Updated weekly · 76 stories
AI news with an
APAC operating lens.
Model releases, regulation, funding, partnerships — each story paired with the AIMenta editorial take on what it means for Asian mid-market operators.
Stories worth your attention
Alibaba releases Qwen 3 with open weights: frontier reasoning for enterprises that cannot use US-hosted models
Alibaba Cloud released Qwen 3, its third-generation large language model family, with open weights for most model sizes including the flagship 235B mixture-of-experts variant. The release includes strong benchmark performance on reasoning tasks and native multilingual support for 7 APAC languages — positioning it as a self-hosted alternative to US frontier models for enterprises with data-residency requirements.
Qwen 3 open weights bring frontier-level reasoning to self-hosted deployments — a credible alternative for enterprises with China data-residency obligations or cost constraints across regulated APAC markets.
EU finalizes GPAI Code of Practice ahead of August deadline
With the August 2026 GPAI obligations approaching, the European Commission published the final Code of Practice for general-purpose AI providers, setting expectations on documentation, copyright due diligence, and systemic-risk assessment.
For APAC providers serving EU users, the Code is the operational benchmark — start your documentation package now, not in July.
Singapore launches AI Verify 2.0 governance framework
IMDA released the second-generation AI Verify framework with expanded coverage for generative AI, agent systems, and post-deployment monitoring requirements.
AI Verify is becoming the de facto ASEAN AI governance baseline. Use it as the starting checklist even outside Singapore.
Anthropic releases Claude with extended reasoning + agent SDK improvements
Anthropic shipped extended-thinking improvements to its Claude model family alongside an updated Claude Agent SDK and new tool-use primitives for production agent deployments.
Production agent reliability now hinges on tool design and eval harnesses, not just model selection. Plan accordingly.
All stories
April 2026 · 30 stories
-
Company via NAVER Cloud
NAVER HyperCLOVA X Expands APAC Enterprise Offering with Korean and Japanese Language AI Models
NAVER expands HyperCLOVA X to target APAC enterprise markets with Korean and Japanese-native LLM, offering an alternative to US providers with in-region data residency. Significant for Korean and Japanese enterprises where English-primary models underperform.
AIMenta take — NAVER expands HyperCLOVA X to target APAC enterprise markets with Korean and Japanese-native LLM, offering an alternative to US providers with in-region data residency. Significant for Korean and Japanese enterprises where English-primary models underperform.
KR -
Company via Grab Tech Blog
Grab Publishes Responsible AI Framework for APAC Deployment — Covering Fairness, Transparency, and Accountability
Grab publishes a responsible AI framework covering fairness, transparency, and accountability for AI systems across Southeast Asia. Signals APAC platform companies building AI governance ahead of regulation — a reference for enterprises deploying consumer-facing AI.
AIMenta take — Grab publishes a responsible AI framework covering fairness, transparency, and accountability for AI systems across Southeast Asia. Signals APAC platform companies building AI governance ahead of regulation — a reference for enterprises deploying consumer-facing AI.
SG -
Company via Sea Group Investor Relations
Sea Group Announces Expanded AI Strategy Across Shopee, SeaMoney, and Garena for APAC Markets
Sea Group announces AI strategy integrating ML across Shopee's recommendations, SeaMoney's credit scoring, and Garena's player matching — placing AI at the centre of its competitive strategy across Southeast Asia's largest consumer internet platform.
AIMenta take — Sea Group announces AI strategy integrating ML across Shopee's recommendations, SeaMoney's credit scoring, and Garena's player matching — placing AI at the centre of its competitive strategy across Southeast Asia's largest consumer internet platform.
SG -
Open source via Hugging Face
Hugging Face Launches APAC Inference Endpoints in Singapore and Tokyo for Open-Source Model Deployment
Hugging Face launches managed inference endpoints in Singapore and Tokyo for open-source model deployment with in-region data residency. Removes infrastructure barriers to Llama, Mistral, and Qwen adoption for APAC teams without dedicated ML engineering capacity.
AIMenta take — Hugging Face launches managed inference endpoints in Singapore and Tokyo for open-source model deployment with in-region data residency. Removes infrastructure barriers to Llama, Mistral, and Qwen adoption for APAC teams without dedicated ML engineering capacity.
-
Open source via Sakana AI
Sakana AI Releases Japanese-Native Open-Source LLM Optimised for APAC Enterprise Deployment
Sakana AI releases Japanese-native open-weights LLM trained on curated Japanese corpora — outperforms English-primary models on Japanese enterprise tasks. Addresses the LLM quality gap blocking adoption at Japanese enterprises with Japanese-language operational workflows.
AIMenta take — Sakana AI releases Japanese-native open-weights LLM trained on curated Japanese corpora — outperforms English-primary models on Japanese enterprise tasks. Addresses the LLM quality gap blocking adoption at Japanese enterprises with Japanese-language operational workflows.
JP -
Open source via Meta AI
Meta Releases Llama 4 with Multimodal Capabilities, Advancing Open-Source LLM Adoption in APAC Enterprise
Meta releases Llama 4 with multimodal capabilities and expanded context. APAC enterprises self-host in-region on AWS/Azure for data sovereignty without proprietary API dependency. Most capable open-weights model at release — significant for APAC cost and customisation.
