Singapore AI News & Daily Briefing

Google DeepMind is tying up with Singapore's government in a major national AI partnership. The deal covers AI tools for doctors, pandemic research using AlphaFold, a running app for blind athletes built with SG Enable, and free Gemini AI access for all teachers. The government estimates this could inject S$3.3 billion into Singapore's economy by 2040 through faster R&D. Teachers get AI training, and researchers will learn to use agentic AI tools for science.

Google's WeatherNext AI model predicted Hurricane Melissa would become a Category 5 storm hitting Jamaica five days in advance—something never done before. The early warning gave Jamaican officials time to evacuate communities and save lives. WeatherNext outperformed all other forecasting models for the 2025 hurricane season and is now expanding to help the Philippines, Taiwan, Indonesia, Vietnam, and other storm-prone regions.

IEEE's TryEngineering OnCampus program expanded from 2 to 7 universities in 2024, teaching students aged 10-18 about AI, robotics, quantum computing, and engineering through hands-on workshops. Over 300 students across Croatia, Egypt, Oman, Malaysia, Greece, and Hong Kong participated in activities ranging from circuit building to deep learning projects. The program aims to spark interest in tech careers early by giving kids real-world engineering experience.

Google unveiled its biggest search rede 25 years at I/O 2026. The search box now accepts images, PDFs, videos, and longer conversational queries — not just keywords. Google merged its AI Overviews and AI Mode into one seamless experience, meaning users get AI-powered answers alongside traditional results without switching screens. The company also introduced 'information agents' that monitor the web 24/7 for specific conditions and alert users when triggered. Under the hood, the new experience runs on Gemini 3.5 Flash, Google's fastest AI model yet.

IBM Research argues that raw AI language models alone won't drive enterprise AI adoption — companies need "agent logic," which are software tools like knowledge graphs and program analysis libraries that guide AI through actual business workflows. In tests, this approach reduced AI token costs by up to 30 times while improving accuracy. IBM shows real examples: understanding old COBOL code, generating software tests, investigating IT incidents, and automating compliance checks. A healthcare case study showed policy enforcement improved task correctness by 15-26% across different AI models.

South Africa sits on 88% of the world's platinum reserves—a critical mineral for semiconductors and data centers. It also hosts Africa's largest data-center market. Yet its draft AI policy treats the country as a passive consumer rather than a negotiating power. Microsoft, Google, Amazon, and Huawei are all building AI infrastructure there, competing for control of systems that will serve the entire continent. The author argues South Africa could leverage its mineral wealth for data sovereignty and technology transfer deals, but its withdrawn policy missed that chance.

Google DeepMind is launching an accelerator program specifically in Asia Pacific to support AI startups tackling environmental risks like climate change, pollution, and natural disasters. The program will provide funding, technical resources, and Google DeepMind expertise to selected companies working on AI-powered environmental solutions in the region.

Melbourne is rapidly becoming a major AI research hub, backed by Australia's largest university-based AI supercomputer MAVERIC at Monash University and expanding hyperscale data center campuses. The city is home to 188 AI firms and hosts major international tech conferences, creating a 'research flywheel' where infrastructure, discovery, and global collaboration reinforce each other. This matters because sovereign AI compute—keeping sensitive data within national borders—is becoming critical for medical research, drug discovery, and advanced engineering.

A Cambridge professor used Google's Co-Scientist AI to hunt for the molecular triggers behind diseases that jump from animals to humans, like Ebola and flu. The AI helped her pinpoint specific amino acids to test in just six months — work that would normally take two to three years. This matters because faster identification of how pathogens cause severe disease could improve outbreak responses and save lives.

A University of Edinburgh professor used Google's Co-Scientist AI to tackle a common liver disease called MASH. The AI figured out why a newly approved liver drug only works for some patients — it identified a specific molecular link never before connected. That hypothesis was then confirmed in lab tests, opening the door to combination drug treatments. The professor says AI feels like a 'jetpack' for scientists, dramatically shortening the path from idea to breakthrough.

AI is being used to tackle one of medicine's hardest problems — understanding aging. Researchers at Calico Life Sciences used Google's Co-Scientist to generate new hypotheses about cellular mechanisms that drive aging and disease. The AI helped cut through thousands of noisy, contradictory scientific studies to find ideas worth testing. One discovery about how cells respond to stress has already led to publishable findings.

Google DeepMind's AI tool Co-Scientist helped a mechanical engineer and a chemical biologist team up to tackle ALS (Lou Gehrig's disease). The AI connected their different expertise areas—tissue engineering and RNA biology—to find new drug targets. The pair is now hunting for RNA-based treatments that could slow or stop this devastating disease that affects nerve and muscle function.

Anthropic released "Cowork," a desktop AI agent that works in your files without needing to code. It can read, edit, and create files on your computer — like organizing your downloads folder, turning receipt screenshots into expense spreadsheets, or drafting reports from scattered notes. Built remarkably fast: reportedly in just 10 days, largely using Claude Code itself (meaning the AI helped build its own sibling product). Currently exclusive to Claude Max subscribers ($100-$200/month) on macOS, with Windows and cross-device sync planned. Anthropic is unusually transparent about risks: the AI can delete files and could potentially be tricked through prompt injection attacks online.

OpenAI has broken ground on a massive one-gigawatt data center in Michigan as part of its Stargate infrastructure program. The project will create jobs and support local communities while expanding the computing power needed to run AI systems like ChatGPT. This is part of a broader build-out to meet surging demand for AI services.

Salesforce rebuilt Slackbot from a basic notification tool into a full AI agent powered by Anthropic's Claude. It can search across Salesforce records, Google Drive, calendar data, and years of Slack conversations — then draft documents, synthesize insights, and create shared Canvas documents. Internal testing with 80,000 Salesforce employees showed the fastest product adoption in company history, with 96% satisfaction rates and users saving 2-20 hours weekly. MrBeast's company and other pilot customers report similar gains. Slackbot is free for Business+ and Enterprise+ customers, positioning Salesforce to compete directly with Microsoft Copilot and Google Gemini in the workplace AI race.

Google has released Gemini 3.5, its latest flagship AI model series that combines top-tier intelligence with the ability to take actions—like browsing the web, writing code, or managing files. Announced at Google I/O, the models are designed to be more capable partners for both everyday users and developers building AI applications.

Railway, a San Francisco cloud startup, raised $100 million in Series B funding to compete directly with AWS and Google Cloud. The company claims deployments in under one second — compared to the 2-3 minutes typical on legacy platforms. Railway built its own data centers from scratch and charges by the second (no idle VM fees), undercutting hyperscalers by roughly 50%. With just 30 employees, Railway processes 10 million deployments monthly and serves 31% of Fortune 500 companies. The pitch: as AI coding assistants generate code in seconds, legacy cloud infrastructure becomes a bottleneck.

Google is consolidating its AI-for-science tools under 'Gemini for Science,' including Co-Scientist for generating research hypotheses, AlphaEvolve for testing thousands of computational experiments in parallel, and NotebookLM features for analyzing scientific papers. Major companies like BASF and Klarna are already using these tools for supply chain optimization and machine learning. Validation papers were published in Nature.

