SAN FRANCISCO — March 5, 2025 — Ceramic.ai emerged from stealth at this time with software program for basis mannequin coaching infrastructure designed to allow enterprises to construct and fine-tune generative AI fashions extra effectively.
Based by Anna Patterson, former Google VP of Engineering and Gradient Ventures founder, Ceramic.ai mentioned it improves AI mannequin coaching pace and cost-efficiency, providing as much as a 2.5x efficiency enhance, accelerated by NVIDIA.
Ceramic.ai additionally mentioned it secured $12 million in seed funding from NEA, IBM, Samsung Subsequent, Earthshot Ventures and Alumni Ventures.
“Within the midst of a surge in AI adoption, too many firms are nonetheless hindered by boundaries to scale – from prohibitive prices to restricted infrastructure,” mentioned Patterson, CEO. “We’re democratizing entry to high-performance AI infrastructure so firms can navigate the complexity of AI coaching with out spending tons of of hundreds of thousands in analysis and engineering sources. However the shift to enterprise AI isn’t nearly higher instruments – it’s about altering how companies work. If AI adoption had been a
baseball sport, we’d nonetheless be singing the nationwide anthem.”
International AI investments are experiencing explosive progress from $16 billion in 2023 to an estimated $143 billion by 2027. Regardless of this surge in spending, 74 % of firms nonetheless battle to scale AI successfully and obtain worth. A serious problem is that constructing AI infrastructure is dear, advanced, and resource-intensive. Whereas tech giants spend billions creating proprietary AI infrastructure, most enterprises lack the engineering sources to optimize and scale their very own AI fashions.
Present AI infrastructure can scale as much as 10x, however not 100x—true exponential progress calls for an entire redesign. Ceramic.ai bridges this hole by offering an enterprise-ready platform that isn’t simply quicker however basically extra scalable to energy the subsequent era of AI, dramatically lowering the complexity and value of AI mannequin coaching.
The software program platform’s mannequin can practice with lengthy contexts and any cluster dimension, enabling enterprises to develop, practice, and scale their very own AI fashions quicker than conventional strategies. For smaller fashions, Ceramic.ai is as much as 2.5x quicker on NVIDIA H100 GPUs than present state-of-the-art platforms, and for large-scale long-context fashions, Ceramic.ai is the one viable selection for quick coaching.
Ceramic.ai has developed a complete platform that addresses the core challenges of enterprise AI deployment:
● Pace and Effectivity: Ceramic.ai’s coaching infrastructure delivers as much as 2.5x larger effectivity than open-source stacks, chopping down coaching prices whereas enhancing mannequin efficiency.
● Unique Lengthy-Context Coaching Functionality: Ceramic.ai is the one platform that may practice massive fashions on long-context information, offering unequalled high quality and efficiency. The corporate outperforms all reported benchmarks for long-context mannequin coaching, sustaining excessive effectivity even for 70B+ parameter fashions.
● Superior Reasoning Mannequin Efficiency: Ceramic educated a reasoning mannequin for problem-solving and achieved an actual match Go@1 rating on GSM8K of 92% tuning Meta’s Llama70B 3.3 base mannequin up from 78% and outperforming DeepSeek’s R1 84%.
● Optimized Information Processing: Ceramic re-orders coaching information, guaranteeing every micro-batch is aligned by matter. Present approaches both masks away different paperwork, dropping the advantage of longer context size or take note of irrelevant paperwork, studying dangerous habits. By re-ordering coaching information so it is available in 64k or 128k contexts, all on the identical matter, we enhance the variety of information factors the place consideration can be taught quadratically.
Constructed by a group of consultants in large-scale infrastructure, Ceramic.ai has already helped present enterprises scale back prices and enhance mannequin coaching effectivity in early trials. They’re partnering with Lambda, AWS and others for accelerated coaching.
“Ceramic.ai is a game-changer for AI builders and enterprises searching for elevated effectivity and superior price-performance,” mentioned Sam Khosroshahi, VP, BD & Strategic Pursuits I AI & Machine Studying at Lambda. “Mixed, our choices present clients with an accelerated full-stack answer, validated and backed by each infrastructure and mannequin experience. This permits clients to realize quicker outcomes, decreased improvement prices, and higher-quality options.”
“AI’s meteoric ascent has been like a rocket tethered to a horse-drawn carriage – till now,” mentioned Lila Tretikov, Associate and Head of AI Technique, NEA. “Anna and her group at Ceramic.ai have algorithmically shattered a vital bottleneck in mannequin coaching, making it quicker, extra environment friendly, and scalable. With Ceramic, firms can scale their already huge AI coaching workloads 100x – with out the corresponding surge in value or complexity.”
“Our funding in Ceramic demonstrates how IBM drives innovation and solidifies partnerships in extremely strategic areas,” mentioned Emily Fontaine, Vice President, IBM International Head of Enterprise Capital. “We’re thrilled to collaborate with Ceramic to deal with a vital want to scale back AI compute prices, making coaching extra environment friendly and accessible.”