AIMenta take — Meta releases Llama 4 with multimodal capabilities and expanded context. APAC enterprises self-host in-region on AWS/Azure for data sovereignty without proprietary API dependency. Most capable open-weights model at release — significant for APAC cost and customisation.
-
Security via Cyber Security Agency of Singapore
CISA and Singapore CSA Issue Joint Guidance on Securing AI Systems for Enterprise Deployment
CISA and Singapore CSA publish joint guidance on securing AI systems in enterprise environments — covering model access controls, data pipeline security, and adversarial mitigations. APAC security teams should audit AI infrastructure against this baseline.
AIMenta take — CISA and Singapore CSA publish joint guidance on securing AI systems in enterprise environments — covering model access controls, data pipeline security, and adversarial mitigations. APAC security teams should audit AI infrastructure against this baseline.
SG -
Partnership via Anthropic
Anthropic and AWS Deepen Partnership to Accelerate Claude Enterprise Adoption in APAC
Anthropic and AWS deepen strategic partnership to accelerate Claude adoption across APAC, prioritising Claude on Amazon Bedrock for enterprise customers. Strengthens the case for Claude as default enterprise LLM for APAC companies already running on AWS infrastructure.
AIMenta take — Anthropic and AWS deepen strategic partnership to accelerate Claude adoption across APAC, prioritising Claude on Amazon Bedrock for enterprise customers. Strengthens the case for Claude as default enterprise LLM for APAC companies already running on AWS infrastructure.
-
Partnership via Singtel Newsroom
Singtel and CrowdStrike Expand APAC Cybersecurity Partnership to Deliver AI-Powered MDR Services
Singtel and CrowdStrike expand APAC managed detection and response partnership, bringing AI-powered EDR and threat intelligence to mid-market enterprises via Singtel's regional network. Signals telco-vendor bundling of cybersecurity AI across APAC mid-market.
AIMenta take — Singtel and CrowdStrike expand APAC managed detection and response partnership, bringing AI-powered EDR and threat intelligence to mid-market enterprises via Singtel's regional network. Signals telco-vendor bundling of cybersecurity AI across APAC mid-market.
SG -
Regulation via Monetary Authority of Singapore
MAS confirms AI model risk management guidelines mandatory for Singapore's largest financial institutions by end-2026
The Monetary Authority of Singapore published its formal response to the AI in Finance industry consultation, confirming that AI model risk management guidelines will become mandatory for D-SIBs (Domestic Systemically Important Banks) and major insurers by Q4 2026, with an expectation of industry-wide adoption for all MAS-regulated entities by mid-2027.
AIMenta take — MAS confirms AI model risk management guidelines mandatory for Singapore's largest banks by end-2026, with industry-wide extension by mid-2027 — the de facto ASEAN benchmark for AI governance in financial services.
SG -
Model release via Alibaba Cloud
Alibaba releases Qwen 3 with open weights: frontier reasoning for enterprises that cannot use US-hosted models
Alibaba Cloud released Qwen 3, its third-generation large language model family, with open weights for most model sizes including the flagship 235B mixture-of-experts variant. The release includes strong benchmark performance on reasoning tasks and native multilingual support for 7 APAC languages — positioning it as a self-hosted alternative to US frontier models for enterprises with data-residency requirements.
AIMenta take — Qwen 3 open weights bring frontier-level reasoning to self-hosted deployments — a credible alternative for enterprises with China data-residency obligations or cost constraints across regulated APAC markets.
CN -
Regulation via Ministry of Economy, Trade and Industry (Japan)
Japan METI updates AI governance guidelines: supply chain transparency now required for enterprise procurement
Japan's Ministry of Economy, Trade and Industry updated its AI Governance Guidelines to version 3.0, introducing supply-chain transparency requirements for enterprises procuring AI systems and aligning the framework with G7 Hiroshima AI process principles. The guidelines are advisory rather than mandatory but carry significant regulatory expectation weight.
AIMenta take — Japan's METI updated its AI governance guidelines to align with the G7 Hiroshima AI process, adding supply-chain transparency requirements and clarifying responsible AI obligations for enterprises procuring third-party AI systems.
JP -
Regulation via Ministry of Science and ICT (Korea)
Korea MSIT releases AI Basic Act implementation guidelines with 2027 compliance timeline
South Korea's Ministry of Science and ICT published detailed implementation guidelines for the AI Basic Act, specifying risk classification criteria, compliance obligations for high-impact AI systems, and sector-specific safe-harbour conditions. Enterprises have until Q1 2027 to achieve full compliance.
AIMenta take — Korea's AI Basic Act implementation guidelines clarify high-impact AI obligations and sector-specific risk classifications, setting a 2027 compliance deadline that enterprise AI teams should begin planning for now.
KR -
Regulation via European Commission
EU finalizes GPAI Code of Practice ahead of August deadline
With the August 2026 GPAI obligations approaching, the European Commission published the final Code of Practice for general-purpose AI providers, setting expectations on documentation, copyright due diligence, and systemic-risk assessment.
AIMenta take — For APAC providers serving EU users, the Code is the operational benchmark — start your documentation package now, not in July.
EU -
Open source via Alibaba Cloud
Alibaba Qwen3 Matches GPT-4o on APAC Language Benchmarks — Open-Source Frontier Moment for the Region
Alibaba's Qwen team has released Qwen3, its third-generation open-source large language model family, with benchmark results showing state-of-the-art performance on Chinese, Japanese, and Korean language understanding and reasoning tasks — matching or exceeding GPT-4o on several APAC-language benchmarks. The Qwen3 family spans model sizes from 0.6B to 235B parameters, with the flagship Qwen3-235B-A22B achieving performance competitive with Claude 3.7 Sonnet and GPT-4o on multilingual coding, mathematical reasoning, and instruction following benchmarks.
