SAN FRANCISCO, Oct. 5, 2026 — Corgi Invest today announced the launch of the MN (Corgi MANGOS ETF), an actively managed exchange-traded fund seeking to provide investors exposure to OpenAI and Anthropic, two private companies in artificial intelligence, alongside Meta Platforms, NVIDIA, Alphabet (Google), and SpaceX. MANGOS is an acronym formed from the first letter of each of the six companies’ names.
MN
“For years, exposure to OpenAI and Anthropic has been reserved for venture investors and insiders,” said Jeff Weniger, Chief Investment Strategist, Corgi Invest. “MN puts that exposure inside a wrapper investors already know how to use a standard, exchange-traded fund, bought and sold like any other ETF, with no lockups and no accreditation requirement.” Shares may trade at a premium or discount to NAV and may have limited liquidity.
Targeted Access to OpenAI and Anthropic
OpenAI and Anthropic are companies in AI, yet neither is publicly traded, ordinarily putting them out of reach for everyday investors. MN is designed to close that gap: within a standard, exchange-traded structure, with no accreditation requirement and no private-fund lockup, the Fund seeks exposure to both companies through cash-settled total return swaps1, rather than direct share purchases.
These private-company swaps2 provide one-for-one exposure, with no leveraged return multiplier3, and are initially priced with reference to perpetual futures contracts4 linked to the respective companies. Combined exposure to OpenAI and Anthropic is limited to 15% of the Fund’s net assets at the time of investment, consistent with the Fund’s liquidity risk management program.
Fund Overview
Beyond its OpenAI and Anthropic exposure, the Fund seeks capital appreciation by investing, under normal market conditions, at least 80% of its net assets in equity securities of all six MANGOS companies and in financial instruments including total return swaps that provide economic exposure to those companies’ equity value.
The Fund offers this exposure through a standard exchange-traded structure, investors can buy and sell shares through a brokerage account like any other ETF. MN carries a total annual operating expense ratio of 0.20%.
About the MN Companies
The Fund seeks to give exposure to six companies: Anthropic, a privately held AI safety company and developer of the Claude family of AI models; OpenAI, a privately held developer of ChatGPT and frontier AI research; Meta Platforms (social media and virtual/augmented reality); NVIDIA (GPUs and AI computing infrastructure); Alphabet/Google (search, cloud, and AI); and SpaceX (launch services, Starlink, and, through its ownership of xAI and X, AI and social media). Portfolio weightings are determined through active management rather than index replication.
Availability
MN began trading on Cboe BZX Exchange on October 2, 2026, and is available through brokerage accounts nationwide.
About Corgi Invest
Corgi Strategies, LLC is an SEC-registered investment adviser founded in 2025. As of June 30, 2026, the firm managed approximately $821 million in assets. Corgi Invest builds actively managed exchange-traded funds designed to give everyday investors access to concentrated, high-conviction themes.
Investors should consider the investment objectives, risks, charges, and expenses of the Fund carefully before investing. This and other information is contained in the Fund’s prospectus, which should be read carefully before investing. Shares are bought and sold at market price, not NAV, and are not individually redeemable from the Fund.
Investing involves risk, including possible loss of principal. Anthropic and OpenAI are privately held companies for which substantially less public information is available; the Fund’s exposure to these companies through swaps involves counterparty, valuation, liquidity, and pricing risks, including reliance on perpetual futures reference pricing that may differ materially from the companies’ actual value, and is limited to 15% of net assets. The Fund is non-diversified and concentrates its investments in six companies and the related industries in which they operate, which may make the Fund more volatile than a more broadly diversified fund. The Fund is newly organized and has no operating history. See the prospectus for a complete description of principal risks.
This press release is not an offer to sell or a solicitation of an offer to buy shares of the Fund, and is not a prospectus. The Fund’s registration statement, including its prospectus, has been filed with the SEC: https://www.sec.gov/Archives/edgar/data/2078265/000207826526000415/cik0002078265-2026 0929.htm. Shares are not FDIC-insured, may lose value, and have no bank guarantee.
This release contains forward-looking statements regarding the Fund and the companies to which it has exposure. Actual results may differ materially from those expressed or implied.
Corgi ETF Trust I. Distributed by Paralel Distributors LLC, member FINRA.
