Happy Thursday! Last week Anthropic launched an AI workbench for drug discovery and revealed Claude's reasoning mirrors human consciousness theory, while Microsoft cut ties with OpenAI and Anthropic models to reduce costs and launched a $2.5B AI consulting firm.

GENERATIVE AI CONSCIOUSNESS RESEARCH

Claude's Hidden Mind Mirrors Human Consciousness Theory

Anthropic published research revealing Claude AI has spontaneously developed an internal structure resembling a leading theory of human consciousness. The 16-author study describes a "J-space" — a small privileged zone where the model holds concepts it can report on and reason with, surrounded by vast automatic processing it cannot access or articulate. The finding is already reshaping how Anthropic monitors its AI systems for safety risks, arriving amid growing scientific debate over whether machines can possess anything resembling a mind. Researchers used a new mathematical technique called a "J-lens" to peer inside Claude's neural network and uncover the structure.

NEW GENERATIVE AI LAUNCH

Claude Cowork Goes Mobile as Most Users Skip Coding

Anthropic has launched Claude Cowork on mobile and web, expanding the tool beyond its desktop-only origins. The rollout begins in beta for Max subscribers, allowing tasks to start on one device and continue autonomously in the background, even after the app is closed. Alongside the launch, Anthropic released usage data from 1.2 million anonymized sessions across 600,000 organizations, revealing that most Claude Cowork users are knowledge workers rather than developers — a finding that signals a clear strategic shift toward the broader professional market.

GENERATIVE AI DRUG DEVELOPMENT

Anthropic Launches AI Workbench for Drug Discovery

Anthropic unveiled Claude Science at its AI for Science event, a new platform designed to unify fragmented scientific tools and datasets into a single environment while generating figures and visuals for researchers. The company, already a leader in coding tools and AI models, positioned the launch around AI's potential to accelerate scientific discovery and healthcare development. Anthropic highlighted a growing list of biotech and pharma customers already using Claude.

NEW AI PLATFORM LAUNCH

Pinecone Nexus Gives AI Agents Enterprise Knowledge

Pinecone Systems has launched the public preview of Pinecone Nexus, a platform designed to curate and distribute enterprise knowledge for AI agents. The system connects to sources like Box and Microsoft OneLake, organizing data into structured knowledge sets so agents can reason across documents without relying on fragmented retrieval methods. In benchmarks, Nexus achieved 95% accuracy on complex support questions and outperformed standard retrieval-augmented generation pipelines, hitting 90% accuracy versus a 65% baseline. Curating 598 documents into 12 structured artifact types cost just $2.31, signaling strong efficiency gains for enterprise AI deployments.

GENERATIVE AI COST STRATEGY

Microsoft Dumps OpenAI, Anthropic Models to Cut Costs

Microsoft is shifting away from OpenAI and Anthropic AI models in favor of its own MAI model family to reduce costs, according to Bloomberg. Tens of thousands of prompts in Excel and Outlook previously handled by third-party models are now routed through MAI models, though this remains a small fraction of overall usage. Microsoft AI CEO Mustafa Suleiman called Anthropic extremely expensive and stated the goal is to eliminate that cost entirely. The move reflects a broader industry trend, with Amazon, Meta, Accenture and Uber also cutting AI spending. Microsoft's OpenAI partnership deal expires in 2032.

NEW AI ENTERPRISE LAUNCH

Microsoft Launches $2.5B Enterprise AI Consulting Firm

Microsoft has launched Microsoft Frontier Company, a consulting organization designed to help enterprises plan and deploy AI solutions. The initiative comes with a $2.5 billion investment and a workforce of 6,000 industry experts. The move signals Microsoft's aggressive push to capture enterprise AI adoption, positioning itself as a hands-on partner for businesses navigating AI deployment challenges.

NEW AI CHIP DEVELOPMENT

DeepSeek Designs Own AI Chips, Ditching Nvidia

DeepSeek is developing custom in-house inference chips, according to a Reuters report citing three sources familiar with the plans. The initiative, roughly a year in the making, involves talks with chip design, foundry, and memory companies, plus efforts to hire experienced chip designers. The move aims to reduce dependence on Nvidia and Huawei, DeepSeek's current primary suppliers. It mirrors strategies by OpenAI and Anthropic, which are also pursuing proprietary silicon. DeepSeek's recent $7.4 billion funding round is expected to support the costly endeavor.

NEW AI CHIP DEVELOPMENT

Anthropic Eyes Custom AI Chip With Samsung

Anthropic is in talks with Samsung to develop a custom AI chip, according to The Information. The company has yet to determine the chip's purpose, power specifications, or server integration. The move follows competitor OpenAI's announcement of its custom inference chip, dubbed Jalapeño, built with Broadcom. Anthropic maintains that its compute strategy will continue to rely on chips from Google, Amazon, and Nvidia, but declined to comment further on the Samsung partnership.

AI CHIP MANUFACTURING DEAL

Meta Eyes $6.5B AI Chip Deal With Samsung

Meta Platforms is reportedly in discussions with Samsung Foundry for a $6.5 billion deal to manufacture its MTIA AI chips. The agreement would mark a significant investment in Meta's custom silicon strategy. The deal highlights Meta's push to reduce reliance on third-party chip suppliers like Nvidia by developing its own AI infrastructure through the Meta Training and Inference Accelerator chip program.

NEW GENERATIVE AI LAUNCH

Meta Debuts AI Image Model With Code, Search Tools

Meta Platforms launched Muse Image, an image generation model with web search and code-writing capabilities. The model, released through the Meta AI chatbot, can generate images from detailed prompts, edit existing photos, and use reasoning before generating output rather than the common best-of-N approach. Muse Image integrates with Meta's Muse Spark reasoning model for complex tasks and features self-refining behavior that emerged during training. Currently available in limited markets, Meta plans to expand it to Facebook, Messenger, and Instagram, with advertiser access through its Advantage+ tools coming soon.

Upcoming Data & AI Conferences

Ai4 2026

Tue, Aug 4 - Thu, Aug 6, 2026 | Las Vegas, NV

Curated with AI by EMC2 AI

Brought to you by Estevan McCalley

Share this issue

Follow Estevan

Like this? Subscribe for the weekly AI roundup.