AIMenta take — Alibaba's Qwen3 achieves state-of-the-art results on Chinese, Japanese, and Korean benchmarks while matching GPT-4o on multilingual reasoning. First open-source APAC-origin model to reach competitive parity with frontier proprietary models across APAC business languages.
CN -
Product launch via DeepL
DeepL Expands Enterprise Offering for APAC Market with Regional Data Residency
DeepL has expanded its enterprise product for APAC customers, introducing data residency options for Japan and South Korea and increasing API throughput capacity for high-volume translation workloads.
AIMenta take — DeepL for Business adds APAC data residency and expands enterprise API capacity. For APAC enterprises translating financial disclosures, legal contracts, and regulated content, DeepL's quality-cost ratio leads for European-Asian language pairs.
APAC -
Regulation via Ministry of Science and ICT, Korea
Korea AI Basic Act Implementation Criteria Published — APAC Enterprises Must Act Before Q3 2026
South Korea's Ministry of Science and ICT has published implementing regulations for the AI Basic Act, confirming high-risk AI classification criteria and mandatory compliance requirements including impact assessments, human oversight, and transparency obligations.
AIMenta take — Korea's AI Basic Act implementation criteria published: high-risk AI in hiring, credit, and customer decisions requires impact assessments before Q3 2026. APAC enterprises with Korea operations must initiate compliance reviews immediately.
KR -
Funding
Chinese foundation-model labs raise combined US$3B+ in Q1 2026
DeepSeek, Zhipu, Moonshot, and MiniMax collectively raised over $3B in the first quarter, signaling continued investor appetite for Chinese sovereign LLM efforts.
AIMenta take — The China-vs-US frontier-model race is now a multi-pole competition; APAC enterprises should evaluate Chinese models alongside US frontier options for cost-sensitive workloads.
CN -
Model release via Anthropic
Claude 3.7 Sonnet Enterprise Adoption Accelerates Across APAC in Q1 2026
Anthropic's Claude 3.7 Sonnet has seen accelerating enterprise adoption across APAC in Q1 2026, with notable uptake in legal technology, financial services, and software development. Extended thinking mode is driving adoption in high-stakes analytical tasks.
AIMenta take — Claude 3.7 Sonnet enterprise adoption accelerates in APAC in Q1 2026, particularly for legal document review, financial analysis, and code generation. Extended thinking mode drives use in complex multi-step reasoning tasks.
APAC -
Research via AI Singapore
Singapore Updates National AI Strategy 2.0 Targets — 15,000 AI Professionals by 2027
AI Singapore has published updated progress metrics confirming 15,000 AI practitioners trained through AIAP and AI4I programmes, 600+ enterprise AI deployment grants awarded, and expanded research infrastructure through the National AI Research Institute.
AIMenta take — AI Singapore updates 2027 targets: 15,000 AI professionals trained, 600+ enterprise deployments co-funded. Singapore reinforces APAC AI hub position — with direct implications for enterprise buyers choosing APAC HQ and CoE locations.
SG -
Regulation via IMDA
Singapore launches AI Verify 2.0 governance framework
IMDA released the second-generation AI Verify framework with expanded coverage for generative AI, agent systems, and post-deployment monitoring requirements.
AIMenta take — AI Verify is becoming the de facto ASEAN AI governance baseline. Use it as the starting checklist even outside Singapore.
SG
March 2026 · 12 stories
-
Model release via Anthropic
Anthropic releases Claude with extended reasoning + agent SDK improvements
Anthropic shipped extended-thinking improvements to its Claude model family alongside an updated Claude Agent SDK and new tool-use primitives for production agent deployments.
AIMenta take — Production agent reliability now hinges on tool design and eval harnesses, not just model selection. Plan accordingly.
-
Regulation via Ministry of Electronics and IT
India Releases Draft AI Bill Proposing Mandatory Registration for High-Risk AI Systems
India's Ministry of Electronics and Information Technology (MeitY) has released a draft Digital India AI Bill for public consultation. The bill proposes mandatory registration for AI systems classified as "high-risk" — defined as AI used in critical infrastructure, healthcare decisions, credit scoring, employment decisions, and education — and establishes algorithmic accountability requirements including impact assessments, bias audits, and explainability standards. If passed in the proposed form, the bill would regulate the largest AI market in South Asia and would require significant compliance investment from APAC technology companies with India-facing AI products and services.
AIMenta take — India releases a draft AI Bill proposing mandatory registration for high-risk AI systems and algorithmic accountability. Would create the largest AI regulatory market in South Asia, directly affecting APAC technology companies with India-facing products.
IN -
Product launch via ServiceNow
ServiceNow AI Agent Orchestrator Launches — Multi-Agent IT and HR Workflows Cut Ticket Resolution by 60%
ServiceNow has launched AI Agent Orchestrator — a platform capability that enables enterprises to design, deploy, and manage multi-agent AI workflows within the Now Platform. AI Agent Orchestrator allows different AI agents (IT operations, HR service delivery, customer operations, risk management) to collaborate on complex workflows that span multiple business domains — for example, an employee onboarding workflow that spans IT provisioning, HR onboarding, and facilities access without human coordination. ServiceNow cites early APAC enterprise deployments reporting 40–60% reduction in mean time to resolve (MTTR) for ITSM tickets and 35% reduction in HR service request handling time when AI agents handle standard requests end-to-end. The platform includes AI-generated workflow templates for 50+ common enterprise use cases.