Boston Children's Hospital has used OpenAI technology to help doctors diagnose over 40 rare diseases that often go unidentified for years. The AI helps reduce the paperwork burden on doctors so they can focus more on patients, and it spots patterns in medical data that humans might miss. For families dealing with undiagnosed children, this could mean faster answers and earlier treatment.

Google is launching Gemini Omni Flash, a new AI model that can create and edit videos using just text, images, audio, or existing video as input — all grounded in Gemini's real-world knowledge. Users can edit videos through natural conversation, change specific elements, or transform entire scenes simply by describing what they want. The feature is rolling out to Google AI subscribers and YouTube Shorts creators.

Companies like Hugging Face, Nvidia, and Alibaba are releasing open-source robotics AI tools, dramatically lowering barriers to building smart robots. Robotics datasets on Hugging Face grew from 1,145 to over 58,000 in months. You no longer need a PhD to make a robot reason and act—the tools are now freely available for anyone to build on.

Anthropic's Claude Code charges up to $200 per month but caps usage with confusing token-based limits. Now developers are turning to Goose, a free open-source AI coding agent built by Block (formerly Square). Goose runs entirely on your local machine using open-source models from Ollama — no cloud, no subscription, no rate limits. It works offline on flights. The trade-off: local models lack Claude Opus's raw power and million-token context window. Setup requires 32GB RAM and some technical know-how.

AI startup Majestic Labs is building Prometheus, a server with 128 terabytes of memory—over 60 times more than Nvidia's top AI server. The memory wall is a major bottleneck for large language models; this design uses DRAM instead of expensive HBM memory to handle larger AI models more efficiently. The company claims it can cut customer costs by 10-50x and reduce power consumption significantly. The server uses custom Ignite chips combining ARM and RISC-V cores and supports existing AI frameworks like PyTorch without code changes. Shipping expected 2027.

Google's AI tool Co-Scientist is helping biologists at MIT find genetic factors that can push aging cells back to a younger state. The AI scanned tens of thousands of research papers and proposed over 20 novel genetic targets to test — work that previously took months now takes days. Lab experiments validated some of these AI-suggested factors, showing cells improving overall function. This could eventually lead to treatments for aging-related conditions affecting skin, hair, and muscle.

OpenAI's most powerful AI models, including Codex for coding tasks, are now directly accessible through Amazon Web Services. This means businesses can use OpenAI's technology within their existing AWS setups without needing separate contracts or integrations. For everyday users, this could mean faster, smarter AI tools baked into the apps and services they already use through work or cloud platforms.

Google has released Antigravity 2.0, a standalone desktop app for interacting with AI agents — no coding background required. New features include scheduled tasks that run automatically (like cron jobs), subagents that work in parallel on complex problems, and voice input with live transcription. Google says millions of developers have already adopted its agent-first approach since launch.

Māori researchers at the University of Waikato built their own text-to-speech AI system for their dialect because Big Tech scraped their language data without permission or community benefit. Using just 7 hours 45 minutes of recordings from a consenting Māori speaker, they achieved word error rates competitive with major AI companies. The voice remains community-owned under tribal governance, setting a blueprint for indigenous data sovereignty in AI.

JetBrains released Mellum2, a 12 billion parameter open-source AI model that uses a Mixture-of-Experts design. It only activates 2.5 billion parameters per token, making it over twice as fast as similar-sized models while delivering comparable results. The model is designed for production workloads like routing requests between AI systems, summarizing documents, powering sub-agents, and running privately on company servers. Released under the permissive Apache 2.0 license.

Google DeepMind is connecting its Project Genie world-model AI to nearly 20 years of Google Street View imagery, letting users create new virtual environments grounded in real-world locations. This could transform how video games, simulations, and training environments are built — potentially allowing anyone to recreate actual cities or neighborhoods as interactive spaces.

Google DeepMind's Co-Scientist AI found a cancer drug called vorinostat that could treat liver scarring (fibrosis), which kills over 1.4 million people yearly. The AI suggested three candidates from existing medicines, and two blocked fibrosis in live human liver cells—one that was barely mentioned in prior research. This shows AI can find hidden gems in medical literature for drug repurposing.

A startup called Wetour Robotics is building wearable interfaces that let people control robots and devices using body signals, gestures, and eye focus—no screens, buttons, or voice commands needed. Their wristband reads muscle signals 50-80 milliseconds before visible movement, letting devices anticipate intent. This could help factory workers, wind turbine technicians, or mobility device users command nearby machines while their hands stay busy.

Johns Hopkins Applied Physics Laboratory researchers are developing AI agents that can coordinate teams of heterogeneous robots using large language models. Their presentation covers architecture design for multi-robot coordination and lessons learned from real hardware demonstrations.

Google is expanding tools that help people identify AI-generated images and videos. With deepfakes becoming more realistic, these new features aim to let users verify whether content was created or edited by AI. This matters because anyone scrolling social media or news sites needs a way to spot fakes before sharing or believing what they see.

HuggingFace published a glossary to standardize confusing AI agent terminology. It explains key concepts: a "model" is just the language engine, a "scaffold" defines behavior through prompts and tools, a "harness" executes the agent loop, and an "agent" combines all three. The guide also covers context engineering, skills, sub-agents, and training concepts like rewards and rollouts. Aimed at clearing up industry confusion after terms like "harness" and "scaffold" were used inconsistently at recent AI conferences.

An AI system called WhaleSpotter uses thermal cameras on Angel Island to detect gray whales surfacing in San Francisco Bay, then alerts nearby ships to slow down and avoid deadly collisions. Ship strikes killed 40% of the 21 gray whales that died in the bay last year. The system has logged over 6,600 detections in its first 10 days. Thermal cameras outperform human spotters in fog and darkness, and the AI reduces strike risk by 90% according to WhaleSpotter. Human verification is still needed to prevent false alarms.

A veteran chip designer shares what it's like moving from academia to the semiconductor industry. The key difference: academia asks "what's possible?" while industry asks "what works reliably at scale?" The ASIC market is booming—expected to hit $38.8 billion by 2033—but academic training often doesn't prepare engineers for industry realities like $30 million lithography masks and exhaustive verification requirements.

Scientists trained radar systems with machine learning to identify pollinator insects like bees and wasps by their unique wingbeat patterns. Using millimeter-wave radar, they achieved 85% accuracy distinguishing five species and 96% accuracy differentiating between bee and wasp families. The non-invasive method could replace harmful insect trapping and enable real-time ecosystem monitoring without killing the insects.

This essay traces how "social engineering"—once a neutral term for shaping human behavior at scale—became associated with manipulation and scams. It argues the term got a bad reputation due to abuses by authoritarian regimes, and today the practice hides behind softer labels like "user experience" and "behavioral nudges." The author contends we need to name these techniques openly to hold companies accountable for designing algorithms that manipulate user behavior for profit.

