Ultrathink

News

Breaking tech news, funding rounds, and AI developments—built for signal.

Back to home
Breaking
Dec 15 • 3 min
NVIDIA Buys Slurm Creator SchedMD, Promising Open-Source Continuity for HPC Workloads
NVIDIA has acquired SchedMD, the company behind Slurm—the workload management system running on more than half of the world's top 100 supercomputers. The move extends NVIDIA's reach deeper into the AI training stack while raising questions about infrastructure consolidation.
Dec 15 • 3 min
OpenAI Buys Neptune: A Bet on Infrastructure as Models Grow Unwieldy
OpenAI just bought an experiment tracking company. That tells you everything about where frontier AI development is headed—not just bigger models, but the unglamorous infrastructure needed to keep them from becoming unmanageable.

Latest

FUNDING • Dec 15 • 5 min
Fal Triples Valuation to $4.5B in Five Months
In a market where AI valuations are under intense scrutiny, Fal just pulled off something remarkable: tripling its valuation to $4.5 billion in five months flat. Sequoia's leading the $140M Series D, and the numbers tell a story worth paying attention to.
FUNDING • Dec 15 • 5 min
Unconventional AI's $475M Seed Shatters Records
A two-month-old startup just raised $475 million at a $4.5 billion valuation. Unconventional AI, founded by former Databricks AI head Naveen Rao, is betting that the future of AI computing looks a lot more like your brain than your GPU.
FUNDING • Dec 15 • 5 min
Enterprise AI Spending Explodes to $37B
Enterprise spending on generative AI has exploded from $1.7 billion in 2023 to $37 billion in 2025, now capturing 6% of the global SaaS market. But here's the real story: startups are winning, taking $2 for every $1 incumbents earn at the application layer.
ANALYSIS • Dec 15 • 5 min
OpenAI Wants AI to Admit When It's Wrong
OpenAI is testing a new training method called 'confessions' that teaches models to self-report their mistakes. If it works, it could fundamentally change how enterprises trust—and verify—AI outputs.
ANALYSIS • Dec 15 • 5 min
OpenAI Trains Models to Admit Their Mistakes
OpenAI is testing a training method called 'confessions' that teaches AI models to admit when they've made mistakes or acted undesirably. It's a direct attack on one of the most persistent problems in production AI: models that confidently lie rather than acknowledge uncertainty.
PRODUCT • Dec 15 • 5 min
Britain Bets Big on Sovereign AI Compute
The UK just secured its own AI compute destiny. OpenAI, NVIDIA, and Nscale are building Britain's largest supercomputer with up to 50,000 GPUs—a direct play for AI sovereignty in an increasingly fragmented global tech landscape.
PRODUCT • Dec 15 • 5 min
OpenAI Releases Open-Weights Models
The company that once warned open-source AI was too dangerous is now releasing open-weights models. OpenAI's pivot isn't charity—it's a calculated response to losing ground in the race for developer mindshare.
ANALYSIS • Dec 14 • 7 min
The Fine-Tuning Revolution
Companies are discovering that fine-tuned 8B models can outperform GPT-4 on their specific tasks—at 1/25th the cost. The era of one-size-fits-all AI is ending.
ANALYSIS • Dec 14 • 5 min
Unsloth Makes LLM Training 2x Faster
A YC startup with 43,000+ GitHub stars just made fine-tuning LLMs accessible to everyone. Unsloth delivers 2x faster training with 70% less VRAM—and it's completely open source.
ANALYSIS • Dec 14 • 5 min
Together AI Raises $305M
Together AI just raised $305M at a $3.3B valuation—backed by NVIDIA, Salesforce, and General Catalyst. They're betting the future of AI is open source, and they're building the infrastructure to prove it.