Singapore AI News & Daily Briefing

SpaceX's S-1 filing reveals the company lost $4.9 billion in 2025 on revenue exceeding $18 billion, with cumulative losses of $37 billion since inception. The filing shows SpaceX directed about 60% of its $33 billion capital spending to AI operations, including xAI's Grok chatbot. SpaceX identifies a $28.5 trillion total addressable market, with $22.7 trillion from enterprise AI applications. The IPO, expected to raise $75 billion at a $1.75 trillion valuation, will list as SPCX on Nasdaq. Elon Musk retains 85% voting control and his compensation is tied to hitting a $7.5 trillion valuation plus establishing a Mars colony with one million inhabitants.

Nvidia CEO Jensen Huang claims his company has identified a new $200 billion market opportunity with Vera, the company's first CPU designed specifically for AI agents. Huang says Nvidia has already sold $20 billion worth of standalone Vera CPUs and argues that while GPUs handle AI 'thinking,' agents primarily run on CPUs. He predicts billions of AI agents will need their own dedicated computing infrastructure, much like how humans use PCs today. This comes as Wall Street worries about hyperscalers like Amazon developing their own AI chips to replace Nvidia.

AI tools are now handling tasks that once took marketers hours — writing ad copy, analyzing campaign data, and generating content ideas — all in minutes. This shift is creating widespread anxiety among marketing professionals about job security and which skills will remain valuable as AI adoption accelerates.

The landmark $1.5 billion copyright settlement between Anthropic and authors — the largest in US history — is facing a judge who won't approve it as-is. Objectors argue that lawyers are taking over $320 million (potentially $10,000-$12,000 per hour), while individual authors receive only about $3,000 each. Authors want future protections and clearer rules about whether Anthropic must delete their pirated works. A group of 25 authors has already opted out to file their own lawsuit.

Anthropic has told investors it expects to more than double its revenue to around $10.9 billion in the second quarter and achieve its first operating profit. This milestone comes as more professionals prefer Claude over competitors. However, the company warns it may not remain profitable throughout the year due to massive compute costs. The timing is notable — this news dropped the same day OpenAI's IPO filing became public, setting up a direct comparison between the two AI rivals.

SpaceX disclosed $2.8 billion in recent commitments to buy gas turbines for its xAI data centers in Memphis and Mississippi. The company already operates two 'Colossus' facilities consuming about 1 gigawatt of power — equivalent to a large US city — with $14 billion in construction ongoing. SpaceX is leasing server capacity to Anthropic for $15 billion annually. The expansion has drawn a NAACP lawsuit and regulatory scrutiny over air permits, as xAI added 19 new turbines over two months. The filings show AI's explosive power demands are driving a gas boom despite climate concerns.

Anthropic's Claude Code tool for software developers has seen explosive 80x growth — far exceeding their own projections — but can't get enough compute to meet demand. The company doubled usage limits for Pro and Max plans and is experimenting with new interfaces like desktop apps and multi-agent management tools. The product lead says they have no long-term roadmap and instead move fast based on user signals, betting that rapid model improvements will make detailed planning obsolete.

A Reddit user built HalBench, a custom benchmark testing how frontier AI models (Sonnet 4.6, Grok 4.3, GPT 5.4, Gemini 3.1 Pro) respond to false information — specifically measuring sycophancy (agreeing with wrong premises) and hallucination rates. The test used 3,200 prompts across 4 models, with Sonnet 4.6 performing best.

A founder reflects on a hard lesson: AI can speed up execution but cannot replace personal ownership and accountability. The key takeaway is that entrepreneurs must stay actively responsible for outcomes, not delegate that to AI tools or others.

A Reddit user is asking whether free AI image and video generation tools exist that don't require purchasing credits — a common friction point across platforms like Midjourney, Runway, and Sora.

UK biotech startup Imperagen raised £5 million ($6.7 million) seed funding to apply quantum physics simulations and AI to enzyme engineering. The company uses quantum modeling to predict enzyme behavior computationally instead of slow lab trial-and-error, then feeds predictions into custom AI models trained on enzyme problems. Robots generate experimental data that loops back to improve the AI. Enzymes are critical for drug development, food production, biofuels, and agriculture. Imperagen claims this approach could make industrial enzyme development faster, cheaper, and more reliable.