AIMenta take — ServiceNow launches AI Agent Orchestrator enabling enterprises to deploy and manage multi-agent AI workflows across IT, HR, and customer operations. Now Platform AI updates reduce ITSM ticket resolution time by up to 60% in early APAC deployments.
GLOBAL -
Product launch via OpenAI
OpenAI extends enterprise data residency to Tokyo and Singapore
OpenAI added Japan and Singapore to its enterprise data residency program, enabling APAC customers to keep prompts and outputs within regional infrastructure.
AIMenta take — Removes a common procurement blocker for Japanese and Singaporean banks and government buyers. Expect faster enterprise deal cycles in those markets.
JP -
Partnership via Reuters
Anthropic and AWS Expand APAC Partnership, Bringing Claude to Japan, Singapore, Korea, and Australia via Bedrock
Anthropic and Amazon Web Services have expanded their strategic partnership to accelerate Claude model availability across APAC markets including Japan, Singapore, South Korea, and Australia. The expanded agreement makes Claude 3.7 Sonnet and Claude 3.5 Haiku available through AWS Bedrock in additional APAC regions, with specific data residency configurations designed to meet the regulatory requirements of financial services and healthcare organisations in each market. The partnership includes joint go-to-market programmes targeting APAC financial institutions, insurance companies, and technology companies that require both Anthropic's model capabilities and AWS's existing enterprise infrastructure relationships.
AIMenta take — Anthropic and AWS expand their APAC partnership, bringing Claude via Bedrock to Japan, Singapore, Korea, and Australia. Signals growing enterprise deployment of Anthropic models in APAC regulated industries through AWS infrastructure and regional compliance frameworks.
APAC -
Open source via Mistral AI
Mistral AI Releases Mistral Large 2 as Open Weights — 123B Parameter Frontier Model Available for On-Premises Deployment
Mistral AI has released Mistral Large 2 as an open-weights model under the Mistral Research License, making a 123 billion parameter language model with a 128K context window available for download, fine-tuning, and on-premises deployment. Mistral Large 2 achieves benchmark scores competitive with Claude 3.5 Sonnet and GPT-4o on standard evaluations (MMLU, HumanEval, MATH, and reasoning tasks) — making it the first open-weights model in the frontier performance tier. For APAC enterprises with data sovereignty requirements, on-premises deployment mandates, or API cost constraints at scale, Mistral Large 2 represents a significant opening — frontier AI capability deployable within enterprise-controlled infrastructure without ongoing API charges.
AIMenta take — Mistral AI releases Mistral Large 2 as open-weights, 123B parameters, 128K context, matching Claude 3.5 Sonnet on most benchmarks. First frontier-class open-weights model available for APAC enterprise on-premises deployment without API costs.
GLOBAL -
Model release via Meta AI
Meta releases Llama 4 family with native multimodal support
Meta's Llama 4 family adds native vision and audio understanding alongside reasoning improvements, all under the existing community license.
AIMenta take — Open-weight multimodal capability matching closed-frontier-model quality changes the build-vs-buy calculus for self-hosted enterprise AI.
-
Funding via Vertex Ventures
Vertex Ventures Launches $300M APAC AI Fund — First Pan-Regional VC Bet on APAC AI Startups
Vertex Ventures, the venture capital arm of Temasek (Singapore's sovereign wealth fund), has launched a US$300 million dedicated APAC AI fund — the first pan-regional VC fund specifically targeting AI companies across Singapore, Indonesia, India, South Korea, and Japan. The fund will invest in Seed through Series B AI companies building applications in enterprise AI, healthcare AI, fintech AI, and AI infrastructure for the APAC market. Vertex has stated a preference for APAC-native companies building for regional markets (language, regulatory, and cultural contexts) over US AI companies expanding to APAC. The fund's launch signals that institutional APAC investors are treating AI as a distinct asset class rather than a technology category within broader venture portfolios.
AIMenta take — Vertex Ventures launches $300M dedicated APAC AI fund targeting early-stage AI companies in Singapore, Indonesia, India, South Korea, and Japan. First institutional APAC-focused AI VC fund signals the region's AI startup ecosystem maturing beyond individual national pockets.
SG -
Partnership via Amazon Web Services
AWS and NVIDIA Expand APAC Partnership — Regional AI Compute Access and SageMaker-NIM Integration
Amazon Web Services and NVIDIA have announced an expanded APAC partnership that will integrate NVIDIA NIM (model inference microservices) directly with Amazon SageMaker, enabling APAC enterprises to deploy NVIDIA-optimised AI models on AWS infrastructure without managing GPU clusters manually. The partnership includes commitment to additional NVIDIA H100 and Blackwell GPU availability in AWS APAC regions (Sydney, Tokyo, Singapore, Mumbai), addressing the GPU compute bottleneck that has constrained APAC enterprise AI deployment. The integration will allow AWS customers to deploy NVIDIA's foundation model catalogue — including models from Mistral, Meta, and Google — through SageMaker with one-click deployment.
