Good morning and welcome back to another edition of The AI Rundown by Lightscape Partners.

• Nvidia has introduced DGX Spark and DGX Station, two AI supercomputers designed for personal and enterprise-level AI development. Built on the Grace Blackwell platform, these systems offer high-speed AI computing with massive memory capabilities, bringing data-center-level performance to the desktop.

• OpenAI has launched new steerable audio models that let developers control tone and delivery in AI-generated speech. The models outperform Whisper in accuracy, especially in noisy or multilingual settings, and are already integrated with OpenAI’s Agent SDK to power next-gen voice agents.

• Global data center investments have surged to $455 billion, up 51% in 2024, as demand for AI infrastructure skyrockets. Hyperscalers like AWS, Microsoft, and Google are leading the charge, while Nvidia dominates the AI processor market, powering the next wave of enterprise and cloud computing.

Stay tuned as we explore these stories and their implications for the future of AI, technology, and innovation.

If you haven’t yet, please support the newsletter by subscribing!

Hardware

NVIDIA introduces DGX Spark and DGX Station personal AI supercomputers. Link.

• NVIDIA's new DGX Spark and DGX Station systems leverage the NVIDIA Grace Blackwell platform for desktop AI development.

• DGX Spark, dubbed the world's smallest AI supercomputer, features the NVIDIA GB10 Grace Blackwell Superchip, enabling up to 1,000 trillion operations per second.

• DGX Station offers data-center-level performance with a massive 784GB coherent memory, designed for large-scale AI workloads.

• Systems will be available for reservation, with manufacturing partnerships including ASUS, Dell, and HP.

Models

Google introduces AI models to transform drug discovery. Link.

• Google unveils the TxGemma models, derived from the Gemma family, aimed at supporting drug discovery through advanced AI capabilities.

• TxGemma can understand text and structures, aiding researchers in predicting the safety and effectiveness of potential therapies.

• The models will be available for scientific evaluation through the Health AI Developer Foundations programme, though commercial use details remain unspecified.

• The rapid adoption of AI in drug discovery has led to regulatory interest, with the FDA and EMA issuing guidance and reflections on the technology's application in the sector.

Enterprise AI Applications

GM partners with Nvidia to integrate AI across factories and self-driving cars. Link.

• General Motors and Nvidia announce an expanded collaboration to implement AI technologies in various branches of GM’s operations, from manufacturing to autonomous vehicles.

• Nvidia will supply essential AI infrastructure to GM, assisting in the development of custom AI systems utilizing Nvidia's advanced products and platforms.

• The partnership highlights the use of Nvidia Omniverse with Cosmos for creating digital twins of factories, aiding GM in training AI models for manufacturing and robotics innovation.

• GM will employ Nvidia Drive AGX technology for upcoming driver-assistance systems, transitioning focus from robotaxi development to enhance its Super Cruise feature.

Product Launches

Meta AI launches in Europe, expanding with new chat features. Link.

• Meta introduces its AI assistant into Europe, aiming to improve user experiences by integrating into popular apps like Facebook, Instagram, WhatsApp, and Messenger.

• The rollout faces challenges from navigating Europe's regulatory frameworks but includes chat functionality in six European languages.

• Meta AI offers new group chat capabilities and personalized content curation, enhancing day-to-day interactions and digital engagement.

• Future plans include expanding its AI offerings globally, building on US features like personalization and memory to enhance social user experiences.

Startup Funding & Valuations

Cognition AI reaches $4 billion valuation with new funding from Lonsdale's 8VC. Link.

• Cognition AI, known for its AI-powered coding assistant Devin, has doubled its valuation to nearly $4 billion in a new funding round led by 8VC.

• The raise indicates robust investor interest in AI-driven coding tools, despite intense competition from established players like GitHub Copilot.

• Existing investors such as Founders Fund and Khosla Ventures participated in this round, showing confidence in Cognition’s potential.