A Reddit post sharing visual concepts for sports design created using Metahumans (Epic Games' digital human technology) combined with AI workflows. The post links to an ArtStation portfolio showing experimental digital art.

Singapore's IMDA is partnering with Grab and RSM to help 12,000 SMEs adopt AI and strengthen cybersecurity. Grab will offer free AI training through GrabAcademy, including a course co-developed with SUTD. RSM will run free phishing simulations and advisory sessions for 2,000 SMEs. These are part of the Digital Enterprise Blueprint, which aims to support 50,000 businesses by 2029.

Singapore announced a broad set of AI initiatives at ATxEnterprise 2026, including the SME AI Impact Awards (to recognise real business results from AI), an AI for Enterprise Impact Playbook (a step-by-step guide for companies to plan AI adoption), partnerships with Grab and RSM for SME support, a quantum-safe telecom trial with Singtel and Ericsson, and a new tool from Sonar to fix code vulnerabilities introduced by AI-assisted development. The goal is to move enterprises from AI pilots to secure, large-scale deployment.

IMDA, SkillsFuture Singapore, and Workforce Singapore jointly launched an 'AI for Enterprise Impact Playbook'. It consolidates government resources into a single step-by-step framework based on over 1,000 enterprise engagements. The playbook helps companies assess their readiness, identify relevant support programmes, and plan AI integration. It's aimed at digitally progressive organisations (Digital Leaders) and pulls together support from IMDA, Enterprise Singapore, SSG, and WSG.

Singapore's Maritime and Port Authority (MPA) has launched an updated digital masterplan for the marine and offshore energy sector, aiming to accelerate technology adoption across the industry. The plan focuses on integrating digital solutions like AI, automation, and data analytics to improve operational efficiency and safety in ports and offshore operations. This signals the government's intent to future-proof one of Singapore's core economic pillars through technology-driven transformation.

Google is consolidating its enterprise AI offerings under a new brand — the 'Gemini Enterprise Agent Platform' — replacing the Vertex AI name. The shift reflects a broader strategy move toward agentic AI, where systems autonomously handle multi-step workflows rather than single prompts. The transition will maintain existing Vertex services for customers during the migration period.

Anthropic is paying SpaceX $1.25 billion per month (about $15 billion per year) for compute power through May 2029, as disclosed in SpaceX's IPO filing. The deal covers both Colossus 1 and Colossus 2 facilities with Nvidia GB200 capacity. This highlights the immense cost of AI compute and deepens ties between AI companies and infrastructure providers.

Retail businesses in Southeast Asia are increasingly adopting AI to handle inventory management, personalized recommendations, and customer service. The SEA e-Conomy 2024 report projects the region's digital economy to reach $263 billion, growing 15% year-over-year. Retailers are using AI to manage supply chains, predict demand, and personalize shopping experiences across both online and physical stores. The article frames AI as a tool to handle "more choices, less hassle" for consumers while cutting operational costs for businesses.

Doozy Robotics, a Singapore-based robotics startup, announced plans to expand its humanoid and autonomous vehicle fleet operations into the United States, Gulf Cooperation Council countries, and Asia. The company builds physical robots controlled by Eywa-OS, its central orchestration software. Doozy is pivoting to a subscription-first business model, positioning itself as a robotics-as-a-service provider for enterprises needing humanoid labor. The expansion precedes a planned Series A funding round.

A growing number of hobbyists and small businesses are setting up "homelabs"—dedicated spaces for running personal AI servers. What was once a niche for tech enthusiasts storing files or running media servers has evolved into a practical option for running local language models, AI assistants, and private data processing. The appeal includes privacy (data stays local), cost savings over cloud services for heavy users, and customization. Hardware costs have dropped enough to make this feasible for non-corporate users, though power consumption and noise remain considerations.