MathWorks is hosting a free webinar on using MATLAB and Simulink to design, train, and deploy AI-based virtual sensor models to embedded hardware. The workflow covers integration, verification, model compression, and code generation for edge AI applications. Target audience appears to be engineers working on embedded systems and IoT devices.

IEEE president Mary Ellen Randall argues that digital systems were built for adults, not children, despite one-third of internet users being under 18. Countries including Indonesia (first in Asia), Brazil, Australia, the EU, and UK are now adopting age-appropriate design regulations to protect kids from addictive features and harmful algorithms. Randall says engineers must rethink how data is collected, how AI influences developing minds, and how interfaces shape attention. IEEE is helping governments translate children's rights principles into actual engineering standards.

A robotics maker built a digital clock using millifluidics — logic circuits powered by pressurized air instead of electronics. The clock's face is soft silicone that morphs as air cavities expand and contract. By offloading computation to the same air that powers movement, the clock needs only 11 valves instead of 29, making it simpler and lighter. The project shows that computation doesn't always need silicon — sometimes flexible silicone and airflow are enough.

Singapore-based data center developer and operator DayOne Data Centers has closed a $4.5 billion Series C funding round — one of the largest data center fundraises in the Asia Pacific region. The company plans to use the funds to expand across Asia Pacific and Europe. This massive investment reflects the insatiable demand for AI computing infrastructure, as companies race to build capacity for training and running AI models.

A British photographer built a bipedal robot called Shadow Walker in 1987 using compressed air muscles instead of motors, pioneering a DIY approach to humanoid robotics. The robot couldn't walk properly despite years of effort, but the project led to Shadow Robot, Britain's oldest robotics company, which now focuses on dexterous robot hands. The article shows how far robotics has come—modern humanoid robots can now compete in sports and do useful tasks that would've seemed impossible then.

OpenAI has upgraded ChatGPT's memory to 'Dreaming V3' — it now builds detailed profiles from your chat history, synthesizes information autonomously, and derives preferences even from casual remarks. The feature is rolling out to all users. The catch: it can generate incorrect assumptions about you (like claiming you use smart home systems you don't), and disabling memory doesn't delete stored data. Your entire chat history remains accessible to the AI even after you turn it off.

PhysicsX, a London-based AI company that builds software to simulate and optimize complex engineering processes in sectors like aerospace, automotive, and semiconductors, has raised $135 million in Series B funding. The round was led by European VC Atomico, with Singapore's Temasek among the participants alongside Siemens, Applied Materials, and General Catalyst. The company, founded by ex-Renault F1 and QuantumBlack alumni, plans to use the funds to expand globally and accelerate AI adoption across heavy industries still relying on legacy engineering tools.

Prime Minister Lawrence Wong warned that AI is making it easier to spread misinformation and disinformation, and that Singapore will use tools like the Protection from Online Falsehoods and Manipulation Act (POFMA) to crack down when content crosses the line. He said this at a dialogue with the Singapore Press Club, addressing a recent case where authorities blocked 14 online posts targeting the Indian community that likely originated from a China-based platform. Wong also said the Government depends on media outlets to operate responsibly and help hold society together amid an era of AI-generated content.

Vietnam's FPT Corporation and Nvidia have released a free AI training dataset for Vietnamese language and culture. The dataset contains 900,000 fictional Vietnamese personas based on real national statistics. It aims to help developers build AI systems that better understand local context — from regional dialects to economic conditions. The data is freely available for commercial use on HuggingFace.

Singapore has switched on Aspire 2B, a new national supercomputer with over 1,500 Nvidia H200 GPUs — four times the power of its predecessors combined. That's enough computing muscle to match 120,000 high-end AI laptops. The machine will let Singapore researchers build bigger AI models for weather forecasting, coastal protection, disease prediction, and Asian-language chatbots without sending data overseas. Public researchers from universities, government agencies, and research institutes — more than 9,000 of them — can now access this infrastructure.

South Korean venture firm SBVA invested in AMI Labs, an AI research institute founded by Meta's Yann LeCun, alongside backers including Bezos Expeditions, Nvidia, and Singapore's Temasek. AMI Labs raised a $1.03 billion seed round to build "world models" that understand the physical world differently from current AI. Doosan and Coupang also joined as limited partners, linking Korean manufacturing and logistics capabilities to cutting-edge AI research with a planned dedicated fund.

Singapore-based Grab has proposed building an open electric vehicle charging network across Hanoi and offering AI tools to help small businesses and households in Vietnam's capital work more efficiently. Grab's president met with Hanoi officials to discuss a formal cooperation agreement, aligning with the city's low-emission zone roadmap. The move would also let Grab leverage its large base of foreign tourists using the app to promote Hanoi to international visitors.

Apple's annual developer conference WWDC kicks off this week, where it will unveil its latest AI software plans. Expected highlights include a major Siri overhaul powered by Google's Gemini AI, improved image generation tools, and custom emoji creation from photos. This marks Apple's shift from OpenAI's ChatGPT to Google as its primary AI partner. It will also be the last WWDC led by CEO Tim Cook, who retires soon.

US data center giant Digital Realty is building a massive AI-ready facility in Cyberjaya, Malaysia, with plans for three connected sites totaling 32 megawatts of power. The campus will host cloud providers, enterprises, and network services, linking them to Digital Realty's existing hubs in Singapore and Jakarta. Malaysia's government sees this as a cornerstone of its ambition to become an AI nation by 2030.

Singapore-founded PatSnap, which helps companies search patents and avoid infringements, is using generative AI to expand beyond patent departments into R&D teams. The unicorn company—valued after Sequoia China and Vertex Ventures funding—has grown revenue at 20% annually since 2023 and crossed US$100 million in annual recurring revenue in 2024, while also turning profitable. Its AI agents help R&D engineers spot drug targets in life sciences and cross-pollinate ideas across industries. CEO Jeffrey Tiong has pushed 'tokenmaxing'—giving employees US$1,000 monthly in AI tokens to encourage adoption—and has ditched PowerPoint entirely for AI-generated HTML discussions.

Singapore launched a new programme to prepare university students for AI roles in banking and finance. The Young Talent Programme for AI in Finance (YTP-AIF) offers AI bootcamps, masterclasses, and internships with major banks like JPMorgan, DBS, OCBC, UBS, and Standard Chartered. Students can earn academic credit while working on real financial AI challenges. The pilot kicks off August 2026 with around 100 students from any Singapore university.

Equinix is pouring more than US$190 million into a new data center in Cyberjaya, Malaysia. The KL2 facility is built for AI workloads, supporting high-density GPU setups and liquid cooling systems needed for modern AI computing. The project aligns with Malaysia's push to attract AI infrastructure, though the country has tightened approvals for new data centers outside AI-specific projects. Cyberjaya is already a regional data hub with over 22 facilities, making it a natural spot for this kind of expansion. For users, this means more cloud capacity for AI services across Southeast Asia, though it could also push up local energy costs.

