travel. It also comes with session logs so users can review what the agent did. There are system-wide limits in place to prevent it from doing anything dangerous, like creating harmful content or accessing restricted parts of the computer. Altman said the safest approach for now is to give the agent access to only what it needs, nothing more.

The company is releasing the tool to ChatGPT Pro, Plus, and Team subscribers first. Enterprise and Education users will get access later this summer. There’s no launch date yet for Europe or Switzerland.

The system is powered by a new AI model that OpenAI trained using reinforcement learning, the same method the company used for its other reasoning tools. The model doesn’t have a specific name, but it was built to use a mix of tools – like a browser, file manager, and terminal – and to handle multi-step tasks. Users can also upload their own data, which the agent can work with directly.

OpenAI combined two previous projects, Operator and Deep Research, to build the new agent. The team behind it includes about 20 to 35 people in product and research. In a demo shown to The Verge, product lead Yash Kumar and research lead Isa Fulford showed the tool planning a night out by checking a Google Calendar and then scanning OpenTable for restaurant options. The user could interrupt the process at any point to adjust the request, like switching cuisines or changing times.

Another demo had the agent create a research report comparing two toy trends. Fulford said she’s used the agent for online shopping and found it more thorough than using previous tools. Kumar mentioned using it for regular tasks like requesting a parking spot each week – something the agent now handles automatically.

Because the agent can access more than just the web – it can use the full computer – it can take on more advanced tasks. That includes putting together PowerPoint slides or analysing Excel files. But with that extra reach comes added risk, with “Watch Mode” purporting to add safeguards. If a user opens a sensitive web page, like a financial account, the agent stops working if the user switches tabs. This is meant to keep users in the loop when there’s potential for misuse.

The agent isn’t especially fast, especially for complex requests. But OpenAI says that speed isn’t the goal. Kumar noted that the team is focused on getting harder jobs right. Fulford added that even if it takes 15 to 30 minutes, it’s still faster than doing the task manually. The idea is that users can start a task, let the agent run in the background, and return when it’s done, presumably so ignoring mis-steps, errors, and hallucinations.

OpenAI has added controls to block the agent from being used in high-risk scenarios, like helping someone make dangerous chemicals or creating biological threats. The company said it hasn’t seen signs of that happening but added the protections anyway. Anthropic, a competing AI firm, introduced similar safeguards earlier this year.

The agent can’t complete financial transactions. Kumar said those functions are off-limits “for now.” Any steps toward that kind of control would require stricter safety checks.

AI agents like this have been a hot topic in the tech world for a while. The idea is to build tools that can go beyond answering questions and actually take action – something like Iron Man’s J.A.R.V.I.S., but more realistic. The term “AI agent” became common in 2023 after Klarna said its own system handled two-thirds of its customer chats in just one month, replacing work normally done by hundreds of people.

Since then, big companies like Amazon, Meta, and Google have all talked publicly about their plans to build similar tools. Google recently hired several leaders from Windsurf, a startup focused on this kind of AI, to help speed up its efforts.

ChatGPT Agent follows OpenAI’s earlier tool, Operator, which could browse the web and interact with buttons and forms. Other companies are moving in the same direction. Anthropic released a similar tool called “Computer Use,” which was built to use a PC like a person would. Several AI companies, including OpenAI, now offer “Deep Research” tools that can pull together detailed reports on just about any topic.

With Agent, OpenAI is trying to show what happens when AI moves beyond writing text and starts managing digital work. But even as the tool gets smarter, the company says it shouldn’t be left to run unsupervised. For now, the message is clear: the agent can be helpful, but it still needs a human nearby.

The post ChatGPT Agent can now do the work for you appeared first on TechWire Asia.

Ingram Micro is working to bring systems back online after a ransomware attack took down parts of its network, disrupting ordering systems and customer service channels across multiple regions.

The incident started on July 3. Customers first noticed problems when they could no longer place orders or reach the company through regular phone lines. Resellers and managed service providers — a major part of Ingram Micro’s client base — were among the first to report issues. Local websites began displaying maintenance pages with basic contact details for sales and support teams.

At first, the cause was unclear. The company described the issue as “technical difficulties.” Attempts to reach executives or press representatives went unanswered. It wasn’t until July 6 that Ingram Micro confirmed it had been hit by ransomware.

“Ingram Micro recently identified ransomware on certain of its internal systems,” the company said in a statement. It explained that after discovering the issue, systems were shut down as a precaution while cybersecurity specialists were brought in. Law enforcement was also notified.

The company added that it’s working to restore affected systems so it can resume processing and shipping orders. It apologised for the disruption caused to customers, vendors, and other business partners.