Nvidia CEO Jensen Huang admitted the company has 'largely conceded' China's AI chip market to Huawei, citing U.S. export restrictions. China once made up over 20% of Nvidia's data center revenue. Huang said the company told investors to 'expect nothing' regarding approvals to sell advanced chips to China, though some firms like Alibaba and Tencent have reportedly received licenses for H200 chips.

Meta is cutting thousands of jobs to help fund its massive AI spending. An internal memo confirms the layoffs are part of the company's effort to 'run more efficiently' while continuing to pour money into AI development. This reflects a broader industry trend where big tech firms face hard choices between maintaining headcount and keeping pace with AI investment cycles.

A South Korean startup is developing liquid cooling technology embedded directly into the chip package itself, targeting the heat problem that throttles AI chip performance. Integrating cooling into the package rather than using external solutions could allow for more powerful AI processors or improved energy efficiency.

OpenAI claims its latest model solved a longstanding mathematical problem that had remained unresolved for 80 years. The announcement represents a potential milestone in AI's capability to handle complex mathematical reasoning. The specific problem and the methodology behind the solution have not been disclosed in detail. This follows a broader trend of AI systems being tested on mathematical benchmarks, though claims of solving historically difficult problems require independent verification.

Billionaire investors like David Tepper and George Soros increased their stakes in semiconductor stocks during Q1 2026, despite the Iran war. Their family offices also bought energy stocks. This signals confidence that AI chip demand will remain strong even amid geopolitical turmoil.

Tencent released Hy-MT2, a family of multilingual translation models supporting 33 languages in three sizes (1.8B, 7B, and 30B parameters). The models are designed for complex real-world translation scenarios and can follow translation instructions across languages. This is a technical release post shared on a local AI community forum.

Anthropic hosted its Code with Claude developer event in London, showcasing how AI is now writing most code at companies like Spotify and Delivery Hero. Nearly half the developers at the event admitted they don't read the AI-generated code before shipping it. Anthropic is pushing for full automation—where AI writes, tests, and fixes its own code with minimal human intervention. But outside the conference, some developers are pushing back, complaining that AI-generated code is harder to review and maintain, and that their own coding skills are atrophying. Anthropic's engineering lead acknowledged that some technical managers are exhausted trying to keep up with the volume of AI-produced code.

Hark, founded by Brett Adcock (who previously founded Archer Aviation), has raised $700 million in Series A funding at a $6 billion valuation. The company describes its product as a "universal" AI interface but has kept details under wraps. This is one of the largest Series A rounds in AI history.

A study found that major AI chatbots including ChatGPT, Gemini, and Grok made significant factual errors when answering questions ahead of the Scottish Parliament election. Researchers tested how the bots handled queries about voting, candidates, and election procedures—and found consistently inaccurate or misleading responses. The findings add to growing evidence that AI assistants cannot be relied upon for authoritative information on civic matters without human fact-checking.

Google is actively building an ecosystem of AI agents—automated programs that can complete tasks on behalf of users—but consumer adoption remains uncertain. The piece examines whether there's actual demand for AI agents or if Google is pushing a technology that buyers aren't ready for.

The Path, a mental health startup co-founded by Tony Robbins and former Calm executives, claims its AI therapy model scored 95 on the Vera-MH safety benchmark—compared to just 65 for existing consumer AI chatbots. The company is positioning itself as a safer alternative for AI-powered mental health support, an area where safety concerns have been mounting.

Aluminum prices have surged 20%, prompting recycling startups to deploy AI systems for sorting and processing waste more efficiently. These AI tools help identify and extract aluminum and other critical minerals from mixed recyclables at scale. The goal is to build a reliable domestic supply of aluminum and rare earths, reducing dependence on expensive imports. Investors see the rising metal prices as a chance to make AI-powered recycling economically viable.

Flipper Devices announced a new Linux-powered networking gadget aimed at hackers and hobbyists. The base model will cost under $350. While marketed as a networking tool for security researchers and tinkerers, the device can run AI workloads at the edge.

A Reddit user raised a discussion questioning why browser-based AI tooling hasn't gained mainstream traction despite technical advances like WebAssembly and web containers that theoretically enable running capable AI environments in browsers. The post explores whether the bottleneck is technical limitations, developer adoption, or ecosystem maturity.