Singapore-based VC firm Iterative has cut nearly half its team (from 16 to 9 employees) and shut down programs like its Scale investment arm and debt funding to focus purely on AI startups. The firm, which has backed 191 companies across two funds worth $65 million, is now zeroing in on "AI-fluent founders" and plans to take startups to San Francisco to connect with US investors. Its latest accelerator cohort accepted just eight startups — one of the smallest since the program launched in 2020 — because few applicants met its new AI standards. Co-founder Brian Ma says Southeast Asia's AI edge lies in applying the technology to logistics, fintech, and education.

AI company Anthropic has filed confidentially for an IPO as it targets a trillion-dollar valuation. The timing is tricky: while Anthropic has captured most of the enterprise B2B market, companies are now scrutinizing whether AI is worth what they're paying. A Bain survey found 40% of 900 companies worldwide saw cost savings below 10% from AI — worse than the 11-20% they expected. A separate BCG survey showed Asia-Pacific leads globally in AI adoption among frontline workers, which could help Anthropic's case. Meanwhile, the US government is now asking AI labs to share new models 30 days before public release, partly because Chinese LLMs already dominate global usage (seven of the top 10 most-used models come from Chinese developers). Anthropic is also expanding in Singapore, though cautiously — it has just four job postings here, far behind OpenAI which opened a Singapore office in 2024.

Nvidia CEO Jensen Huang says agentic AI — AI systems that can reason, use tools, and complete tasks on their own — is creating a new computing architecture for PCs, robots, and data centers. He pointed to RTX Spark, a new class of Windows PCs with up to 1 petaflop of AI power and 128GB of memory designed for personal AI agents. This means future computing will split work across local devices, home systems, and cloud instead of relying on one environment. For everyday users, this could mean AI assistants that run more tasks on your own devices without constant internet access.

Meta has confirmed that at least 20,225 Instagram accounts were compromised through a security flaw in its AI support chatbot. For nearly seven weeks, the chatbot sent password reset links to random email addresses without verifying ownership — the opposite of what it was supposed to do. Meta had marketed the tool as improving account security.

Digital Realty has opened a major data center campus in Cyberjaya, Malaysia, with three interconnected facilities designed to support AI and cloud computing across Southeast Asia. The expansion brings more regional capacity to handle AI workloads for businesses in the area.

India's Tata Consultancy Services (TCS) partnered with investment firm TPG to pour up to $2 billion into HyperVault, a new AI data center business. The plan is to build AI-ready facilities in India exceeding one gigawatt of capacity across locations including Navi Mumbai, Chennai, and Hyderabad. However, analysts note the $6.6 billion budget requirement means heavy borrowing, and key elements like grid allocation and renewable power agreements remain unsigned—raising questions about execution.

Cisco unveiled its Silicon One G300 switch chip, built on TSMC's 3-nanometer process, designed to speed up data transfer in large AI data centers. The chip claims up to 28% faster AI processing and 33% higher network efficiency compared to existing techniques. Cisco is targeting hyperscalers—the massive cloud operators—with $1.3 billion in AI infrastructure orders already booked. The move positions Ethernet as a stronger competitor to Nvidia's InfiniBand for AI clusters.

Microsoft has tightened its rules for cloud services in conflict zones after an investigation into how Israel's military used Azure. The report revealed AI-powered target selection and mass surveillance capabilities were deployed in Gaza, raising human rights concerns. Microsoft is now implementing new human rights checks, though key questions about the actual data processed remain unanswered.

Global tin demand from AI data centers is expected to triple by 2030, but supply is tightening. Tin is essential for soldering components in servers. Prices have risen over 55% in a year as data center builders compete for raw materials. Most tin comes from volatile mining regions, creating supply chain risks. This could push up costs for AI services and new electronics.

Cohesity, a data security company, has joined Anthropic's Project Glasswing program, gaining access to Claude Mythos Preview—one of Anthropic's most advanced frontier models. Cohesity plans to use the AI to automatically find, verify, and fix security weaknesses across its platform. The partnership reflects a growing trend of security firms tapping frontier AI to bolster cyber defenses beyond traditional methods.

A £40,000 AI-powered robot named D4RYL has been rejected for membership to Britain's Magic Circle society of magicians. The society's president said the robot was 'extraordinary' but lacked the human spontaneity and empathy needed for real magic. The robot's owner argues D4RYL creates genuine wonder and emotional connection with audiences.

Israeli AI chipmaker Hailo has laid off about half its workforce, according to reports. The company makes chips designed to speed up AI tasks on edge devices like cameras and sensors. Raspberry Pi uses Hailo chips in its AI add-on products. The cuts suggest even companies making specialized AI hardware are struggling in a crowded market. It signals consolidation pressure in the AI chip space.

Nvidia is betting big that AI-powered PCs will move beyond tech enthusiasts into mainstream households, but analysts question whether regular users actually need or want these features. The chipmaker has been pushing its AI PC vision, though the broader PC market has been sluggish. The success of this push could reshape consumer hardware spending and determine whether AI capabilities become standard in everyday computers.

A growing number of tech companies are abandoning 'tokenmaxxing' — the practice of tracking how many AI tokens employees use and tying it to performance reviews. The CEO of AI chip firm Cerebras called unlimited token budgets 'boneheaded,' and companies like Amazon and Uber are now capping AI spending per employee. The shift reflects growing concern that the practice wastes money and doesn't actually improve work quality.

# India's TCS, TPG partner to invest $2b in AI data centers [News](https://www.techinasia.com/)Premium * [News](https://www.techinasia.com/) * [Jobs](https://www.techinasia.com/jobs/search) * [Companies](https://www.techinasia.com/companies) * [Events](https://www.techinasia.com/events/live) * [About](https://www.techinasia.com/about) * [advertise](https://www.techinasia.com/advertise?utm_source=techinasia.com&utm_medium=website&utm_campaign=intl-navbar) * [Asia Economic Summit](https://aes.techinasia.com/?utm_source=navbar-tia-homepage&utm_medium=website&utm_campaign=events-flagship-aes) * [Sign Up](https://www.techinasia.com/auth) * [Log In](https://www.techinasia.com/auth?ref=nav_login) [Subscribe](https://www.techinasia.com/subscription) [I'm fundraising](https://www.techinasia.com/com

Anthropic has opened its first Asia-Pacific office in Tokyo and signed a memorandum of cooperation with the Japan AI Safety Institute. The US-based AI startup aims to collaborate on AI evaluation methodologies and monitor trends in the field. Anthropic’s CEO Dario Amodei met with Japanese Prime Minister Takaichi and local officials during the visit. The company has also joined the Hiroshima AI Process Friends Group, which seeks to promote safe and trustworthy AI development. Anthropic is working with Japanese firms, including Rakuten, Nomura Research Institute, and Panasonic, to integrate its AI model, Claude, into business operations. Anthropic plans to expand further in the Asia-Pacific region and will bring its approach to Seoul and Bengaluru. _🔗 Source: [Anthropic](https://www.anthropi

A technical blog post exploring emergent behaviors in multi-model AI systems, examining how control and complexity arise when five AI models interact in a simulated economy. Likely covers research on autonomous AI behavior and system-level dynamics.

