Led · Equity · Apr 2026

Founded in 2017 and based in Palo Alto, SambaNova designs and sells AI systems such as the SN40L and the next-generation SN50 for high-performance inference and model serving. The SN40L launched in September 2023 and became available on-premises in November 2023; the SN50 was unveiled in February 2026 and is due to begin shipping to customers in the second half of 2026 with SoftBank as its first deployment partner. SambaNova positions itself as a provider of “premium inference,” claiming the ability to fit multi-trillion-parameter models onto a single rack to run them quickly. The company has commercial wins including JPMorgan Chase, Saudi Aramco, Intel, and other Japanese firms. It has a multi-year partnership with Intel that includes co-developing products and go-to-market efforts, and it says current funding will help scale operations and secure its supply chain to meet strong demand.

Participated · Series F · May 2024

Scale AI produces and labels data used to train large language models and other AI systems. The company has begun hiring PhD scientists and senior software engineers to generate higher-quality data for frontier AI labs. Scale confirmed a significant investment from Meta that values the startup at $29 billion. The company said Meta’s investment will be used to pay investors and shareholders and to fuel growth, and emphasized it remains an independent entity. Co-founder and CEO Alexandr Wang is stepping down to join Meta, with Jason Droege — Scale’s chief strategy officer — serving as interim CEO; Wang will remain on Scale’s board as a director. The article notes Scale raised $1 billion last year from investors including Amazon and Meta at a $13.8 billion valuation and that SingalFire data showed a 4.3% loss of top talent last year. Scale AI provides data-labeling and annotation services, meshing machine learning with human-in-the-loop oversight to prepare large, often unstructured, datasets for AI training across industries. It specializes for different use cases such as camera and lidar data for self-driving cars and annotated text for NLP. Customers include Microsoft, Toyota, GM, Meta, the U.S. Department of Defense and OpenAI. Founded in 2016, the company says it will use new capital to accelerate the creation of “frontier data” and support scaling large language models toward future generational advances. The company experienced a 20% workforce reduction last year amid broader headwinds. After the Series F, Scale’s valuation doubled relative to earlier rounds, reflecting strong investor interest in data infrastructure for AI. Scale AI builds data-labeling and information-curation services for AI applications, serving enterprise clients across industries. Founded in 2016 by Alexandr Wang while he was at MIT, the company counts customers including General Motors, Nvidia, Nuro and Zoox. TechCrunch reports Scale is a break-even business. The company plans a large-scale increase in headcount as it builds new products and expands into additional markets. Alexandr Wang, a former Quora and Addepar employee, remains CEO as Scale scales its offerings. Scale AI is a four-year-old startup that built a visual data-labeling platform combining software and tens of thousands of human contractors to prepare image, text, voice and video data for machine-learning models. Its customer base began with autonomous-vehicle companies and now includes legacy automakers (GM, Toyota), chipmaker Nvidia, AV startups (Nuro, Zoox) and enterprise customers such as Airbnb, OpenAI, DoorDash and Pinterest. The company has expanded beyond labeling with Nucleus, an AI development platform for organizing, curating and managing large datasets, and acquired a four-person startup called Helia to support real-time video and neural-network training work. Scale is focused on building Nucleus into a fully integrated platform and on entering new markets and adding product and platform capabilities. Financially, Scale raised $155 million in its latest funding round, pushing its post-money valuation to more than $3.5 billion, and the company says it is now break-even. It plans to grow headcount from about 200 employees to roughly 350 by the end of next year (not counting its contractors). Scale AI supplies labeled training data (text, audio, images and video) to train machine-learning systems, using a mix of software and human annotation. The company delivers data via an API and applies human contractors to segment images, label frames, and add contextual language annotations to improve model accuracy. It combines a relatively small full-time staff (around 100 employees) with nearly 30,000 contractors who perform the bulk of labeling and quality control. Customers named in the article include Waymo, OpenAI, Airbnb and Lyft. The startup is three years old, and CEO Alexandr Wang (22) said the new funding will let Scale capitalize on its ambitions and that he hopes it will be the company’s last fundraise. While Scale aims to reduce reliance on humans over time through automation, the article emphasizes that human insight remains critical today to minimize bias and ensure high-quality labels, and Bloomberg reported the round values the company above $1 billion.

Participated · Series A · Apr 2024

Rivos develops RISC-V-based server chipsets with a data-parallel accelerator designed to speed AI and big-data analytics. Its prototype chip was made using TSMC’s 3nm process and the company is also building self-contained Open Compute Project–style data center hardware plus a firmware-to-app software stack. Rivos targets customers building generative AI and other data-driven solutions and plans to monetize via hardware sales and complementary software to large data center operators. The company is pre-revenue, has roughly 375 employees, and CEO Puneet Kumar said it holds hundreds of millions of dollars in the bank. Rivos was founded in 2021 and previously faced litigation with Apple that settled in February, a legal shadow that had complicated hiring and fundraising. The company does not expect mass production of its chip until sometime next year and emphasizes a “recompile-not-redesign” approach to ease customer adoption.