• Founded in 2023, Cognition continues to scale rapidly, with its CEO Scott Wu bringing experience from co-founding the networking tool Lunchclub.

Dataminr secures $85M to fuel AI and international expansion. Link.

• Dataminr receives $85M in funding from NightDragon and HSBC, aimed at expanding AI capabilities and international presence.

• The capital will assist in developing Generative and Agentic AI for real-time decision-making amidst global uncertainties.

• Dataminr plans to leverage its proprietary AI models to enhance product offerings in Europe, the Middle East, and Asia.

• NightDragon's Dave DeWalt joins Dataminr’s board, underscoring the strategic significance of this investment.

Graphite secures $52M led by Accel, with backing from Anthropic for their AI code review platform. Link.

• Graphite raised $52 million in a Series B round led by Accel, supported by Anthropic's Anthology Fund, Menlo Ventures, and others, aiming to enhance their AI code review capabilities.

• Founded by former engineers from Facebook, Airbnb, and Google, Graphite began as mobile development tooling before pivoting to AI-driven code review, addressing errors, and suggesting code changes.

• The company plans to introduce 'Diamond', a standalone AI tool for automatically detecting bugs, capitalizing on burgeoning demand for reliable AI-powered coding assistants.

• With revenue increasing 20x in 2024 and a client base including Shopify and Snowflake, Graphite aims to use the latest funds to further product development and expand its NYC-based team.

Content Creation

Stability AI launches Stable Virtual Camera to transform photos into 3D video scenes. Link.

• Stable Virtual Camera uses AI to convert 2D images into videos with realistic depth, providing novel views from specified camera angles.

• Users can create videos using up to 32 images, featuring dynamic camera paths like 'Spiral' and 'Dolly Zoom.'

• The current version supports various aspect ratios and up to 1,000 frames but may vary in quality with complex scenes or dynamic textures.

• The model, available on Hugging Face under a noncommercial license, comes amid past financial challenges for Stability AI.

Data Centers + Energy

Data center investments hit $455B as AI drives expansion. Link.

• Global investments in data centers rose by 51% in 2024, reaching $455 billion, with forecasts of a further 30% growth in infrastructure spending this year.

• Over half of the spending came from the top 10 hyperscalers, with AWS, Microsoft, and Google leading substantial investments in AI infrastructure to meet the soaring demand.

• Enterprises increased their hardware spending by 25% to $150 billion, the highest growth rate tracked since 2014, with rising demand affecting rental rates in key markets like Northern Virginia.

• Nvidia dominated the AI processor market, with its GPUs accounting for over 90% of shipments; server revenue, driven by AI infrastructure needs, hit a record $236 billion for the year.

OpenAI

OpenAI launches steerable audio models for voice agents. Link.

• OpenAI released new speech-to-text and text-to-speech models that outperform Whisper in accuracy, especially in noisy and multilingual settings.

• The new text-to-speech model allows developers to control tone and delivery—for example, making agents sound empathetic or dramatic.

• Models are built on GPT-4o architectures and trained using reinforcement learning and self-play to mimic realistic user interactions.

• These audio tools are now available in the API and integrate with OpenAI’s Agent SDK for streamlined voice agent development.

ChatGPT faces complaint for defamation and breaching data accuracy. Link.

• Austrian privacy group Noyb filed a complaint against ChatGPT, alleging it fabricated a story about a user committing child murder, violating GDPR data accuracy standards.

• Noyb claims OpenAI's disclaimers are insufficient and that the AI's data inaccuracies can lead to damaging misinformation about individuals.

• The complaint was filed with Norway’s data protection authority, requesting fines and model adjustments for accuracy improvements.

• ChatGPT's information inaccuracies allegedly occur 3-27% of the time, raising concerns over AI’s reliability in Europe where a task force addresses AI-related privacy issues.

Thank you for reading the AI Rundown by Lightscape Partners. Please send any questions, comments, or suggestions to [email protected].