Microsoft used its Build 2026 conference to announce it is building its own AI stack, reducing dependence on partners like OpenAI and Anthropic. The company launched seven in-house AI models under the MAI brand — including a flagship reasoning model — and introduced new server processors co-designed with its own Maia AI chips. Microsoft also rolled out an agent platform spanning Windows, Azure, and GitHub, and demonstrated Frontier Tuning, which lets companies customize AI models inside their own compliance boundaries. The strategy positions Microsoft alongside Google and Amazon, which already own their full AI stack from chips to models. Benchmarks and performance claims come from Microsoft's own testing and remain largely unverified by third parties.

Nvidia is building a new campus called Constellation in Taipei's technology park, spanning nearly four hectares. CEO Jensen Huang announced the plans at an employee meeting. Nvidia is already hiring heavily in Taipei as its Taiwan supply chain grows. Foxconn and TSMC are already using Nvidia tools for factory digital twins and robotics.

South Korea's Naver raised $1.1 billion through dollar and euro bonds to fund AI and data center expansion. This includes eco-friendly data centers and projects to boost energy efficiency. Orders from investors were nine times the issue size. Naver plans to invest over $680 million in GPUs this year and is exploring AI data center projects in Europe.

Xiaomi has started testing humanoid robots at its electric vehicle factory in China. The robots use AI to load nuts and move materials with about 90% success. CEO Lei Jun wants large-scale deployment within five years. This puts Xiaomi in direct competition with Tesla, Figure AI, and Agility Robotics in the humanoid robot race.

AI coding tools are enabling non-engineers to build working applications without programming experience. An Anthropic employee mentioned a company lawyer with no coding background successfully built a functional app using AI. This shift toward 'citizen developers' is changing how enterprises build internal tools, potentially reducing dependence on traditional software teams.

Qualcomm's CEO says robotics will become a major business opportunity for the company within two years, powered by AI advances. The chipmaker launched a new Dragonwing robotics processor in January designed to work across multiple robot platforms, similar to how Snapdragon powers smartphones. Robotics is seen as the next frontier beyond phones and PCs, with analysts estimating a potential $370 billion market by 2040.

LG Group has launched South Korea's first corporate-accredited AI graduate school at its Seoul campus. Employees from LG Electronics, Energy Solution, Innotek, Display, and Chem can earn master's or doctoral degrees with full tuition support. The program draws on LG AI Research and partners with universities like Seoul National University and KAIST. This move lets LG train its own AI talent pipeline rather than competing in the expensive external job market.

Chinese AI firm Moonshot AI, backed by Alibaba, just released a powerful new version of its Kimi chatbot. The K2.5 model can handle text, images, and videos together, and includes an automated coding tool to compete with Anthropic's Claude Code. The company is raising funds and seeking a valuation near US$5 billion.

LG CNS, the IT arm of South Korea's LG Group, is opening a new research center in Silicon Valley focused on AI and robotics foundation models. This is LG Group's second R&D hub in the Valley after LG Electronics. The move comes as LG invests in AI startups like Skild AI and partners with companies like Canada's Cohere.

Tony Fadell, the inventor of the iPod and co-creator of the iPhone, is warning tech builders not to rely too heavily on AI. He argues that building truly new products requires human storytelling, judgment calls, and moral choices—not just data and team consensus. While AI can speed up early testing, blindly trusting AI to write code creates security risks and weak products. His message: use AI as a helpful tool, but do not surrender your critical thinking to the machine.

This opinion piece examines how dozens of AI companies are now offering nearly identical products — think chatbots, writing assistants, and image generators — creating a crowded marketplace where it's hard to stand out. The "gold rush" analogy suggests most players are competing for the same customers using the same playbook. For everyday users, this means more choices but potentially lower quality as companies rush to market with copycat features. It may also mean subscription prices could drop as competition heats up.

A startup working on adult-themed AI products went viral after posting job listings for unusual roles. The company reportedly received an overwhelming number of applications, highlighting the unusual ways AI startups are approaching hiring and product development.

South Korea has created a government-backed working group with Samsung, Hyundai, SK Telecom, and Naver Cloud to deploy up to 260,000 Nvidia Blackwell GPUs across the country. This would increase Korea's total AI GPU capacity from 65,000 to over 300,000 units, positioning it as a major AI computing hub. The initiative signals aggressive national investment in AI infrastructure comparable to efforts in the US and China.

OpenAI has filed a confidential S-1 with the SEC, signaling it is preparing for a potential public stock offering. The company says it may take time to go public as some goals are easier to achieve as a private company, but this gives them the option to list sooner if needed. This could become one of the biggest tech IPOs in history.

Anthropic has opened its first Asia-Pacific office in Tokyo, marking a major push into the Japanese market. The company, backed by Amazon, has signed deals with major Japanese firms including Rakuten and Nomura Research Institute to integrate its Claude AI model into business operations. CEO Dario Amodei met with Japan's Prime Minister during the visit, and Anthropic has joined Japan's Hiroshima AI Process Friends Group focused on safe AI development. The company plans to expand further to Seoul and Bengaluru. This signals growing international competition among AI companies to establish footholds in Asia's enterprise market.

Amazon software engineers showed up at Seattle City Council meetings to oppose new AI data centers—while their own employer is spending $200 billion on AI infrastructure and has laid off 30,000 corporate workers in eight months. Seattle responded by passing a one-year moratorium on large AI data center construction. This highlights growing public frustration that big tech companies are betting big on AI while cutting human workers.

Google has signed a massive $30 billion deal to rent SpaceX's AI computing power through June 2029 — that's $920 million per month. Google will tap roughly 110,000 Nvidia GPUs owned by SpaceX to power its AI infrastructure. The deal comes just days after Anthropic struck its own SpaceX compute deal, signaling a scramble among Big Tech to lock up AI capacity.

ArXiv, the popular open-access repository for academic preprint papers, will now ban researchers for one year if they submit obviously AI-generated work without reviewing it. Penalties apply when papers contain telltale signs of unchecked AI output — like hallucinated references, misleading content, or even LLM meta-comments left in the text. After the ban, researchers must have subsequent submissions peer-reviewed at reputable venues before posting.

The UK government announced a £1.2 billion ($1.5 billion) plan to build AI infrastructure, including a new supercomputer and chip funding. The initiative aims to position Britain as a major AI player and reduce dependence on US AI compute resources. It follows similar sovereign AI ambitions from the EU and other nations.

This is a comprehensive preview of what Apple is expected to announce at WWDC 2026: a major Siri overhaul powered by Google Gemini, making the voice assistant conversational and context-aware; a standalone Siri app to compete with ChatGPT; an AI agent app store integration for tasks like booking reservations; upgraded Image Playground with better quality and editing; enhanced Photos AI editing; and Apple Wallet improvements including smart bill-splitting from receipt photos. Rollout is expected later this year.

