🤖 The Blockbrain Brief #35

AI is shifting toward efficient enterprise agents and natural-language "vibe coding," while local deployment, hardware-agnostic infrastructure, and digital sovereignty redefine the global competitive landscape.

📩 High-level Summary

• Price War for Enterprise Agents Intensifies: With the arrival of OpenAI's ChatGPT Work, Meta’s Muse Spark 1.1, and SpaceXAI’s Grok 4.5, frontier labs are no longer just competing on raw intelligence. The core battlefield has shifted to offering multi-step agentic capabilities, token efficiency, and reliable performance at drastically lower price points.
• Vibe Coding and Open-Source: Massive commercial momentum behind platforms like Lovable ($13.2B valuation talk) and Ollama ($65M Series B) proves that the future of software development relies heavily on natural language. Simultaneously, millions of developers are shifting workloads to local, open-weight environments to control costs and ensure data privacy.
• Cost Pressures lead to Foreign and Diverse Models: Skyrocketing inference costs for top-tier U.S. models are forcing enterprises to adopt practical routing strategies. U.S. companies are increasingly shifting routine tasks to highly competitive, significantly cheaper Chinese models like DeepSeek and Z.ai’s GLM, optimizing for "good enough" performance per dollar.
• Infrastructure Flexibility: Startups like ZML are introducing free, multi-chip inference servers to let enterprises run open-source models seamlessly across Nvidia, AMD, Apple, and Intel silicon. This hardware-agnostic layer helps companies dodge vendor lock-in and optimize cloud spending.
• AI Sovereignty and Regulatory Guardrails: From Ukraine demanding self-hosted, on-premise AI infrastructure to avoid foreign provider control, to European central bankers calling for trading "circuit breakers," the focus is rapidly shifting toward national resilience, safety, and strict governance for autonomous agents.

OpenAI Launches ChatGPT Work as a competitor of Claude Cowork

OpenAI launched ChatGPT Work, a new AI agent designed to help users create documents, presentations, and websites using AI coding capabilities. The tool combines ChatGPT with Codex and is powered by OpenAI’s newly released GPT-5.6 model. OpenAI is positioning ChatGPT Work as a direct competitor to Anthropic’s Claude Cowork and Microsoft’s Copilot Cowork, especially for users who want AI agents that can plan and execute multi-step tasks with minimal human input.

OpenAI emphasized that GPT-5.6 is meant to be cheaper, faster, and more broadly available than rival offerings, with the smaller version reportedly able to complete certain tasks at much lower cost.

Meta Enters the Coding Agent Market With Muse Spark 1.1

Meta launched Muse Spark 1.1, a multimodal AI model built for agentic coding and enterprise workflow automation. The model can handle multi-step reasoning, complex processes, digital workflows, bug fixing, feature deployment, and large code migrations.

While Meta is entering this space later than OpenAI and Anthropic, its low pricing makes it a serious competitor. The model is reportedly priced at $1.25 per million input tokens and $4.25 per million output tokens, placing it near other cost-efficient coding models.

Grok 4.5 Positions Itself as a Lower-Cost “Opus-Class” Workhorse

SpaceXAI released Grok 4.5, describing it as a general-purpose workhorse for coding, app-building, office work, research, writing, and clerical tasks. Elon Musk positioned it as an “Opus-class” model that is faster, more token-efficient, and cheaper than comparable high-end models. TechCrunch notes that SpaceXAI priced Grok 4.5 at $2 per million input tokens and $6 per million output tokens, making it cheaper than Anthropic’s Opus 4.7 and OpenAI’s highest-end Sol model.

The model does not claim outright leadership, but positions itself with strong enough performance at a more enterprise-friendly cost. Grok 4.5 could appeal to businesses trying to reduce AI spend without dropping to a lightweight model.

Ollama Raises $65M as Local Open-Weight AI Goes Mainstream

Ollama, the popular open-source developer tool for running open-weight models locally, raised a $65 million Series B led by Theory Ventures, bringing total funding to $88 million. Founded in 2023 by Jeff Morgan and Michael Chiang, Ollama helps developers run open-weight models on their own computers in minutes. It now has 176,000 GitHub stars, nearly 17,000 forks, over 8.9 million monthly developers, and usage across 85% of the Fortune 500.

Lovable’s Reported $13.2B Valuation Shows Vibe Coding Is Still Booming

Lovable, a Swedish vibe-coding startup, is reportedly in talks to raise $300 million at a $13.2 billion valuation, doubling its previous $6.6 billion valuation. The company, which is less than three years old, reportedly reached $500 million in annualized revenue run rate in June. Its users include founders, designers, and salespeople building websites and e-commerce storefronts, while enterprise customers include Workday, Asana, and Nvidia.

ZML Launches Free Inference Server to Break AI Hardware Monopolies

Paris-based ZML launched ZML/LLMD, a free LLM inference server designed to help open-source models run efficiently across different chips, including Nvidia, AMD, Google TPUs, Apple Metal, and Intel Arc. The goal is to break hardware silos and make it easier for enterprises and cloud providers to use mixed-chip environments. ZML wants to improve inference performance while helping companies reduce costs, increase energy efficiency, and avoid being locked into one hardware vendor.