AIMenta take — AWS and NVIDIA announce expanded APAC partnership to deliver GPU compute and AI model access through AWS for APAC enterprise customers. Creates a regional AI infrastructure pathway for APAC enterprises unable to access US hyperscaler capacity directly.
APAC -
Company via Nikkei Asia
ByteDance Establishes Dedicated APAC Enterprise AI Division, Targets B2B Market with Doubao and Coze
ByteDance has announced the formation of a dedicated APAC Enterprise AI division, consolidating its business-facing AI products — Doubao (enterprise LLM), Coze (AI agent platform), and Lark (enterprise communication suite with AI features) — under a unified enterprise go-to-market organisation targeting Singapore, Japan, South Korea, and Australia. The division will operate separately from ByteDance's consumer business (TikTok, Douyin) and is designed to address enterprise concerns about consumer platform associations in regulated APAC markets.
AIMenta take — ByteDance establishes a dedicated APAC enterprise AI division targeting corporate customers in Singapore, Japan, Korea, and Australia with Doubao, Coze, and Lark suite AI. Signals ByteDance's pivot from consumer to B2B AI revenue across international markets.
APAC -
Model release via Google DeepMind
Google launches Gemini 3 Pro with 2M token context
Google DeepMind shipped Gemini 3 Pro to Vertex AI with a 2-million-token context window and native code-interpreter integration.
AIMenta take — Long-context advantage matters most for legal, contract, and codebase use cases. Test with your actual workloads — quality varies materially across context positions.
-
Model release via DeepSeek
DeepSeek releases R2 reasoning model with open weights
DeepSeek's R2 reasoning model matches frontier closed models on math and code benchmarks at a fraction of the inference cost, with weights released under the MIT license.
AIMenta take — The cost-quality frontier has moved again. Re-run your inference economics if you priced workloads more than 6 months ago.
CN
February 2026 · 11 stories
-
Research via Nanyang Technological University
NTU Singapore Publishes Rapid Robot Learning Breakthrough — Manipulation Tasks Learned in 30 Minutes vs Weeks
Researchers at Nanyang Technological University (NTU) Singapore's College of Computing and Data Science have published a new imitation learning framework that enables robotic manipulators to learn complex assembly and manipulation tasks from 30 minutes of human demonstration — a 20× reduction in learning time versus comparable prior methods. The research, published in Nature Machine Intelligence, uses a hierarchical skill decomposition approach that breaks complex manipulation tasks into reusable primitive skills — allowing the robot to learn new task combinations from minimal human demonstration once the primitive skill library is built. The research was conducted in collaboration with Singapore's Agency for Science, Technology and Research (A*STAR) and has attracted international attention as an example of APAC-originated AI robotics research.
AIMenta take — Nanyang Technological University researchers publish AI-robotics breakthrough enabling robots to learn manipulation tasks from 30 minutes of human demonstration versus weeks of prior methods. Signals Singapore's emergence as a tier-1 APAC robotics AI research centre.
SG -
Research via AI Singapore
AI Singapore Releases SEA-HELM: First Systematic LLM Benchmark for Southeast Asian Languages
AI Singapore has published SEA-HELM (Southeast Asian Holistic Evaluation of Language Models) — a comprehensive benchmark evaluating 20 large language models across Thai, Vietnamese, Bahasa Indonesia, Malay, Filipino, and Tamil. The benchmark reveals consistent 15–25% performance gaps between models' English-language and Southeast Asian-language capabilities, and identifies which commercially available and open-source models perform best on each regional language — providing APAC enterprises with the first systematic regional evidence base for model selection decisions.
AIMenta take — AI Singapore publishes SEA-HELM: a systematic evaluation of 20 LLMs across Thai, Vietnamese, Bahasa Indonesia, and Filipino. Results show 15-25% performance gaps vs English benchmarks, giving APAC enterprises the first regional evidence base for model selection.
SG -
Regulation via MSIT Korea
Korea AI Basic Act takes effect — high-impact AI obligations now binding
Korea's AI Framework Act came into force on January 22, with enforcement guidance from MSIT covering high-impact AI classification, generative AI labeling, and foreign-provider obligations.
AIMenta take — Foreign AI providers serving Korean users need a domestic representative. Confirm your contractual chain now if you operate in Korea.
KR -
Funding via Sarvam AI
Sarvam AI Raises Series B to Scale Multilingual AI Across 10 Indian Languages
Sarvam AI, the Bangalore-based AI company building large language models for Indian languages, has raised a Series B funding round to scale its multilingual AI platform across Hindi, Tamil, Telugu, Kannada, Malayalam, Bengali, Marathi, Gujarati, Odia, and Punjabi. Sarvam's models combine speech recognition, text translation, and conversational AI for each language — enabling AI-powered services for the 1.2 billion Indians who are not proficient in English. The funding will accelerate integration of Sarvam's API with Indian enterprise customers in financial services (voice-based banking for rural India), healthcare (vernacular patient communication), and government (citizen services in local languages). Sarvam represents a new generation of Indian AI companies building foundation models for India rather than fine-tuning US models for Indian use cases.
AIMenta take — Sarvam AI raises Series B to scale multilingual AI covering 10 Indian languages. Addresses the AI gap for India's 1.4B population where English-only AI fails most users. Signals India's maturation as an AI model developer alongside services market.