OpenAI is planning to transform ChatGPT into a 'super app' before its IPO, bundling coding tools and AI agents into one platform. A senior OpenAI employee told the Financial Times that 'chat is dead' — the company is pivoting from simple chatbots to autonomous AI systems that can complete complex tasks on behalf of users.

Two-thirds of America's planned AI datacenters — 517 out of 809 — are being built in drought-stricken areas already struggling with water shortages. These facilities can require up to 5 million gallons of water daily for cooling. By 2028, US datacenters alone could demand 73 billion gallons of water per year, raising fears of clashes with farmers and communities over water access.

Apple's annual developer conference kicked off with major AI announcements. The star: a completely overhauled Siri powered by Google's Gemini, turning it into a proper AI assistant that can understand your screen context and act on your behalf. Other highlights include macOS dropping Intel chip support, requiring Apple M1 or newer, and new AI writing tools for iOS. It's also Tim Cook's final WWDC before handing the CEO role to John Ternus.

Uber has hit the brakes on AI spending after burning through its entire 2026 coding budget in just four months. The ride-hailing giant capped engineers at $1,500 per month for AI coding tools like Claude Code and Cursor, even as CEO Dara Khosrowshahi confirmed AI agents now write roughly 10% of the company's code. The core problem: while AI costs per task keep dropping, the sheer volume of tasks being automated is causing total bills to skyrocket — and companies are struggling to connect token spending to actual business results.

Xiaomi has achieved a major AI speed milestone by reaching over 1,000 tokens per second on a 1-trillion-parameter model using just standard 8-GPU servers — no specialised hardware required. They used clever tricks like compressing the model to use less memory (FP4 quantization) and a smarter method where a smaller "draft" model guesses tokens for the big model to verify in batches. For everyday users, this means AI could respond nearly instantly — think coding assistants that write code in real-time, chatbots that feel like talking to a person, or AI tools that could assist in time-critical situations like medical analysis.

Nvidia is partnering with South Korea's Naver—essentially Korea's version of Google—to build massive AI data centers called "AI factories." Starting at 55 megawatts and aiming for gigawatt-scale operations, these facilities will run on Nvidia's full AI stack platform. Jensen Huang says this will help Korea build "sovereign intelligence" for AI agents and real-world services using Naver's regional AI models.

OpenAI has proposed creating a sovereign-wealth-style fund that would give everyday Americans a financial stake in the AI industry. The idea is to let the public share in the massive valuations being generated by AI companies—even though most AI firms are currently unprofitable. If implemented, this could set a precedent for how AI wealth gets distributed broadly, rather than concentrating gains among investors and tech workers.

Nvidia's CEO Jensen Huang revealed that the company's new Vera CPU, designed for AI systems, will use memory from SK Hynix. Nvidia sees a massive $200 billion market for this new chip category, and notably, Huang included China in that market estimate despite ongoing US tech restrictions. The Vera CPU is designed to work alongside Nvidia's GPUs in AI infrastructure, opening new opportunities beyond graphics processing.

OpenAI is preparing a major overhaul of the ChatGPT app interface, with the company reportedly planning to file its IPO soon. The redesign aims to transform ChatGPT into a 'superapp' that handles multiple tasks beyond simple conversations. This could make AI assistance more accessible and integrated into daily life for millions of users.

Samsung's chip manufacturing division is expected to turn profitable again by late 2026 after struggling with low yields on its advanced 2-nanometer chips. The company has now improved its 2nm chip yield above 60%, which is a key milestone for competing with rivals like TSMC. This matters because Samsung fabricates chips for many devices, from phones to servers, and better yields mean more reliable supply chains.

This opinion piece examines how dozens of AI companies are now offering nearly identical products — think chatbots, writing assistants, and image generators — creating a crowded marketplace where it's hard to stand out. The "gold rush" analogy suggests most players are competing for the same customers using the same playbook. For everyday users, this means more choices but potentially lower quality as companies rush to market with copycat features. It may also mean subscription prices could drop as competition heats up.

Singapore launched a new programme to prepare university students for AI roles in banking and finance. The Young Talent Programme for AI in Finance (YTP-AIF) offers AI bootcamps, masterclasses, and internships with major banks like JPMorgan, DBS, OCBC, UBS, and Standard Chartered. Students can earn academic credit while working on real financial AI challenges. The pilot kicks off August 2026 with around 100 students from any Singapore university.

Qualcomm's CEO says robotics will become a major business opportunity for the company within two years, powered by AI advances. The chipmaker launched a new Dragonwing robotics processor in January designed to work across multiple robot platforms, similar to how Snapdragon powers smartphones. Robotics is seen as the next frontier beyond phones and PCs, with analysts estimating a potential $370 billion market by 2040.

AI company Anthropic has filed confidentially for an IPO as it targets a trillion-dollar valuation. The timing is tricky: while Anthropic has captured most of the enterprise B2B market, companies are now scrutinizing whether AI is worth what they're paying. A Bain survey found 40% of 900 companies worldwide saw cost savings below 10% from AI — worse than the 11-20% they expected. A separate BCG survey showed Asia-Pacific leads globally in AI adoption among frontline workers, which could help Anthropic's case. Meanwhile, the US government is now asking AI labs to share new models 30 days before public release, partly because Chinese LLMs already dominate global usage (seven of the top 10 most-used models come from Chinese developers). Anthropic is also expanding in Singapore, though cautiously — it has just four job postings here, far behind OpenAI which opened a Singapore office in 2024.

Nvidia is building a new campus called Constellation in Taipei's technology park, spanning nearly four hectares. CEO Jensen Huang announced the plans at an employee meeting. Nvidia is already hiring heavily in Taipei as its Taiwan supply chain grows. Foxconn and TSMC are already using Nvidia tools for factory digital twins and robotics.

South Korea's Naver raised $1.1 billion through dollar and euro bonds to fund AI and data center expansion. This includes eco-friendly data centers and projects to boost energy efficiency. Orders from investors were nine times the issue size. Naver plans to invest over $680 million in GPUs this year and is exploring AI data center projects in Europe.

Xiaomi has started testing humanoid robots at its electric vehicle factory in China. The robots use AI to load nuts and move materials with about 90% success. CEO Lei Jun wants large-scale deployment within five years. This puts Xiaomi in direct competition with Tesla, Figure AI, and Agility Robotics in the humanoid robot race.

LG CNS, the IT arm of South Korea's LG Group, is opening a new research center in Silicon Valley focused on AI and robotics foundation models. This is LG Group's second R&D hub in the Valley after LG Electronics. The move comes as LG invests in AI startups like Skild AI and partners with companies like Canada's Cohere.

Singapore-based VC firm Iterative has cut nearly half its team (from 16 to 9 employees) and shut down programs like its Scale investment arm and debt funding to focus purely on AI startups. The firm, which has backed 191 companies across two funds worth $65 million, is now zeroing in on "AI-fluent founders" and plans to take startups to San Francisco to connect with US investors. Its latest accelerator cohort accepted just eight startups — one of the smallest since the program launched in 2020 — because few applicants met its new AI standards. Co-founder Brian Ma says Southeast Asia's AI edge lies in applying the technology to logistics, fintech, and education.