As of now, the outage continues to affect both physical product orders and license management systems, including Microsoft 365 and Dropbox. A source familiar with the situation said staff at Ingram Micro’s Bulgaria-based service centre were told to stay home on July 4 and avoid connecting their work laptops to the company network. Internal systems were still down at the time.

The disruption is significant given the scale of Ingram Micro’s operations. The company brought in $48 billion in revenue last year and posted $262.2 million in profit. It sells a wide range of products and services, from hardware and software to cloud licenses, IT asset disposal, returns management, logistics, and product remarketing. In Q1 this year, which ended on March 29, the company recorded $12.28 billion in revenue and $69.2 million in net income.

A group calling itself SafePay has claimed responsibility. Cybersecurity site Bleeping Computer obtained a copy of the ransom note, which accuses Ingram Micro of poor network security. The group claims it was able to access the company’s systems and stay inside undetected for some time.

“It was the misconfiguration of your network that allowed our experts to attack you,” the note says. “Treat this situation as simply as a paid training session for your system administrators.”

SafePay says it gained access to sensitive documents, including financial records, intellectual property, customer and employee data, bank account details, transaction history, and legal complaints. The group claims it encrypted all critical files and moved copies to a remote server. It says those files could be published online if no deal is reached.

“WE ARE THE ONES WHO CAN CORRECTLY DECRYPT YOUR DATA AND RESTORE YOUR INFRASTRUCTURE IN A SHORT TIME,” the note reads in all caps.

The attackers said the motive is purely financial. Ingram Micro has been given seven days to start negotiations.

At this point, the full extent of the breach remains unclear. Claims made by ransomware gangs often include exaggerations and should be verified independently. It’s also not confirmed how the attackers got in. One possible entry point, according to sources speaking to Bleeping Computer, is Ingram’s GlobalProtect VPN platform. That theory has not been confirmed.

SafePay was the most active ransomware group in May, according to threat intelligence company Fortra. The firm linked 70 separate incidents to the gang or its affiliates during that month alone. One of the group’s more prominent victims was fleet management software firm Microlise, which was attacked in October last year.

For now, Ingram Micro continues to operate with limited system access while recovery efforts are underway.

The post Ingram Micro struggles to restore services after ransomware breach appeared first on TechWire Asia.

Thousands of small and medium-sized businesses (SMBs) encountered cyberattacks in 2025 involving fake versions of popular productivity tools, according to new data from Kaspersky. Nearly 8,500 users were affected by malicious or unwanted software posing as legitimate apps — most often Zoom and Microsoft Office. Attackers also began using AI tools like ChatGPT and DeepSeek to trick users into downloading harmful files.

Kaspersky looked at how often threats were disguised as common online tools. Across 12 apps examined, researchers found over 4,000 unique malicious or suspicious files in 2025. A noticeable rise came from files pretending to be AI tools. ChatGPT-related threats jumped 115% in the first four months of the year compared to the same period in 2024. Kaspersky identified 177 files pretending to be ChatGPT and 83 mimicking DeepSeek, a large language model released in 2025.

Kaspersky’s Vasily Kolesnikov said attackers tend to go after tools that are widely talked about. “The more publicity and conversation there is around a tool, the more likely a user will come across a fake package on the internet,” he said. Kolesnikov advised SMB employees and everyday users to double-check URLs and avoid suspicious email links or software offers that seem too generous.

Aside from AI tools, collaboration platforms remain a common disguise for malware. Fake Zoom files rose nearly 13% to 1,652 this year. Threats mimicking Microsoft Teams and Google Drive also climbed — by 100% and 12%, respectively — with 206 and 132 files flagged. These tools have become essential for distributed teams, making them easy targets for impersonation.

Among the apps reviewed, Zoom stood out as the most copied, accounting for 41% of all detected threats. Microsoft Office apps were also high on the list: Outlook and PowerPoint each made up 16%, Excel nearly 12%, while Word and Teams followed at 9% and 5%.

The most common types of threats aimed at SMBs in 2025 were downloaders, trojans, and adware.

Phishing and spam tactics also on the rise

Kaspersky also reported a steady stream of phishing scams and spam aimed at SMBs. Many scams attempt to grab login credentials for services like banking platforms or delivery apps. One example involved a fake Google login page offering to promote a business on X — a scheme built to steal user credentials.

Spam continues to flood inboxes as well. Some messages now claim to offer AI-powered business automation. Others promote email marketing tools, business loans, or services like lead generation and reputation management — all crafted to appeal to small business owners.

According to Kaspersky, attackers are tailoring these phishing and spam campaigns to match what SMBs typically search for online, making them harder to spot at a glance.

The post Fake productivity apps and AI tools used to target SMBs in 2025 appeared first on TechWire Asia.