Fractile raises $720M for AI chips
The British startup is scaling the development of purpose-built inference hardware.
ddNews Β· a daily AI briefing for Dirk (generative video/ddStudio, Lit
The British startup is scaling the development of purpose-built inference hardware.

A specialized model designed for high-speed closed-decision evaluations across text, image, audio, and video.

The provider is moving toward customer prepayments and shorter contracts to fuel rapid AI growth.

The Ramp-AI-Index shows October growth at only 1.65%, with a small group of firms driving the bulk of revenue for top labs.
A new security framework for AI agents implementing "deny-by-default" to prevent prompt injection and data exfiltration.
The provider is deepening open-source contributions to solve production bottlenecks for frontier model serving.

Google DeepMind's AI mathematically proved the security of Signal's core code in four weeks.
Amazon releases an open-source emulator allowing developers to validate cloud code locally without AWS connectivity.

Coding agents now run in GitHub Actions with built-in isolation and budget controls.
The orchestrator is shifting from container management to the primary layer for running enterprise AI agents.
A next-generation vision-language-action (VLA) model aimed at creating a more capable "AI brain" for robots.

Research shows that asking "Are you sure?" can flip correct answers to wrong ones in models like Qwen2.5-1.5B.

CEO Ali Ghodsi is using a new agent to synthesize unvarnished organizational truth from 12,000 employees.
The company is focusing on bridging the gap between enterprise AI ambitions and existing infrastructure.

A European AI platform using agents to condense sensor data into weapon system target suggestions.

New model preserves tone, emotion, and speaking pace during real-time translations.

The company used its "Mythos" model to port Claude to AMD GPUs in just three days with minimal human input.

Reports indicate $500M in training data, including RL environments, sold to Chinese labs.
A next-gen quantum processor designed to execute 10,000 operations.

Despite a 50% revenue increase, the company reported a massive adjusted operating loss of $10.9B.

A DeepMind model that maps how DNA changes affect individual cells to uncover the roots of disease.

New experimental features allow users to navigate 3D spaces within photos and interview AI characters.
New server processors specifically optimized for AI agent workloads.
The neocloud provider raises full-year guidance, signaling sustained high demand for AI compute infrastructure.
Funding focused on scaling human-in-the-loop evaluation for enterprise AI agents.
The company is seeking to raise $64 billion through its initial public offering.

The AI lab has acquired a travel startup founded by the creator of Sonder.

Major developers must now register and disclose safety practices or face significant fines.

Introduction of a Business Hub and "Business Agents" allowing companies to manage profiles and customer interaction data.
New industry focus on the security risks of autonomous agents taking actions via credentials and tools.

Anthropic enables partner apps (Adobe, Atlassian) to be embedded directly within Claude.

Built on Apache Iceberg and R2, providing open-format data management.

New robotics model that learns tasks by transferring human hand movements from video to robot arms without manual teleoperation.

Research shows that legacy technical debt in codebases can derail AI coding agents.

New model featuring "Agent-Swarm" capabilities, claiming top benchmark performance at a lower cost.

A new workplace tool that transforms live discussions into actionable plans, timelines, and tasks.

New server CPUs specifically tuned to optimize performance for agentic AI workloads.
Zen 6 architecture brought to embedded systems for semiconductor validation.

Google DeepMind and Apptronik are testing humanoid robot control on active Mercedes-Benz production lines.
Purpose-built AI agents designed to automate cybersecurity threat hunting.

New specialized chips are launching to support confidential model training and increased memory capacity.
Troubleshooting agents now support virtual machines in addition to Kubernetes clusters.
Startup raises $4.6M to develop high-speed optical links for chiplet-based AI hardware.
A new platform designed to scale the production volume of training data for physical AI models.

New agentic workflows for Photoshop, Premiere, and Firefly, including a prototype builder to compete with Figma.

Shares are expected to list between $70 and $78, potentially valuing the company at $1.404 trillion.
Graph query engine enabling AI agents to query complex enterprise data as a graph.
A new Linux-based OS with built-in AI models and agents, targeting local AI power users.

New partnership allows Adobe AI tools to function directly within Gemini and Google Drive.
A new insurance product covering losses caused by AI agents deployed by small and mid-sized businesses.

A new Linux-based desktop alternative to Windows/macOS built specifically for running local AI models.

Consumer indoor security drone and an enterprise security robot utilizing former Amazon Astro technology.

The company is launching a home security drone and a rolling patrol robot for business security.

Eight states and major labs (OpenAI, Anthropic, Mistral) commit to implementing "kill switches" for AI.
A new runtime specifically designed to bring AI agents into production infrastructure securely.
Security software that vets and intercepts AI agent actions before they execute in enterprise environments.

Exec warns that cyberattacks may force tighter security restrictions on open AI platforms.

The company targets a valuation of up to $1.4 trillion with a price range of $70β$78 per share.

The company confirmed a potential IPO with a share price between $70 and $78.

New data shows it significantly outperforms GPT 6 Astra and Opus 5.5 in real-world work.

Verified brand agents will allow users to perform comparison shopping and interact with e-commerce services.

Implementation of resilient batch intelligence for nightly inventory forecasting across 4,700 stores.

Adobe is enhancing object removal tools across Photoshop, Lightroom, and Premiere for faster editing.

The new "strongest model" features a "swarm" agent capability for cybersecurity and general public use.

The company expands its enterprise offering with a dedicated suite of productivity tools.

New API-driven billing budgets allow pausing services automatically when monthly limits are hit.

Strategic shift toward automating repetitive creative tasks via autonomous design assistants.

Projections indicate systems spending will "go through the roof" in 2027.

New frontier models capable of solving open physics problems and outperforming GPT-6 Astra in professional freelance tasks.

Analysis suggests the primary bottleneck for AI scaling is shifting from chip availability to electricity grid capacity.

A highly efficient small model optimized for low-cost agent routing, tool calls, and data extraction.

TPU accelerators are being redesigned to better support broader AI inference workloads.

Technical benchmark shows continuous batching outperforms static/dynamic methods by up to 2.91Γ.

A general-purpose physical-reasoning model for robot control, Real2Sim construction, and policy improvement.

Development of a public CA for Merkle Tree Certificates to reduce bandwidth overhead in post-quantum crypto.

Native PyTorch backend allowing models to run on AWS Trainium chips without code modifications.

Migration to a fine-tuned small model resulted in measurable increases in email and push engagement.