Press "Enter" to skip to content
DALL-E 3

AI Startups: Lightmatter Raises $400M, and a Frenchman’s Accent Troubles

AI startups with interesting business models that have recently raised capital. Some of the founders came from Intel Labs, Google Search, or Oracle Cloud.

Lightmatter

Mountain View, Calif.-based Lightmatter is the developer of what it describes as the world’s fastest photonic engine, ‘Passage.’ This technology uses light to move data between chips. It aims to solve the problem of traditional electronic interconnects that are bottlenecks for scaling AI workloads, since these cannot keep up with the growing demand for high-bandwidth, low-latency movement of data.

‘Passage’ leverages 3D-stacked photonics chips to move data, which the startup said dramatically increases AI cluster bandwidth and performance while decreasing power consumption. By enabling leaps in computing power, “this is how you get to AGI,” said co-founder Nicholas Harris, at a Sequoia Capital talk earlier this year.

Founders: Nicholas Harris, Darius Bunandar and Thomas Graham

Fundraise: $400 million, Series D

Investors: New investors advised by T. Rowe Price Associates, with participation from existing investors, including Fidelity Management and Research Company and GV (Google Ventures)

Use of proceeds: To ready mass deployment of ‘Passage’ in partner data centers, enabling scaling required for sustained AI innovation

“AI is evolving faster than anyone could have predicted, pushing the limits of data center technology,” said Erik Nordlander, general partner at GV, in a statement. “Photonics isn’t just a breakthrough; it’s the future of million-xPU data centers for AI. … Lightmatter is the definitive leader in data center photonics, and we’re excited to stand behind them as they unlock the next era of AI innovation and scale.”

Gladia

Parisian startup Gladia provides real-time AI transcription in over 100 languages, along with enhanced support for accents and the unique ability to adapt to different languages on the fly.

Founders: Jean-Louis Quéguiner and Jonathan Soto

Fundraise: $16 million, Series A

Investors: Led by XAnge, with participation from Illuminate Financial, XTX Ventures, Athletico Ventures, Gaingels, Mana Ventures, Motier Ventures, Roosh Ventures, and Soma Capital. Earlier seed investments were led by New Wave, Sequoia Capital (as part of the First Sequoia Arc program), Cocoa and GFC.

Use of proceeds: Advance its R&D efforts and bring to market a one-stop AI toolkit for audio, and expand its product offering with additional à la carte models. It is currently piloting a contact center agent-assist solution powered by Gladia’s real-time AI engine. The startup also plans to expand its workforce as it prepares to tap international markets.

“I founded Gladia for a very personal reason – I was frustrated that existing audio transcription services were not able to understand my French accent,” said Quéguiner, in a statement. “Our international team and customers often switch between languages during meetings, but finding a transcription solution that can handle different languages and accents simultaneously was impossible.”

DataCrunch.io

Finnish startup DataCrunch.io is an AI-focused cloud provider and aims to become Europe’s first hyperscaler. Founded in 2020, the startup’s services have been used by developers at OpenAI, Sony, 1x.tech, Freepik, Nex.art, Manifest.ai, Premai.io. 

Founder: Ruben Bryon

Fundraise: $13 million, seed funding

Investors: Led by byFounders, J12 Ventures and angels including Aiven founder Oskari Saarenmaa, Tuomo Riekki, founder of Smartly, and former AI researchers and founders from DeepMind and Elo Health. There is also support from Finnish insurer Local Tapiola and Nordic bank Nordea.

Use of proceeds: The startup plans to use the funds to scale infrastructure, expand its team and broaden its service within Europe and elsewhere. It is set to to launch its use of Nvidia H200 servers and clusters, and will adopt GB200 NVL72 clusters next year, further expanding the speed, capacity, and capabilities of its infrastructure.

“DataCrunch was born out of frustration with the existing hyperscaler offerings. AI companies deserve better access to computing without the complexity and high costs that have become the industry norm,” said Bryon, in a statement. “With this new round of funding, we are scaling our infrastructure to meet the growing demand, and firmly positioning ourselves as Europe’s leading provider of AI infrastructure.”

ApertureData

San Francisco-based ApertureData aims to solve the problem of serving multimodal AI models with data (text, images, video) that are siloed in different parts of an organization. Its solution manages and accesses complex multimodal datasets through one central interface. The startup said its ApertureDB product is 35x faster than existing disparate solutions and 2x to 4x faster than other open-source vector databases.

Founders: Vishakha Gupta and Luis Remis, who came from Intel Labs. (Their firsthand experience with the complexities of visual data management led to the creation of ApertureDB.)

Fundraise: $8.25 million, seed round (oversubscribed)

Investors: Led by TQ Ventures with participation from Westwave Capital, Interwoven Ventures, and a group of high-caliber angel investors. Existing investors also reaffirmed their commitment.

Use of proceeds: Scale its production deployments, enhance user experience through improved documentation and sandbox environments, focus on ecosystem integrations, and significantly expand its sales and marketing efforts.

“ApertureData has steadily built an amazing business with a wide view on the tech stack. They knew early on that traditional databases, which are geared toward textual data, would be insufficient for managing more complex multimodal data,” said Andrew Marks, general partner at TQ Ventures, said in a statement.

“The quantum of multimodal data and the desire to leverage it for analysis and machine learning is likely to explode over the coming decade as we are already seeing with the growth in use cases for generative and multimodal AI.”

Simplismart

San Francisco-based Simplismart developed a fast inference engine that lets organizations run machine learning models at what it describes as “lightning speeds.” For example, its software-level optimization helps run Llama 3.1 (8B) at a throughput of over 440 tokens per second.

Founders: Amritanshu Jain, formerly at Oracle Cloud, and Devansh Ghatak, formerly at Google Search

Fundraise: $7 million, Series A

Investors: Led by Accel, with participation from Shastra VC, Titan Capital, and angels including Akshay Kothari, co-founder of Notion

Use of proceeds: R&D and growth for its enterprise-focused MLOps orchestration platform.

“As GenAI undergoes its Cambrian explosion moment, developers are starting to realize that customizing and deploying open-source models on their infrastructure carries significant merit; it unlocks control over performance, costs, customizability over proprietary data, flexibility in the backend stack, and high levels of privacy/security,” said Anand Daniel, partner at Accel, in a statement.

“We were happy to see that Simplismart’s team saw this opportunity quite early, but what blew us away was how their tiny team had already begun serving some of the fastest-growing GenAI companies in production. It furthered our belief that Simplismart has a shot at winning in the massive but fiercely competitive global AI infrastructure market.”

Author

Be First to Comment

Leave a Reply

Your email address will not be published. Required fields are marked *