Large Language Models

Today's Stories

1. 1

   Salesforce launches Agentic Advisor for wealth advisors, but industry experts are raising concerns about costs, benefits, and implementation complexity.

   Salesforce has introduced 'Agentic Advisor,' a new AI tool designed for registered investment advisors (RIAs) to help with their work. The company is positioning this as part of its effort to catch up in the AI space and address a decline in its stock price. Wealth advisors are under pressure to adopt AI tools to remain competitive, so a Salesforce offering tailored to their workflow could be significant. However, experts cited in the article are raising questions about whether the actual benefits will justify the costs and the effort required to integrate the tool into existing systems.

   Industry reaction will determine whether RIAs view Agentic Advisor as a must-have tool or as another costly implementation burden. The concerns flagged by experts—spanning cost, real-world value, and operational friction—suggest the product's success is not assured despite Salesforce's backing.

2. 2

   TrueFoundry acquires MLOps pioneer Seldon AI to combine infrastructure for deploying AI agents at scale, addressing a gap where only 14% of enterprises have moved AI pilots into production.

   TrueFoundry Inc. announced it has acquired Seldon Technologies Ltd., an open-source machine learning platform founded in 2014 that raised around $33 million（約53億円）. The deal brings Seldon Core, a production-grade solution for AI inference and machine learning deployments, under TrueFoundry's control. Enterprise customers including PayPal Inc., Johnson & Johnson Inc., Audi AG and Experian plc rely on Seldon for production ML workloads. TrueFoundry's control plane for agentic AI (self-directed AI systems) combined with Seldon's foundation gives enterprises one place to deploy, observe and govern these systems—directly addressing a market gap where 78% of enterprise customers have AI agent pilots but only 14% have reached production scale.

   The MLOps market is valued at roughly $3 billion（約4800億円） to $6 billion（約9600億円） in 2026, with analysts projecting it to reach $32 billion（約5.1兆円） to $56 billion（約9兆円） by the early 2030s. TrueFoundry recently launched Agent Gateway, a unified control interface for enterprise teams to deploy AI agents with policy enforcement and observability.

3. 3

   Visa partners with AI and stablecoin firms to build payments infrastructure for autonomous agents and digital assets, diversifying revenue beyond traditional card fees.

   Visa announced partnerships with AI developer Alchemy and cross-border payments platform Nuvion, and a stablecoin wallet collaboration with Opera's MiniPay. Alchemy's AgentCard—which enables AI agents to make online purchases on behalf of humans without a checkout screen—recorded 78,000 sign-ups in its first 48 hours. Visa is building what it calls 'value-added services' to reduce reliance on point-of-sale card fees, which face pressure from regulation and fintech competition. Investors view these partnerships as a 'tailwind' supporting long-term revenue growth, as both Visa and Mastercard reported more than 20% year-over-year value-added services growth in their most recent quarter.

   The technology is moving from experimentation to real-world uses, according to Visa's global head of growth products and partnerships. Visa's Trusted Agent Protocol framework partners include Adyen, Ant, Checkout.com, Coinbase, Shopify, Stripe, and Worldpay, positioning agentic commerce as a complementary infrastructure layer for long-term market expansion.

4. 4

   Fujitsu develops PHOTON, a new AI architecture that uses GPU resources up to 475 times more efficiently than the standard Transformer model.

   Fujitsu has developed PHOTON, a new large-language model architecture designed to use GPUs more efficiently than conventional Transformer models, achieving up to 475 times greater efficiency in specific scenarios. GPU resources are a major cost and bottleneck for companies running AI systems; a more efficient architecture could reduce the computational expense of training and operating large language models, making AI deployment more economical for businesses.

   The specific performance gains and real-world deployment timeline for PHOTON remain to be seen; availability and adoption by cloud providers or enterprises will determine whether this efficiency advantage translates into practical cost savings.

5. 5

   SmartDB Launches Trial of "Gemini on Google Distributed Cloud" Provided by KDDI - Enabling Advanced AI Business Applications While Ensuring Domestic Data Sovereignty

   SmartDB Launches Trial of "Gemini on Google Distributed Cloud" Provided by KDDI - Enabling Advanced AI Business Applications While Ensuring Domestic Data Sovereignty

6. 6

   DeepSeek releases R1, a reasoning model that matches top-tier performance at a fraction of the computational cost, potentially reshaping the economics of AI development.

   DeepSeek released R1, an open-weights reasoning model that achieves performance comparable to OpenAI's o1 while using 27% of the FLOPs (computational operations) compared with DeepSeek-V3.2, and the model is now available for use. The breakthrough demonstrates that reasoning capability does not require proportional increases in computational spending, which may shift how companies and researchers approach AI system design and cost efficiency.

   R1 is available as an open-weights model that developers and researchers can download and run on their own systems, removing a traditional barrier to accessing frontier reasoning performance.