Definitions
1. Total return swaps: Contracts under which the Fund receives an investment’s gains and income, pays its losses and typically pays financing costs, without owning the investment directly.
2. Private-company swaps: Total return swaps linked to companies whose shares are not publicly traded.
3. One-for-one exposure, with no leveraged return multiplier: The swap is designed to reflect the referenced investment’s gains or losses at a 1:1 rate, before fees and costs, without magnifying them. The Fund’s overall return may differ.
4. Perpetual futures contracts: Contracts with no fixed expiration date that use periodic payments between traders to help keep their prices aligned with a referenced asset. Their prices may differ from its actual value.
-Clockwork.io recauda 31 millones de dólares; LinkedIn, Together AI y WhiteFiber adoptan su software de resiliencia para evitar el desperdicio de horas de GPU
LinkedIn evita decenas de miles de horas de inactividad de la GPU mensualmente; las nuevas innovaciones de TorchPass preservan el progreso de la carga de trabajo de IA sin cambios de código y aceleran el aprendizaje por refuerzo.
PALO ALTO, California, 5 de octubre de 2026 — Clockwork.io, cuyo software de tolerancia a fallos mantiene en funcionamiento las cargas de trabajo de entrenamiento de IA, aprendizaje por refuerzo e inferencia ante fallos de infraestructura, anunció hoy una nueva financiación de 31 millones de dólares, despliegues en producción en LinkedIn y Together AI, y una mayor adopción por parte de WhiteFiber.
New TorchPass innovations preserve AI workload progress without code changes and speed reinforcement learning.
La empresa también presentó dos nuevas funcionalidades para su solución TorchPass, cada una de las cuales captura el estado de una tarea de IA distribuida en ejecución. Las instantáneas de plataforma multimodo, una primicia en el sector para la formación, guardan la tarea completa en todos los nodos sin necesidad de modificar el código de entrenamiento y la preservan para su posterior recuperación. Por su parte, los puntos de control de aplicación rápidos y asíncronos, que se realizan en segundo plano mientras la tarea está en curso, aceleran el aprendizaje por refuerzo. Estos suministran pesos del modelo actualizados a las réplicas de inferencia encargadas de generar los rollouts, los ejemplos a partir de los cuales aprende el modelo, lo que permite que dichas réplicas pasen menos tiempo esperando o trabajando con un modelo desactualizado.
La tolerancia a fallos se ha convertido en un requisito para la IA a gran escala
Las grandes cargas de trabajo de IA distribuida pueden abarcar miles de GPU que deben mantenerse sincronizadas: el fallo de una sola GPU, la interrupción de un enlace o el bloqueo de un servidor pueden paralizar la tarea por completo. Meta informó de interrupciones inesperadas con una frecuencia media aproximada de una cada tres horas durante un periodo de 54 días de entrenamiento de Llama 3 utilizando 16.384 GPU.
La respuesta habitual consiste en recargar un punto de control, es decir, una copia guardada del progreso de la tarea. El proceso de recuperación puede durar hasta 90 minutos, deja inactivas las GPU que funcionan correctamente y obliga a repetir el trabajo realizado desde dicho punto de control. Los clientes pagan por GPU ociosas y por cálculos redundantes, y los modelos tardan más en completarse. A medida que las tareas aumentan de tamaño, cada reinicio pone en riesgo una mayor cantidad de tiempo de uso de las GPU.
A esta escala, garantizar la continuidad del trabajo útil a pesar de los fallos es un requisito de infraestructura. Clockwork.io satisface esta necesidad mediante un conjunto de herramientas de tolerancia a fallos que los equipos de plataforma despliegan como una capa intermedia entre el hardware y la carga de trabajo. LinkPass redirige el tráfico para evitar el enlace averiado, de modo que la tarea no llega a detectar el fallo. TorchPass transfiere la carga de trabajo de una GPU defectuosa a una operativa, permitiendo que el entrenamiento continúe sin necesidad de retroceder a un estado anterior. Ambas soluciones están ya en fase de producción. La nueva funcionalidad de snapshots de plataforma de TorchPass, anunciada hoy, captura el estado de una tarea distribuida en ejecución; esto permite restaurar la tarea completa cuando el fallo es demasiado grave para solucionarlo mediante una simple migración.
