OpenAI partners with Cerebras for 750MW AI Inference Deployment

AI enterprises, take note: Ziroh Labs’ Kompact AI is paving the way for CPU-native LLM inference—offering up to 80% cost savings and faster deployment, all on existing infrastructure. Is this the future of scalable enterprise AI? Read more via @StartupTalky: startuptalky What benefits and challenges do you see in shifting from GPU to CPU-native AI? https://lnkd.in/gVWZKbwu #startupnews #growth

Nvidia, Others in Talks for OpenAI Funding, invest as much as $60 billion in OpenAI round of as much as $100 billion valuation of about $750 billion to $830 billion, Information Says--- generative artificial intelligence ---

OpenAI just made a multibillion-dollar computing deal with Cerebras Systems. And the most interesting part is not the money. 👇 First: the unit of this agreement is megawatts. OpenAI committed to buy up to 750 MW of compute over three years. This is not “how many GPUs”, this is “how much power does your data center consume”. That alone tells a lot about where AI scaling is today ⚡ Second: Cerebras is weirdly fast, at least so far. 🚀 Out of curiosity, I tried to challenge https://chat.cerebras.ai/ with my daughter’s homework: Solve the crossword: to bother or annoy someone repeatedly, seven letters, the third is o. The response came almost instantly. Subjectively, it felt tens or even hundreds of times faster than what im used to — yes, yet?, but still impressive. Why this matters. We are slowly moving from a world where AI progress is limited by models and ideas to a world where it is limited by power, hardware architecture, and supply chains. Nvidia is still dominant, but deals like this show that the industry is actively looking for exits from the single-vendor bottleneck. If Cerebras delivers at OpenAI scale, this may become a real shift, not just an experiment. 👀 Compute is the new oil — and now we measure it in megawatts, not chips 🔌🤖

https://lnkd.in/ewrp4p77 "Cerebras' $10 billion deal with OpenAI positions the startup and its wafer-scale engine as a challenger to Nvidia in the AI chip market, while helping OpenAI try to accelerate the performance of its large AI models"

OpenAI Releases a Research Preview of GPT‑5.3-Codex-Spark: A 15x Faster AI Coding Model Delivering Over 1000 Tokens Per Second on Cerebras Hardware OpenAI has launched GPT-5.3 Codex-Spark, a research preview optimized for near-instant coding by delivering over 1000 tokens per second—a 15x speed increase over the flagship model. This massive performance jump is powered by the Cerebras Wafer-Scale Engine 3 (WSE-3), which eliminates traditional GPU bottlenecks by keeping all compute on a single silicon wafer, paired with a new persistent WebSocket connection that reduces networking overhead by 80%..... Full analysis: https://lnkd.in/gCvBfM47 Technical details: https://lnkd.in/gn_AkgmQ OpenAI Cerebras

As multi-modal models become the norm I think we are going to see a lot more impact from these "old school" adversarial ML attacks. If your whole AI red team approach is natural language prompt injection and you don't understand how these models actually work you are going to be missing a large attack surface.

Alibaba just dropped the hammer on Google and Nvidia, proving that in 2026, raw parameter count is a vanity metric while inference efficiency is the only metric that actually pays the bills. Here is your high-velocity technical breakdown of the last 24 hours in AI. 1. Alibaba Decouples Intelligence From Size with "RynnBrain" The Drop: Alibaba’s DAMO Academy released RynnBrain, an open-source embodied AI model stack optimized for robotics, available now on Hugging Face. The Specs: Built on the Qwen3-VL vision-language backbone, RynnBrain activates only 3 billion parameters during inference yet reportedly crushes Google’s Gemini Robotics-ER 1.5 on 16 benchmarks. Architecture: Seven variants released, ranging from a tiny 2B model to a complex Mixture-of-Experts (MoE) architecture. Technical Impact: This is a localized inference breakthrough. By keeping active parameters at 3B, Alibaba enables sophisticated embodied reasoning directly on robot hardware (edge compute) without the latency crippling cloud-dependent rivals. 2. Mistral Shreds Latency with "Voxtral" The Drop: French labs released Voxtral Realtime and Voxtral Mini Transcribe V2 under Apache 2.0. The Specs: A 4 billion parameter model capable of processing speech-to-text with 200 milliseconds of latency. Supports 13 languages. Target: Edge devices (phones/laptops). Technical Impact: 200ms is the "magic number" for natural human interruption. Mistral has effectively solved the turn-taking lag that makes most voice bots feel robotic. This is the new baseline for edge-native voice agents. 3. The Code War: OpenAI Strikes Back with GPT-5.3 Codex The Drop: Following Anthropic's Claude Opus 4.6 release, reports confirm OpenAI has deployed GPT-5.3 Codex. The Specs: Early telemetry shows a Terminal Bench score of 77.3 (vs. Opus 4.6's 65.4). Efficiency: Significant reduction in token consumption for complex autonomous loops. Technical Impact: We are seeing a divergence in model distinctiveness. While Opus focuses on reasoning depth, GPT-5.3 Codex is optimizing for execution. The massive token reduction implies better internal "thought compression," allowing agents to run longer autonomous loops before hitting context windows or cost barriers. 4. The "Speed King" Benchmarks: Liquid Takes the Crown The Drop: Independent "Agentic AI" benchmarks released Feb 10. The Leader: Liquid LFM 2.5 is officially the fastest model in its class, hitting ~359 tokens/second. The Value Play: Ministral 3B clocked ~293 tokens/sec, offering the best price-to-performance ratio. Technical Impact: For agentic workflows requiring multiple steps (reflection, planning, execution), inference speed is the bottleneck. Liquid LFM 2.5's throughput allows for "thought-heavy" agent architectures that would be too slow on transformer-based architectures. #EmbodiedAI #EdgeAI #InferenceEfficiency #ArtificialIntelligence #TechNews

🚀 NVIDIA MAKES ITS LARGEST INVESTMENT EVER - IN OPENAI Jensen Huang just announced in Taipei that Nvidia will participate in OpenAI's latest funding round with what could be "the largest investment we've ever made." - Huang called OpenAI "one of the most consequential companies of our time" - The GPU giant is now directly investing in its biggest customer - This signals a major shift in how AI infrastructure is financed - The company that sells the picks and shovels is now buying into the gold mine The AI supply chain is consolidating faster than anyone expected. Sources: - https://lnkd.in/gNxEsTix #Nvidia #OpenAI #AIInvestment #TechNews #ArtificialIntelligence

Project Stargate is a $500B push from OpenAI, Nvidia & Oracle to dominate AI with national-scale data centers. But could centralizing AI power risk stifling competition? #TechMonopoly #AIPolicy https://lnkd.in/gysQKvHT

Nvidia and OpenAI have been lynchpins of the AI boom, but their relationship has come under new scrutiny in recent days amid reports of tensions between the two firms.