While ZML/LLMD is not open source, it is launching for free so the company can study adoption before deciding how to monetize. The startup has raised $20 million and has backing from notable AI and developer-tool figures, including Yann LeCun and leaders from Hugging Face.

Ukraine Prioritizes AI Models It Can Run Under Its Own Control

Ukraine is shifting toward AI models it can deploy on its own servers, especially for government, business, and military use. It wants to avoid depending on remote systems that providers or governments could restrict or shut off. The policy favors self-hosted or on-premise models, which may limit the use of remote-only flagship systems from companies like OpenAI and Anthropic.

The decision was reinforced after the U.S. government ordered Anthropic to cut access to powerful models, which Ukraine saw as evidence that AI sovereignty is an operational necessity. Ukraine currently uses Google’s Gemini inside the Diia government app, but removes personal data before sending queries because it does not control the model. It is also developing its own model with Kyivstar based on Google’s open Gemma model, intended for government, private-sector, and military use.

US Enterprises Pivot to Chinese AI Models Amid Rising Costs

US companies are increasingly experimenting with Chinese-built AI models from companies such as DeepSeek, Z.ai, and Alibaba’s Qwen as they search for cheaper alternatives to expensive U.S. frontier models. The article notes that Chinese open-source and open-weight models are often reported to be 60% to 90% cheaper than leading U.S. models, making them attractive for companies trying to manage rising token costs.

Companies are no longer defaulting to the most powerful model for every task. Instead, they are increasingly sending work to the cheapest model that is good enough. Some companies have reportedly moved major workloads from Claude to DeepSeek, while platforms such as OpenRouter, Vercel, and LaunchLemonade are seeing stronger adoption of Chinese models. This trend also creates geopolitical tension, since U.S. proprietary models remain powerful but expensive and potentially subject to access restrictions, while Chinese models are becoming cheaper and more flexible alternatives.

Station F Strengthens Paris as a Launchpad for European AI Startups

Station F, the Paris startup hub founded by Xavier Niel, is preparing the second batch of its F/ai accelerator program. The program is designed to help AI startups move from early product development to real revenue quickly, with a target of reaching €1 million in revenue within six months. They host around 1,000 companies annually and that its Future 40 selection increasingly highlights startups with AI at the center of their business.

The first F/ai cohort was backed by major tech players including AMD, Anthropic, AWS, Google, Hugging Face, Meta, Microsoft, Mistral AI, OpenAI, OVHcloud, Snowflake, and Qualcomm. The second cohort will add partners such as ElevenLabs, Nebius, Rippling, OpenRouter, HubSpot, and GitHub. The first batch already raised $34 million in pre-seed funding, with 80% of startups founded by repeat entrepreneurs and around one-third of founders holding PhDs.

Finto Raises $3.4M to Build Accounting Agents From Munich

Finto, a German startup building AI agents for accounting, raised a £3.4 million seed round from Y Combinator, Gradient, and Lightspeed. The company deliberately chose Munich over Silicon Valley after going through Y Combinator, arguing that European finance teams need European solutions built by people who understand local regulations, enterprise systems, and ERP workflows.

Founded in 2025, Finto builds agents that handle invoice verification, account coding, purchase-order matching, and integrations with SAP, Microsoft Dynamics, and DATEV. Its customers include German football club Arminia Bielefeld and Eat Happy Group. The company’s location strategy is also important: Munich gives Finto access to European industrial customers, TU Munich talent, and nearby enterprise-software infrastructure such as SAP.

European Regulators Warn AI Is Moving Faster Than Financial Rules

European bankers and regulators are warning that AI development is moving faster than financial regulation can adapt. The concern is especially strong around agentic AI, which could affect financial crime, cybersecurity, market integrity, and trading systems. U.K. Financial Conduct Authority CEO Nikhil Rathi said traditional rulemaking is too slow for technology that now evolves in weeks or months rather than years.

Christine Lagarde, president of the European Central Bank, described AI as both a major productivity opportunity and a fast-moving risk, particularly around cybersecurity, hacking, and data theft. Bank of England deputy governor Sarah Breeden warned that agentic AI could increase volatility during market stress if used more widely in trading. She suggested that safeguards similar to circuit breakers or kill switches may eventually be needed if faulty AI systems disrupt markets.

UN Agency Launches Global Initiative to Secure Trust in AI Agents

The International Telecommunication Union, the UN agency for digital technologies, launched a new initiative to improve trust and accountability around AI agents. The ITU is creating a Focus Group to develop frameworks that keep AI agents identifiable, trustworthy, and subject to meaningful human control. The group will focus especially on sensitive areas such as financial transactions and critical infrastructure.

The initiative responds to concerns that autonomous AI agents could impersonate people, make unauthorized decisions, or operate without enough oversight. The Focus Group will include technical, policy, and legal experts, with its first meeting scheduled for Paris in November and a second meeting in Geneva in January.