🤖 The Blockbrain Brief #17

Agentic AI gets skilled, enterprise models battle for supremacy, massive funding fuels infrastructure wars, and regulatory pressure builds on walled gardens.

📩 High-level Summary

• Agentic Shift & Skills: OpenAI and Glean are moving beyond simple chatbots by developing "Skills" for repeatable workflows and interactive research, while Stack AI leverages Cerebras chips for sub-second agent inference.
• Enterprise AI Battle: Anthropic raises a massive $30B to challenge OpenAI’s dominance, while enterprise search firm Glean claims its specialized retrieval outperforms both ChatGPT and Claude in corporate environments.
• Infrastructure & Data Strategy: The race for compute and data intensifies with ByteDance building its own AI chip, Mistral investing in a €1.2B sovereign EU data center, Databricks eyeing an IPO, and Amazon exploring a marketplace for licensed training content.
• Market Impact & Regulation: Fears of AI disruption cause European insurer stocks to slide

ChatGPT Tests Skills & Upgrades Deep Research

OpenAI is actively evolving ChatGPT from a simple chatbot into a more comprehensive agentic workspace. They have introduced a revamped "Deep Research" experience to clarify intent and refine the scope before diving into complex tasks.

Furthermore, reports suggest a new first-party "Skills" layer is in development. This feature would allow users to install and edit specific workflow instructions which are packaged repeatable procedures, constraints, and expected outputs into a "skill."

Glean Launches "Skills" to Package Enterprise Expertise

Glean has launched a new feature supporting the open "Agent Skills" standard. This allows employees to create, reuse, and share specific units of expertise (Skills) across the company. Instead of knowledge staying trapped in an individual’s head or a single chat session, it becomes a standardized asset.

Glean positions these Skills as complementary to AI Agents: Skills encode the "how-to" for specific tasks, while Agents orchestrate the broader end-to-end workflow, deciding when to invoke those skills. This aligns with their vision of mining real work patterns to scale automation without sacrificing control.

Glean Claims Superiority Over OpenAI and Claude in Enterprise Contexts

In a recent blind evaluation of ~280 complex enterprise queries (such as debugging code or analyzing internal data), human graders preferred Glean’s answers significantly more often than those from ChatGPT or Claude. Glean claims their answers were rated "correct" nearly 2x more often than ChatGPT’s.

The key differentiator appears to be retrieval quality. Generalist models like Claude and ChatGPT often over-call tools or struggle to resolve ambiguous internal queries, whereas Glean’s specialized retrieval grounded the answers more effectively in company-specific context.

Stack AI Partners with Cerebras for Instant Inference

Stack AI has partnered with chipmaker Cerebras to supercharge enterprise AI agents. By utilizing Cerebras’ specialized hardware, they are achieving sub-second latency on complex, multi-step workflows where performance levels that standard GPUs reportedly struggle to match without sacrificing quality.

Anthropic Raises $30B in Series G, Valuation Hits $380B

Anthropic has secured a massive $30 billion funding round led by GIC and Coatue, propelling its post-money valuation to $380 billion. Despite having earned its first dollar only three years ago, the company now boasts a run-rate revenue of $14 billion, growing over 10x annually. The capital is earmarked for expanding infrastructure and fueling the "frontier research" that has made them a market leader in enterprise AI and coding.

Databricks Raises $5B at $134B Valuation

Data giant Databricks has raised $5 billion in funding with an additional $2 billion in debt capacity. With annualized revenue exceeding $5.4 billion (up 65% YoY), the company is signaling it is ready to go public "when the time is right." This reinforces the trend that data readiness is the prerequisite for successful AI adoption.

European Insurer Stocks Slide on AI Disruption Fears

European insurance stocks (including AXA and Hiscox) took a hit following a selloff in US insurance brokers. Investors are reacting to fears that new AI tools, specifically recent updates from players like Anthropiccould automate significant portions of the insurance value chain, disrupting traditional business models.

TikTok Parent ByteDance is Building an AI Chip

ByteDance (TikTok) is reportedly developing its own AI chips in partnership with Samsung to reduce reliance on Nvidia. They plan to produce up to 350,000 units of the "SeedChip" this year. This move comes as tech giants increasingly seek to control their own hardware destiny amidst skyrocketing compute costs.

Exclusive: ByteDance developing AI chip, in manufacturing talks with Samsung, sources say | John Buckoke AuDHD

Vertical Integration: The New Moat in the AI Era ByteDance’s reported development of SeedChip, their first in-house AI inference processor—is a textbook example of “Strategic Resilience” in action. While the headlines focus on the tech rivalry, the real story for UK business leaders is about Hardware Sovereignty. In an era where U.S. export controls can shift overnight and global memory supplies (HBM) are tighter than ever, relying solely on third-party silicon is becoming a business continuity risk. Why this matters for the UK Tech & Supply Chain sectors: 1. From Software to Silicon: We are seeing the death of the “asset-light” tech giant. To win in AI, companies are finding they must own the full stack, from the underlying transistors up to the algorithm. 2. The “Inference” Pivot: ByteDance isn’t just trying to out-train OpenAI; they are trying to out-serve them. By focusing on inference chips, they are optimising for the daily operational cost of AI, targeting 100,000 to 350,000 units this year to slash their reliance on the “Nvidia tax.” 3. Supply Chain Diversification: The move to partner with Samsung rather than relying exclusively on TSMC is a masterclass in risk de-concentration. It secures not just manufacturing, but critical access to high-bandwidth memory. Building domestic AI capabilities is no longer a ‘nice to have’ it is a strategic necessity for resilience.” For those of us in the UK, where we have a world-class heritage in chip design (ARM, Graphcore), this raises a vital question: How is your organisation approaching hardware sovereignty? Are you looking at RISC-V to avoid IP lock-in? Is your cloud strategy diversified enough to withstand a sudden “compute-choke”? The race isn’t just about who has the best model anymore, it’s about who controls the sand it runs on. #AI #Semiconductors #TechInnovation #SupplyChain #BusinessStrategy #UKTech #ArtificialIntelligence #ByteDance https://lnkd.in/eFfm-Te9

LinkedInJohn Buckoke AuDHD

Mistral AI x EcoDataCenter Invest €1.2B in Swedish AI Hub

French champion Mistral AI is partnering with EcoDataCenter to build a €1.2 billion AI-focused data center in Sweden. This massive investment highlights the growing importance of sovereign, sustainable compute infrastructure within the European Union.

Amazon May Launch Marketplace for Media Content Licensing

Amazon is exploring a marketplace where publishers can sell their content directly to AI companies for training data. This aims to solve the "messy" copyright landscape by providing a legally safe, standardized way for AI models to access high-quality text and media, similar to Microsoft’s recent initiatives.