Microsoft used its Build 2026 conference to announce it is building its own AI stack, reducing dependence on partners like OpenAI and Anthropic. The company launched seven in-house AI models under the MAI brand — including a flagship reasoning model — and introduced new server processors co-designed with its own Maia AI chips. Microsoft also rolled out an agent platform spanning Windows, Azure, and GitHub, and demonstrated Frontier Tuning, which lets companies customize AI models inside their own compliance boundaries. The strategy positions Microsoft alongside Google and Amazon, which already own their full AI stack from chips to models. Benchmarks and performance claims come from Microsoft's own testing and remain largely unverified by third parties.

LG Group has launched South Korea's first corporate-accredited AI graduate school at its Seoul campus. Employees from LG Electronics, Energy Solution, Innotek, Display, and Chem can earn master's or doctoral degrees with full tuition support. The program draws on LG AI Research and partners with universities like Seoul National University and KAIST. This move lets LG train its own AI talent pipeline rather than competing in the expensive external job market.

Chinese AI firm Moonshot AI, backed by Alibaba, just released a powerful new version of its Kimi chatbot. The K2.5 model can handle text, images, and videos together, and includes an automated coding tool to compete with Anthropic's Claude Code. The company is raising funds and seeking a valuation near US$5 billion.

A startup working on adult-themed AI products went viral after posting job listings for unusual roles. The company reportedly received an overwhelming number of applications, highlighting the unusual ways AI startups are approaching hiring and product development.

Tony Fadell, the inventor of the iPod and co-creator of the iPhone, is warning tech builders not to rely too heavily on AI. He argues that building truly new products requires human storytelling, judgment calls, and moral choices—not just data and team consensus. While AI can speed up early testing, blindly trusting AI to write code creates security risks and weak products. His message: use AI as a helpful tool, but do not surrender your critical thinking to the machine.

Singapore has switched on Aspire 2B, a new national supercomputer with over 1,500 Nvidia H200 GPUs — four times the power of its predecessors combined. That's enough computing muscle to match 120,000 high-end AI laptops. The machine will let Singapore researchers build bigger AI models for weather forecasting, coastal protection, disease prediction, and Asian-language chatbots without sending data overseas. Public researchers from universities, government agencies, and research institutes — more than 9,000 of them — can now access this infrastructure.

Singapore-based Grab has proposed building an open electric vehicle charging network across Hanoi and offering AI tools to help small businesses and households in Vietnam's capital work more efficiently. Grab's president met with Hanoi officials to discuss a formal cooperation agreement, aligning with the city's low-emission zone roadmap. The move would also let Grab leverage its large base of foreign tourists using the app to promote Hanoi to international visitors.

US data center giant Digital Realty is building a massive AI-ready facility in Cyberjaya, Malaysia, with plans for three connected sites totaling 32 megawatts of power. The campus will host cloud providers, enterprises, and network services, linking them to Digital Realty's existing hubs in Singapore and Jakarta. Malaysia's government sees this as a cornerstone of its ambition to become an AI nation by 2030.

Equinix is pouring more than US$190 million into a new data center in Cyberjaya, Malaysia. The KL2 facility is built for AI workloads, supporting high-density GPU setups and liquid cooling systems needed for modern AI computing. The project aligns with Malaysia's push to attract AI infrastructure, though the country has tightened approvals for new data centers outside AI-specific projects. Cyberjaya is already a regional data hub with over 22 facilities, making it a natural spot for this kind of expansion. For users, this means more cloud capacity for AI services across Southeast Asia, though it could also push up local energy costs.

Nvidia CEO Jensen Huang says agentic AI — AI systems that can reason, use tools, and complete tasks on their own — is creating a new computing architecture for PCs, robots, and data centers. He pointed to RTX Spark, a new class of Windows PCs with up to 1 petaflop of AI power and 128GB of memory designed for personal AI agents. This means future computing will split work across local devices, home systems, and cloud instead of relying on one environment. For everyday users, this could mean AI assistants that run more tasks on your own devices without constant internet access.

AI coding tools are enabling non-engineers to build working applications without programming experience. An Anthropic employee mentioned a company lawyer with no coding background successfully built a functional app using AI. This shift toward 'citizen developers' is changing how enterprises build internal tools, potentially reducing dependence on traditional software teams.

PhysicsX, a London-based AI company that builds software to simulate and optimize complex engineering processes in sectors like aerospace, automotive, and semiconductors, has raised $135 million in Series B funding. The round was led by European VC Atomico, with Singapore's Temasek among the participants alongside Siemens, Applied Materials, and General Catalyst. The company, founded by ex-Renault F1 and QuantumBlack alumni, plans to use the funds to expand globally and accelerate AI adoption across heavy industries still relying on legacy engineering tools.

Meta has confirmed that at least 20,225 Instagram accounts were compromised through a security flaw in its AI support chatbot. For nearly seven weeks, the chatbot sent password reset links to random email addresses without verifying ownership — the opposite of what it was supposed to do. Meta had marketed the tool as improving account security.

A growing number of tech companies are abandoning 'tokenmaxxing' — the practice of tracking how many AI tokens employees use and tying it to performance reviews. The CEO of AI chip firm Cerebras called unlimited token budgets 'boneheaded,' and companies like Amazon and Uber are now capping AI spending per employee. The shift reflects growing concern that the practice wastes money and doesn't actually improve work quality.

Global tin demand from AI data centers is expected to triple by 2030, but supply is tightening. Tin is essential for soldering components in servers. Prices have risen over 55% in a year as data center builders compete for raw materials. Most tin comes from volatile mining regions, creating supply chain risks. This could push up costs for AI services and new electronics.

Anthropic has opened its first Asia-Pacific office in Tokyo and signed a memorandum of cooperation with the Japan AI Safety Institute. The US-based AI startup aims to collaborate on AI evaluation methodologies and monitor trends in the field. Anthropic’s CEO Dario Amodei met with Japanese Prime Minister Takaichi and local officials during the visit. The company has also joined the Hiroshima AI Process Friends Group, which seeks to promote safe and trustworthy AI development. Anthropic is working with Japanese firms, including Rakuten, Nomura Research Institute, and Panasonic, to integrate its AI model, Claude, into business operations. Anthropic plans to expand further in the Asia-Pacific region and will bring its approach to Seoul and Bengaluru. _🔗 Source: [Anthropic](https://www.anthropi

Vietnam's FPT Corporation and Nvidia have released a free AI training dataset for Vietnamese language and culture. The dataset contains 900,000 fictional Vietnamese personas based on real national statistics. It aims to help developers build AI systems that better understand local context — from regional dialects to economic conditions. The data is freely available for commercial use on HuggingFace.

South Korean venture firm SBVA invested in AMI Labs, an AI research institute founded by Meta's Yann LeCun, alongside backers including Bezos Expeditions, Nvidia, and Singapore's Temasek. AMI Labs raised a $1.03 billion seed round to build "world models" that understand the physical world differently from current AI. Doosan and Coupang also joined as limited partners, linking Korean manufacturing and logistics capabilities to cutting-edge AI research with a planned dedicated fund.