IN -
Security via Cyber Security Agency Singapore
Singapore CSA Issues AI Security Framework — Red-Teaming Now Required for Critical Infrastructure AI
Singapore's Cyber Security Agency (CSA) has published its Artificial Intelligence Security Guidelines, establishing formal security requirements for AI systems deployed in Singapore's Critical Information Infrastructure (CII) sectors — finance, healthcare, energy, water, transport, and government. The guidelines require CII owners to conduct adversarial robustness testing (red-teaming) against AI systems before production deployment and to implement ongoing adversarial monitoring in production. This makes Singapore the first APAC nation to issue specific AI security standards as an operational compliance requirement, moving beyond the governance and ethics frameworks issued by MAS and PDPC.
AIMenta take — Singapore CSA issues AI security framework requiring adversarial testing before production deployment. First APAC national AI security standard making red-teaming a compliance requirement for critical information infrastructure owners.
SG -
Regulation via HKMA
HKMA publishes generative AI principles for authorized institutions
The Hong Kong Monetary Authority issued principles-based guidance on responsible use of generative AI by authorized institutions, covering governance, model risk, and customer impact.
AIMenta take — Hong Kong banks now have a clear regulatory anchor for AI deployment. The next 6 months will see the procurement floodgates open for compliant vendors.
HK -
Funding via DealStreetAsia
Singapore AI Healthtech Holmusk Raises $45M Series C for APAC Mental Health Data Platform
Holmusk, a Singapore-based digital health company specialising in mental health data science and AI, has raised a $45M Series C led by MSD (Merck Sharp & Dohme) with participation from Temasek Holdings and B Capital. The company operates NeuroBlu, a real-world evidence platform for mental health research built on data from psychiatric health systems across the US and APAC. The Series C will fund expansion of NeuroBlu into Japan, South Korea, and Australia, where mental health data infrastructure is underdeveloped relative to the scale of the treatment need.
AIMenta take — Singapore-based AI healthtech Holmusk raises $45M Series C for its mental health data platform serving APAC health systems. The round expands Holmusk's real-world evidence and AI-driven mental health analytics to Japan, South Korea, and Australia.
SG -
Model release via Mistral AI
Mistral Large 3 ships with sovereign deployment options for EU enterprise
Mistral's Large 3 release pairs frontier capability with on-premise and EU-sovereign cloud deployment options.
AIMenta take — European-sovereignty-focused buyers now have a credible non-US frontier-model option. Expect more aggressive procurement comparisons.
-
Partnership via Microsoft
Microsoft and OpenAI Commit $3B to APAC AI Infrastructure — Japan, Australia, and Singapore Data Centres Announced
Microsoft has announced a US$3 billion expansion of its APAC AI infrastructure in partnership with OpenAI, committing to new Azure OpenAI Service data centres in Japan (Tokyo), Australia (Sydney), and Singapore. The investment represents the largest single Microsoft infrastructure commitment in APAC history and will provide enterprise customers in the region with direct access to OpenAI models — GPT-4o, o1, and next-generation models — from in-region infrastructure, addressing data residency concerns that have slowed enterprise AI adoption. The announcement includes a commitment to train one million APAC developers and IT professionals on Azure AI services by end of 2027.
AIMenta take — Microsoft and OpenAI deepen APAC partnership with $3B regional infrastructure investment — new Azure OpenAI data centres in Japan, Australia, and Singapore. Signals accelerating enterprise AI demand across the region as hyperscalers compete for APAC AI workloads.
APAC -
Product launch via Microsoft
Microsoft Copilot Studio Expands APAC with Japanese, Korean, and Simplified Chinese AI Agent Builder
Microsoft has expanded Copilot Studio — its low-code platform for building custom AI agents and Copilot extensions — with full language support for Japanese, Korean, and Simplified Chinese, along with data residency commitments for Japan, South Korea, and China markets. Copilot Studio enables Microsoft 365 enterprises to build custom AI agents that connect to internal data sources, automate workflows, and integrate with Microsoft Teams, SharePoint, and Dynamics 365 — without requiring ML engineering capability. The APAC expansion with native language support is significant for Japan and South Korea, where enterprise AI adoption has been constrained by limited Japanese and Korean language quality in enterprise AI tools.
AIMenta take — Microsoft Copilot Studio expands APAC availability with new low-code AI agent builder in Japanese, Korean, and simplified Chinese. Enables APAC enterprises to build custom AI agents without ML expertise. Major unlock for large enterprise Microsoft shops across the region.
APAC -
APAC via Kominfo
Indonesia unveils national AI strategy with sovereign LLM mandate
Indonesia's government launched a national AI strategy emphasizing Bahasa Indonesia foundation models, public-sector AI adoption, and skills development.
AIMenta take — Indonesian-language AI is structurally underserved. Multinational vendors should plan localized model development if Indonesia is in their growth plan.
ID
January 2026 · 9 stories
-
APAC via ASEAN Secretariat
ASEAN Releases AI Governance Compendium 2026 — Nine Nations Align on Responsible AI Principles
The Association of Southeast Asian Nations (ASEAN) has released the ASEAN AI Governance Compendium 2026 — the first multilateral AI governance framework endorsed by all nine active ASEAN member states (Singapore, Malaysia, Indonesia, Thailand, Vietnam, Philippines, Brunei, Myanmar, Cambodia, with Laos as observer). The Compendium establishes shared principles for responsible AI deployment in ASEAN, covering algorithmic transparency, data governance, human oversight requirements, and sector-specific guidance for financial services, healthcare, and government. The framework is non-binding but creates a common baseline for AI governance discussions across the region.