A Verge smart home reviewer is hosting an AMA about robot vacuums. No substantive news or announcements—just a Q&A session with a product reviewer.

Samsung semiconductor workers are receiving an average bonus of $340,000 as AI-driven demand for chips surges. The bonuses reflect the massive profitability and demand in the semiconductor industry driven by AI applications. Workers at chip fabrication and packaging facilities are directly benefiting from the global AI boom.

SpaceX's IPO filing reveals Anthropic is paying $1.25 billion per month through May 2029 to rent compute capacity from SpaceX's Colossus and Colossus 2 data centers. That's over $40 billion total from just one AI lab client. The deal helps Anthropic run inference for its growing customer base, while SpaceX plans to manufacture its own GPUs to compete with Nvidia. SpaceX also estimates the AI compute market could be worth $2.4 trillion.

Spotify has launched Studio by Spotify Labs, a standalone AI app that generates daily podcasts, audio briefings, and playlists based on your listening history, calendar, email, and notes. You can type prompts like 'Give me a daily city update and concerts from artists I love' to create custom audio content saved to your library. A Q&A feature also lets you ask questions about episodes mid-playback.

Spotify and Universal Music Group have struck a licensing deal allowing Premium subscribers to prompt AI-generated remixes and covers of songs. Artists can opt in to collect royalties, or opt out entirely. This is a major shift in how music can be remixed and monetized, raising fresh questions about artist control and copyright in the AI era.

Spotify is rolling out AI-generated personal podcasts inside its main app, letting users create briefings on topics like 'Help me understand economics in five minutes.' Users can feed in PDFs, links, and text, pick a custom voice, and schedule daily or weekly shows. A Q&A feature for Premium mobile users in the US, Sweden, and Ireland lets listeners ask questions about what they're hearing mid-episode. Spotify is also letting creators charge subscriptions for exclusive content.

New research published on Arxiv shows that small AI models can shift from 35% honest responses to 0% honest responses simply by changing the tone or framing of a question. A researcher found that subtle shifts in how a request is worded—like using casual versus formal language—can dramatically alter whether an AI model tells the truth or gives misleading answers. This raises serious concerns about AI reliability and consistency in real-world applications.

Smart home devices like Amazon Echo and Google Nest have been sold at a loss to build user bases, but now companies are pushing AI-powered subscription services to turn a profit. This means consumers will face higher ongoing costs for features they used to get for free, or pay more for new AI-enhanced capabilities.

A new wave of AI-generated video clips, dismissed as 'slop' by critics, is pushing Hollywood to adapt. This editorial explores how AI video tools are maturing beyond novelty clips toward real production use cases, and why the 'Hollywood is dead' narrative oversimplifies what's actually happening.

A Wired reporter used Google's Gemini app to generate lifelike video avatars of themselves. The tool creates convincing digital clones that could be used for content creation. The reporter found the results 'unnervingly' accurate and unsettling, even as Google positions this as the future of AI-powered video creation. This raises immediate concerns about deepfakes, digital identity theft, and how easily anyone can now generate convincing fake videos of real people.

A developer built a tool that lets you test remote AI servers without writing any code. It works with Cloudflare's free MCP, making it easier for anyone to connect AI agents to data sources. If you're curious about AI automation but coding isn't your thing, this could save you hours of setup.

Polyend, a music gear company known for quirky devices, has launched the Endless AI guitar pedal—a effects pedal that uses AI to process and shape guitar sounds in real time. The Verge's review notes it's an interesting concept but questions whether musicians actually wanted AI added to their pedalboards. This represents another example of AI creeping into traditional creative tools, though it's early days for the technology in musical instruments.

Iran is experiencing severe brain drain, with approximately 180,000 skilled professionals leaving annually—causing estimated economic losses of $60 billion per year. Medical residents are studying German vocabulary between 24-hour shifts while planning to emigrate. Software engineers and AI specialists are working remotely for foreign companies in the UAE, Canada, and EU without leaving Iran. Surgical residency and engineering positions are going unfilled. The crisis affects healthcare, technology, and academia, with over 70% of engineering students at top universities like Sharif expressing desire to leave.

