The latest technology news from across Asia-Pacific
Hyperscalers are locking in nuclear and gas deals as AI's always-on power demands dwarf cryptocurrency mining's flexible load. The shift is real: Bitcoin miners are pivoting facilities to AI hosting, with mining revenue projected to drop from 85% to under 20% by late 2026. Energy availability, not compute capacity, now determines who can scale.
AI tools ship code faster but stumble on performance-critical paths like N+1 queries and over-allocation. Enterprise leaders report the real constraint isn't compute or latency—it's engineers who can design sustainable systems. The gap between coders who use AI and those who think in first principles is widening.
LangChain's summarization middleware automatically compresses conversation history when hitting token limits, using cheaper models to keep agents scalable. Worth watching for long-running enterprise agents, though analytical tasks may lose numerical precision in the compression.
Short-form video app UpScrolled says it added 2.35 million users in three weeks, capitalising on TikTok's US ownership turmoil and late-January outages. Founder Issam Hijazi announced the milestone at Web Summit Qatar, though the claim relies on his statements without independent verification. The growth is a fraction of TikTok's 1.6 billion global users.
Britain's statutory Fuel Finder scheme launched Monday, requiring every petrol station to submit price changes to a central API within half an hour. The Competition and Markets Authority's open data play mirrors TfL's transport API - which powers 600+ apps and saves Londoners £130M annually. Claims of £40 household savings depend on whether drivers actually shop around.
Layer 2 networks compress hundreds of transactions into single Ethereum settlements, splitting L1 costs across users. The 2024 EIP-4844 blob storage update dropped fees to $0.01 by treating data as temporary, not permanent. Trade-off: most L2s run centralized sequencers.
Birmingham startup Linq closed a $20M Series A from TQ Ventures to expand its API that lets AI assistants send blue-bubble iMessages instead of gray business texts. ARR doubled in eight months after launch, but the real question is whether native messaging solves a customer problem or just a vendor aesthetic issue.
Research code hardwired to specific hardware becomes untestable when the equipment is booked, broken, or belongs to another lab. The Dependency Inversion Principle—abstracting hardware dependencies through interfaces—lets scientists test algorithms without physical devices. It's the same pattern enterprise teams use to build resilient systems.
An enterprise developer's reflection on switching from C++ to Rust reveals how compile-time safety checking shifts cognitive load from avoiding bugs to solving problems. With Microsoft and Google attributing 70% of security vulnerabilities to memory safety issues, the mental model matters as much as the tooling.
For 45 years, programming offered neurodivergent coders something rare: a literal, unambiguous language where instructions meant exactly what they said. LLMs and AI coding assistants are changing that fundamental contract—replacing deterministic logic with probabilistic ambiguity. The implications go beyond individual developers.
Elon Musk is moving to combine SpaceX with his AI company xAI, according to Bloomberg sources. The merger would give xAI access to SpaceX's capital and computing resources as it burns through cash building foundational models—while raising questions about conflicts of interest and governance.
Microsoft's guidance is clear: star schemas outperform snowflake schemas in Power BI semantic models. For CTOs managing enterprise data warehouses, the choice between normalized and denormalized dimension tables directly impacts query speed and DAX complexity. The trade-off isn't subtle.