Apple's annual developer conference WWDC kicks off this week, where it will unveil its latest AI software plans. Expected highlights include a major Siri overhaul powered by Google's Gemini AI, improved image generation tools, and custom emoji creation from photos. This marks Apple's shift from OpenAI's ChatGPT to Google as its primary AI partner. It will also be the last WWDC led by CEO Tim Cook, who retires soon.

India's Tata Consultancy Services (TCS) partnered with investment firm TPG to pour up to $2 billion into HyperVault, a new AI data center business. The plan is to build AI-ready facilities in India exceeding one gigawatt of capacity across locations including Navi Mumbai, Chennai, and Hyderabad. However, analysts note the $6.6 billion budget requirement means heavy borrowing, and key elements like grid allocation and renewable power agreements remain unsigned—raising questions about execution.

Cisco unveiled its Silicon One G300 switch chip, built on TSMC's 3-nanometer process, designed to speed up data transfer in large AI data centers. The chip claims up to 28% faster AI processing and 33% higher network efficiency compared to existing techniques. Cisco is targeting hyperscalers—the massive cloud operators—with $1.3 billion in AI infrastructure orders already booked. The move positions Ethernet as a stronger competitor to Nvidia's InfiniBand for AI clusters.

Israeli AI chipmaker Hailo has laid off about half its workforce, according to reports. The company makes chips designed to speed up AI tasks on edge devices like cameras and sensors. Raspberry Pi uses Hailo chips in its AI add-on products. The cuts suggest even companies making specialized AI hardware are struggling in a crowded market. It signals consolidation pressure in the AI chip space.

# India's TCS, TPG partner to invest $2b in AI data centers [News](https://www.techinasia.com/)Premium * [News](https://www.techinasia.com/) * [Jobs](https://www.techinasia.com/jobs/search) * [Companies](https://www.techinasia.com/companies) * [Events](https://www.techinasia.com/events/live) * [About](https://www.techinasia.com/about) * [advertise](https://www.techinasia.com/advertise?utm_source=techinasia.com&utm_medium=website&utm_campaign=intl-navbar) * [Asia Economic Summit](https://aes.techinasia.com/?utm_source=navbar-tia-homepage&utm_medium=website&utm_campaign=events-flagship-aes) * [Sign Up](https://www.techinasia.com/auth) * [Log In](https://www.techinasia.com/auth?ref=nav_login) [Subscribe](https://www.techinasia.com/subscription) [I'm fundraising](https://www.techinasia.com/com

Singapore-founded PatSnap, which helps companies search patents and avoid infringements, is using generative AI to expand beyond patent departments into R&D teams. The unicorn company—valued after Sequoia China and Vertex Ventures funding—has grown revenue at 20% annually since 2023 and crossed US$100 million in annual recurring revenue in 2024, while also turning profitable. Its AI agents help R&D engineers spot drug targets in life sciences and cross-pollinate ideas across industries. CEO Jeffrey Tiong has pushed 'tokenmaxing'—giving employees US$1,000 monthly in AI tokens to encourage adoption—and has ditched PowerPoint entirely for AI-generated HTML discussions.

Cohesity, a data security company, has joined Anthropic's Project Glasswing program, gaining access to Claude Mythos Preview—one of Anthropic's most advanced frontier models. Cohesity plans to use the AI to automatically find, verify, and fix security weaknesses across its platform. The partnership reflects a growing trend of security firms tapping frontier AI to bolster cyber defenses beyond traditional methods.

Nvidia is betting big that AI-powered PCs will move beyond tech enthusiasts into mainstream households, but analysts question whether regular users actually need or want these features. The chipmaker has been pushing its AI PC vision, though the broader PC market has been sluggish. The success of this push could reshape consumer hardware spending and determine whether AI capabilities become standard in everyday computers.

A £40,000 AI-powered robot named D4RYL has been rejected for membership to Britain's Magic Circle society of magicians. The society's president said the robot was 'extraordinary' but lacked the human spontaneity and empathy needed for real magic. The robot's owner argues D4RYL creates genuine wonder and emotional connection with audiences.

Microsoft has tightened its rules for cloud services in conflict zones after an investigation into how Israel's military used Azure. The report revealed AI-powered target selection and mass surveillance capabilities were deployed in Gaza, raising human rights concerns. Microsoft is now implementing new human rights checks, though key questions about the actual data processed remain unanswered.

A technical blog post exploring emergent behaviors in multi-model AI systems, examining how control and complexity arise when five AI models interact in a simulated economy. Likely covers research on autonomous AI behavior and system-level dynamics.

Prime Minister Lawrence Wong warned that AI is making it easier to spread misinformation and disinformation, and that Singapore will use tools like the Protection from Online Falsehoods and Manipulation Act (POFMA) to crack down when content crosses the line. He said this at a dialogue with the Singapore Press Club, addressing a recent case where authorities blocked 14 online posts targeting the Indian community that likely originated from a China-based platform. Wong also said the Government depends on media outlets to operate responsibly and help hold society together amid an era of AI-generated content.

Digital Realty has opened a major data center campus in Cyberjaya, Malaysia, with three interconnected facilities designed to support AI and cloud computing across Southeast Asia. The expansion brings more regional capacity to handle AI workloads for businesses in the area.

Coinbase CEO Brian Armstrong is using AI tools to cut costs while the cryptocurrency exchange grows its user base. The strategy aims to keep operating expenses relatively flat even as more people use the platform.

Singapore has launched ASPIRE 2B, a powerful new supercomputer with over 1,500 Nvidia H200 GPUs. The system delivers 115 petaFLOPs of computing power, roughly 100 times more than its predecessor. It will help researchers train larger AI models locally, run climate simulations, and support projects in quantum computing and maritime engineering. The system is operated by the National Supercomputing Centre and was launched by Minister Josephine Teo.

Apple has finally unveiled its long-delayed AI overhaul for the iPhone, partnering with Google to power its new 'Siri AI' using Google's Gemini model. The revamped assistant can now track information across apps, find specific photos, and carry out complex tasks using personal context. Many features were already available on Android phones, but Apple is betting it doesn't need to spend hundreds of billions on AI infrastructure to compete. Some AI features will have daily usage limits, and Siri AI won't launch initially in Europe or China due to regulatory issues.

Apple has finally delivered on its 2024 promise to overhaul Siri, unveiling 'Siri AI' at WWDC 2026 powered by Google's Gemini model after a rare year-long delay. The new assistant can find specific photos, addresses, and information across apps using personal context. It will exist as a standalone app with a new expressive voice. Apple paid $250 million to settle a lawsuit over the delayed AI features. CEO Tim Cook's final WWDC marked the close of this chapter before his successor John Ternus takes over in September.

Apple is positioning its AI features as more personal and user-friendly compared to competitors. The pitch emphasizes ease of use and personalization rather than raw capability.