AIMenta take — ASEAN AI Governance Compendium 2026 aligns nine Southeast Asian nations on shared principles for responsible AI in government and regulated industries. The first multilateral ASEAN AI governance document signals the region's maturity beyond individual national AI strategies.
APAC -
Funding via Upstage
Korean AI Startup Upstage Raises Series C to Scale Solar LLM Across APAC Enterprise Markets
Upstage, the South Korean AI company behind the Solar LLM family, has raised a Series C funding round to accelerate APAC enterprise expansion. Solar Pro and Solar Mini are instruction-tuned LLMs optimised for document processing, code generation, and enterprise workflow tasks — benchmarking competitively with GPT-4o class models on structured enterprise tasks while being deployable on-premises in APAC data centres. Upstage has established enterprise customers in South Korea across financial services, telecommunications, and healthcare, and is expanding into Japan, Singapore, and Southeast Asia with localised model variants and enterprise licensing. The Series C funding will accelerate the APAC go-to-market and fund development of additional language models for Japanese and Southeast Asian languages.
AIMenta take — South Korean AI startup Upstage raises Series C to expand its Solar LLM platform into APAC enterprise markets. Positions Korea as an APAC AI model developer capable of competing with US foundation models for regional enterprise deployment.
KR -
Company via TSMC
TSMC begins volume production of 2nm process for AI accelerators
TSMC's N2 node is now in volume production, with NVIDIA, AMD, and Apple confirmed as initial customers for next-generation AI silicon.
AIMenta take — A new compute generation typically lands in cloud GPUs 9-12 months after silicon ramp. Plan capacity refresh windows accordingly.
TW -
Product launch via Salesforce
Salesforce Agentforce Enters General Availability in APAC — Pre-Built Agents for Financial Services, Retail, and Manufacturing
Salesforce has announced general availability of Agentforce in APAC markets, with the release of pre-built AI agents for financial services (wealth management and insurance), retail, and manufacturing sectors. Agentforce is Salesforce's autonomous AI agent platform — distinct from Einstein Copilot (AI assistant) in that Agentforce agents operate independently to complete multi-step tasks without ongoing human direction. The APAC GA release includes agents that handle customer onboarding workflows, order and returns management, and field service scheduling — tasks that currently consume significant human customer service agent time in APAC Salesforce deployments. Salesforce cited early APAC pilots with 40–60% reduction in handling time for automated task categories.
AIMenta take — Salesforce Agentforce launches APAC-specific AI agents for financial services, retail, and manufacturing. Pre-built autonomous agents for customer service and sales workflows enter general availability. Significant for APAC Salesforce customers evaluating AI automation.
APAC -
Funding via TechCrunch
Data Integration Platform Airbyte Raises $100M Series C, Plans APAC Expansion for Enterprise AI Data Pipelines
Airbyte, the open-source data integration platform, has closed a $100M Series C funding round with participation from Andreessen Horowitz, Sequoia Capital, and several Singapore-based growth investors. The company plans to use the funding to expand its enterprise sales presence across APAC — specifically targeting Japan, Singapore, and Southeast Asian markets where enterprise AI adoption is driving demand for automated data pipeline infrastructure that can feed AI models and analytics platforms.
AIMenta take — APAC AI startup Airbyte raises $100M Series C to expand data pipeline automation across Southeast Asia and Japan. The round signals continued investor confidence in APAC data infrastructure companies enabling the AI adoption wave among mid-market enterprises.
APAC -
Company via Nikkei
SoftBank breaks ground on US$15B Stargate Japan data center campus
SoftBank, OpenAI, and Oracle began construction of a major AI training and inference data center campus in Hokkaido as part of the global Stargate buildout.
AIMenta take — Domestic Japanese AI inference capacity is expanding fast. For Japan-resident workloads, capacity ceiling concerns ease through 2027.
JP -
Research via Google DeepMind
Google DeepMind's AlphaProteo Designs Novel Proteins with 88% Success Rate — AI Drug Discovery Enters a New Phase
Google DeepMind has published research on AlphaProteo, an AI system that designs novel protein binders — molecules that bind to specific target proteins with high specificity. In testing against seven different target proteins, AlphaProteo achieved an 88% success rate in designing binders that work as intended, compared to 11–20% success rates for leading prior methods. AlphaProteo represents a progression from AlphaFold (which predicts the 3D structure of existing proteins) to AI that designs new proteins with specific desired properties — a capability with direct implications for drug discovery, therapeutic development, and vaccine design. For APAC pharmaceutical and biotechnology companies, the research signals that AI-designed therapeutics may enter clinical development pipelines within the 2026–2030 timeframe.
AIMenta take — Google DeepMind's AlphaProteo generates novel protein binders with 88% success rate, outperforming all prior methods on seven targets. Marks a shift from protein structure prediction (AlphaFold) to AI-driven protein design with direct pharmaceutical drug discovery applications.
GLOBAL -
Open source via Anthropic
Model Context Protocol 1.1 spec released with multi-server orchestration
The MCP 1.1 specification adds first-class support for multi-server orchestration, capability discovery, and standardized authentication flows.