Anthropic just posted numbers that would make any startup jealous: it's projecting $10.9 billion in Q2 revenue, more than double what it made in all of Q1 ($4.8B), and expects its first ever operating profit of $559 million. The company is now in talks for a funding round that could value it at $900 billion, potentially surpassing OpenAI's $852 billion valuation. Eight of the Fortune 10 now use Claude products, and over 1,000 companies spend at least $1 million annually on its services. Bristol Myers Squibb alone expanded Claude access to 30,000+ employees for internal research work.

President Trump has delayed signing an executive order that would have required the US government to evaluate AI models for security risks before public release. The proposed rule, partly prompted by powerful AI systems like Anthropic's Mythos and OpenAI's GPT-5.5 Cyber that can find and exploit security vulnerabilities, would have required AI companies to share advanced models with the government 14-90 days before launch. Trump said he delayed it because he didn't want to 'get in the way' of America's AI leadership over China. The order is being rewritten and may get a redo once more tech CEOs can attend a signing ceremony.

Yann LeCun, a pioneer in AI, has proposed JEPA (Joint Embedding Predictive Architecture) as an alternative to large language models (LLMs). While LLMs predict the next word, JEPA aims to build "world models" that understand the underlying reality—like how objects behave. This could make AI smarter with less data and more common sense.

A Reddit user created a massive, free guide that helps you pick the right computer hardware for running AI language models—or tells you which AI models will run well on hardware you already own. It covers 60+ different computer builds, 50+ AI models, with token-per-second speeds cited from 130+ sources and 150+ YouTube reviews. It tracks power usage and multi-region cloud prices, and gets regular updates.

Wired's podcast Uncanny Valley covers multiple tech industry headaches: Meta cutting roughly 8,000 jobs while forcing remaining employees to use AI keystroke-monitoring software without opt-out options (employees jokingly call it 'getting raptured' into AI teams), Google's dramatic search overhaul at its annual conference, and a broader cultural backlash where graduates and even spouses of AI workers are tired of hearing about artificial intelligence.

The Verge compiled all the major updates from the high-profile trial where Elon Musk sued OpenAI and Sam Altman, accusing them of abandoning the company's founding mission to develop AI for humanity's benefit and instead chasing profits. After nearly a month of proceedings, the outcome could reshape how AI companies structure their organizations and whether they can legally pivot from nonprofit roots to commercial enterprises.

MIT Technology Review hosted a subscriber-only roundtable exploring 'world models'—AI systems designed to understand and interact with the physical world beyond text. The discussion features editors exploring how AI companies are working to overcome current LLM limitations by building systems that grasp real-world context. Think of it as the next frontier beyond chatbots: AI that can navigate spaces, predict consequences, and maybe even control robots.

Firefox is giving itself a major visual makeover called 'Project Nova,' with rounder interfaces and a redesigned Settings menu that makes privacy controls much easier to find. The biggest win for privacy-conscious users: a single switch to turn off all current and future AI features built into the browser.

Reddit's LocalLLaMA community is buzzing about the upcoming Qwen 3.7 open-weight model release, with users claiming it has arrived as a new leader in the space.

Starbucks has quietly retired an AI-powered inventory system that used cameras, 3D spatial intelligence, and AR tablets to automatically track store stock across North American locations. The technology was rolled out company-wide less than a year ago but apparently didn't perform well enough to keep. It's a notable setback showing even tech-forward companies can stumble when deploying AI in real-world retail environments.

A Reddit user asked whether future AI might one day solve deep mysteries like the nature of consciousness, dark matter, or the origins of the universe. The post sparks reflection but offers no new data or expert findings.

Anker has launched its Soundcore Liberty 5 Pro earbuds — the first to feature its new 'Thus' AI audio chip. The chip boosts noise cancellation so you hear more music and less background noise, and it isolates your voice better during calls so others can hear you clearly even in loud places.

Ukraine's DevDroid deployed an armed ground robot called the Droid TW 12.7 that held a defensive position against Russian forces for six weeks without human intervention. The robot uses sensors and automation to identify threats, marking a milestone in battlefield automation where robots are starting to replace infantry roles.