Participated · Series D · Aug 2023

Hugging Face operates a GitHub-like hub for AI code repositories, models and datasets alongside libraries for dataset processing and model evaluation, plus web apps to demo AI applications. Its paid offerings include AutoTrain for automated model training, an Inference API for hosted model serving, Infinity for faster in-production inference, and an enterprise hub supporting SaaS and on-prem deployments. The company also contributes to open-source models and initiatives such as BigScience and Bloom and has released models like StarCoder and SafeCoder. Hugging Face reports 10,000 customers, more than 50,000 organizations on the platform, and a model hub hosting over 1 million repositories; it has about 170 employees. Financially, the company was valued at $4.5 billion in its latest round and the valuation was reported to be more than 100 times its annualized revenue. Management says it plans to "double down" on research, enterprise and startup support and will recruit additional talent. Hugging Face hosts thousands of pre-trained machine-learning models, datasets, and developer-contributed repositories and positions itself as the “GitHub of machine learning.” Its Transformers library (on GitHub) has attracted significant attention, with 62,000 stars and 14,000 forks, and the site lets developers create, discover, and collaborate on models and apps. The company also offers hosted services, including an Inference API to use models via an API and an AutoTrain feature to train models. Ten thousand companies are using Hugging Face in some capacity, underscoring enterprise adoption. With the new funding, the company plans to continue expanding these developer- and enterprise-facing offerings rather than pivoting product strategy. Hugging Face maintains Transformers, an open-source NLP library that exposes popular models such as BERT, GPT, XLNet, T5 and DistilBERT. The Transformers GitHub repository has about 42,000 stars and 10,000 forks. Around 5,000 companies use Hugging Face in some capacity, including Microsoft (Bing) and Monzo; paying customers include Bloomberg and Typeform. The company recently launched paid offerings for prioritized support, private model management and a hosted inference API. Hugging Face reported being cash-flow positive in January and February 2021 and that roughly 90% of its prior $15 million round remains in the bank. It plans to triple headcount in New York and Paris and hire remote positions. Hugging Face began with a mobile chatbot app and has since released Transformers, an open-source library for natural language processing. Transformers has been downloaded more than one million times and its GitHub project has about 19,000 stars. Researchers at Google, Microsoft and Facebook have experimented with the library, and companies including Monzo and Microsoft Bing use it in production for tasks like classification, summarization, generation and conversational AI. The company plans to triple its headcount across New York and Paris with proceeds from the latest funding. The article frames Hugging Face’s core product as a widely adopted open-source NLP framework that underpins both research and commercial deployments. No revenue figures or other financial metrics are reported in the article. Hugging Face builds a chatbot app with a strong, personality-driven focus on emotions and entertainment for bored teenagers. The app is available in the App Store and on Kik, and the company is launching Hugging Face on Messenger to bring new users. It accepts text, photos, and emojis so the bot can interpret selfies and emotional cues. The service handles about 1 million messages per day and has processed over 100 million messages to date. The team has published AI research papers and plans to double its engineering staff in the coming months. The company recently raised a $4 million seed round to support product growth and hiring.

Participated · Series B · Jul 2023

Proprio builds Paradigm, a real-time surgery guidance device that maps pre-operative scans to a live 3D representation and shows surgeons sub-millimeter guidance during spinal procedures. The system’s Prism imaging array combines multiple RGB cameras, infrared stereo cameras and a dedicated depth sensor to continuously register vertebrae without physical markers. Proprio received FDA 510(k) clearance for stereotaxic navigation for spine surgery and plans to commercialize by deploying the platform in operating rooms and taking on first in-human cases this summer at partnering hospitals. Commercial rollout will require scaled manufacturing, testing, training and field support. Beyond spine surgery the company sees applications in other difficult procedures (for example joint reconstruction) and anticipates a data opportunity to collect high-fidelity surgical recordings for training, outcome analysis and retrospective review. The company announced $43 million in new funding and maintains an office in Seattle where the team and co-founders including CEO Gabriel Jones and CMO Dr. Samuel Browd are based. Proprio develops a computational imaging system that provides enhanced visualization to enable precise surgical execution. The platform integrates robotics, artificial intelligence and computer vision to simplify surgical workflow, improve procedural accuracy and reduce or eliminate exposure to radiation. Proprio’s system is intended to increase surgical productivity while improving precision. The company plans to use the new funding to expand its development teams, accelerate clinical and regulatory timelines, build commercialization capabilities and install the first Proprio systems in leading institutions. Leadership additions and technical advisors have been added to support commercialization and product development, including hires with expertise in regulatory affairs and operations and advisors from Microsoft and academia. The co‑founders and leadership team include CEO Gabriel Jones, Ken Denman, Dr. Joshua Smith, Head of Engineering James Youngquist and Chief Medical Officer Dr. Samuel Browd. Proprio Vision develops a light-field platform that captures real-time volumetric video, enabling surgeons to change viewpoints in immersive, accurate 3D. The system couples light-field arrays with software similar to a powerful video-game engine and is designed to work with existing medical imaging such as MRIs and CT scans. Immediate use cases include teaching via playback from different angles, surgical planning and remote collaboration, and the company aspires to supplant tools like surgical microscopes and fluoroscopy. Proprio emerged from stealth in Seattle and started as a secretive operation at the University of Washington; it has clinical research partnerships with Seattle Children’s and UW Medicine. The startup has 10 employees, plans to double headcount this year and hire AR/VR specialists, and is opening a workspace in Seattle’s Lower Queen Anne neighborhood. Proprio positions itself against incumbents such as Intuitive Surgical and emphasizes improving surgeon user experience through AR/VR and computer vision.