AIMenta take — MCP is becoming the lingua franca for agent tool integration. Treating MCP server compatibility as a procurement criterion now is reasonable.
-
Model release via ByteDance
ByteDance ships Doubao 1.5 Pro with aggressive pricing for enterprise
ByteDance's Doubao 1.5 Pro launched with pricing roughly 60% below comparable Chinese competitors, intensifying the domestic price war.
AIMenta take — For non-China-resident workloads the implications are limited; for in-China deployments, this is a meaningful cost lever worth re-evaluating.
CN
December 2025 · 5 stories
-
Regulation via European Commission
Reminder: EU AI Act prohibited practices took effect Feb 2 2025
The first major compliance milestone of the EU AI Act — the ban on prohibited practices like social scoring and untargeted facial-image scraping — has been in force since February 2025.
AIMenta take — Most enterprise AI is unaffected by the prohibited-practices article, but biometric and recommendation systems should re-confirm scope.
EU -
Funding via Anthropic
Anthropic closes major late-stage funding round at elevated valuation
Anthropic raised additional growth capital to fund expanded compute capacity and global enterprise GTM, with strategic participation from cloud and APAC investors.
AIMenta take — Frontier model capacity is no longer the binding constraint on Claude availability. Expect aggressive APAC enterprise expansion through 2026.
-
Model release via OpenAI
OpenAI launches GPT-5 with unified reasoning and tool-use surface
OpenAI shipped GPT-5 with a single API surface for reasoning, vision, code interpretation, and tool use, simplifying the previously fragmented model lineup.
AIMenta take — The unified surface reduces integration complexity but adds router-cost considerations. Production teams should monitor latency and cost per task type.
-
APAC via AI Singapore
AI Singapore launches third national AI talent and startup fund
AISG's third fund cycle commits $200M+ to talent development, applied AI research, and ASEAN AI startup acceleration.
AIMenta take — Singapore continues to lead ASEAN on AI ecosystem investment. Worth monitoring for partnership and acquisition deal flow.
SG -
APAC
India launches BharatGen sovereign multilingual LLM initiative
Indian government and academic consortium launched BharatGen, a multilingual foundation model effort covering 22 official Indian languages.
AIMenta take — India is the largest underserved language AI market globally. Even partial Bharat-language coverage materially expands addressable market for vendors.
IN
November 2025 · 3 stories
-
Product launch via AWS
AWS unveils Trainium3 at re:Invent with 4x training throughput
AWS announced Trainium3 with major performance improvements vs Trainium2, alongside expanded Bedrock model coverage and Q product roadmap.
AIMenta take — Trainium economics increasingly compelling for sustained training workloads. Worth a comparison if you spend >$5M/year on training.
-
Regulation via BIS
US updates AI chip export controls; China responds with rare-earth measures
New US BIS export control updates further restrict frontier AI chips to China; China responded with expanded rare-earth export controls affecting global semiconductor supply.
AIMenta take — Compute supply chains remain a geopolitical risk. APAC enterprises should consider multi-vendor strategies and document end-use cases carefully.
CN -
Product launch via Salesforce
Salesforce ships Agentforce 2.0 with multi-agent orchestration
Agentforce 2.0 adds multi-agent orchestration, an expanded action library, and pricing changes targeting per-conversation rather than per-action billing.
AIMenta take — CRM-anchored agent deployments are maturing. For Salesforce-standardized enterprises, Agentforce should be in the build-vs-buy bake-off.
October 2025 · 3 stories
-
Product launch via Microsoft
Microsoft opens Copilot Studio agent marketplace to third parties
Microsoft opened the Copilot Studio agent marketplace to third-party publishers, with revenue sharing and enterprise admin controls.
AIMenta take — M365-anchored enterprises gain a procurement-friendly path to verified agents. Expect rapid catalog expansion through 2026.
-
Research via Sakana AI
Sakana AI publishes evolutionary model-merging research with production results
Tokyo-based Sakana AI published research demonstrating evolutionary model merging applied to Japanese-language model improvements, with measurable production gains.
AIMenta take — For Japanese-language workloads, model merging is now a credible technique alongside fine-tuning. Sakana's open methods are reusable.
JP -
APAC via McKinsey
McKinsey APAC banking GenAI survey: 70% of APAC banks now have production GenAI
McKinsey's 2025 APAC banking GenAI survey shows 70% of regional banks have at least one production GenAI use case, with average ROI of 1.4x in year one.
AIMenta take — APAC banking has crossed the production threshold. The competitive question has shifted from "do we deploy GenAI" to "how fast do we scale."
September 2025 · 2 stories
-
Product launch via Google
Google ships Deep Research as standalone Gemini agent
Google's Deep Research feature graduated to a standalone agent capable of multi-hour research tasks with structured output and citation tracking.
AIMenta take — Long-running research agents are now production-ready. Knowledge-worker workflows in legal, consulting, and finance gain a credible automation path.
-
Security via OWASP
OWASP catalogs first wave of real-world indirect prompt injection incidents
The OWASP GenAI Security Project published its 2025 incident retrospective covering documented indirect prompt injection attacks against deployed agent systems.
AIMenta take — Indirect prompt injection is no longer theoretical. Any agent with email, web-browsing, or third-party data access needs explicit threat modeling.
Want this in your inbox?
We publish a weekly APAC AI digest — the few stories that actually matter for enterprise leaders, with our editorial take.