“Los fallos son inevitables a la escala de la IA; perder horas de trabajo útil debido a ellos no debería ocurrir”, afirmó Suresh Vasudevan, consejero delegado de Clockwork.io. “La tolerancia a fallos actúa como un multiplicador del rendimiento útil: permite que las GPU sigan realizando trabajo productivo en lugar de esperar a procesos de recuperación o repetir tareas ya completadas. Desarrollamos nuestro software en colaboración con empresas y proveedores de nube que operan algunas de las mayores flotas de GPU, por lo que es capaz de gestionar los fallos que realmente experimentan. Esa protección debe formar parte de la infraestructura de la que dependen a diario dichas empresas y proveedores”.
La adopción se extiende a empresas, hiperescaladores y neoclouds
Las empresas que gestionan sus propias flotas de GPU, así como los hiperescaladores y los proveedores de “neocloud”, están adoptando Clockwork.io por la misma razón: logran que una mayor parte de sus horas de GPU se destine a trabajo útil.
LinkedIn ha implementado la tolerancia a fallos de red LinkPass en toda su flota de infraestructura de IA, evitando así la pérdida de decenas de miles de horas de GPU por inactividad cada mes.
“A escala de infraestructura de IA, un único problema de red nunca debería dejar fuera de servicio a GPUs que funcionan correctamente ni interrumpir las cargas de trabajo en ejecución. Antes de Clockwork.io, una oscilación en InfiniBand NIC podía retirar del servicio un servidor con ocho GPU, mientras que una oscilación en el puerto de un conmutador podía dejar inoperativo un segundo servidor, duplicando el impacto a 16 GPU”, afirmó Raghu Hiremagalur, vicepresidente sénior y director de tecnología de Infraestructura en LinkedIn. “Clockwork.io ayudó a transformar ese modelo operativo. Su tecnología de tolerancia a fallos de red redirige automáticamente el tráfico hacia rutas operativas, permitiendo que las tareas continúen sin interrupciones mientras se reparan los fallos en enlaces, componentes ópticos, cables o tarjetas de red. En conjunto, Clockwork.io evita decenas de miles de horas de inactividad de GPU al mes en toda nuestra flota. Al convertir lo que antes eran incidentes operativos disruptivos en eventos de mantenimiento manejables, Clockwork.io ha contribuido a mejorar la utilización de la infraestructura y la eficiencia operativa”.
Together AI lanza TorchPass al mercado como un servicio en sus clústeres de GPU. Durante la conferencia PyTorch, ambas empresas realizarán una demostración en vivo de una tarea de entrenamiento multinodo que continúa ejecutándose —sin necesidad de reinicio— a pesar de la introducción deliberada de fallos en la red y en las GPU.
“Nuestros clientes nos evalúan en función del goodput, es decir, la proporción de sus horas de GPU que realmente hacen avanzar el modelo”, afirmó Pavneet Ahluwalia, responsable de producto de Together AI. “La función de reparación de nodos ya detecta fallos y aprovisiona capacidad de sustitución de forma automática. Las soluciones TorchPass y LinkPass de Clockwork.io se basan en esa infraestructura y están diseñadas para mantener la ejecución de las tareas a pesar de los fallos de GPU y de enlace, preservando así el progreso realizado. Las estamos lanzando al mercado como la siguiente capa de resiliencia de la plataforma”.
WhiteFiber (NASDAQ: WYFI), cliente actual de la compañía, está ampliando el uso del software de Clockwork.io en su creciente infraestructura global de GPU como servicio.
“Someter a pruebas de estrés la fiabilidad de un clúster antes de que pase a producción es fundamental, ya que el cliente que recibe un sistema con un fallo oculto en la red de interconexión acaba pagando las consecuencias en forma de trabajos fallidos y horas de GPU perdidas”, afirmó Tom Sanfilippo, director de tecnología de WhiteFiber. “Componentes ópticos de calidad marginal, tarjetas de red (NIC) mal configuradas y enlaces que superan una prueba básica pero se degradan bajo carga pueden pasar desapercibidos. La auditoría automatizada de flotas de Clockwork.io valida simultáneamente cada enlace y nodo, localiza fallos en cuestión de minutos y nos permite corregirlos antes de la entrega. Ponemos los clústeres en marcha con mayor rapidez, de modo que la primera ejecución de entrenamiento del cliente se realiza sobre una red validada de extremo a extremo, y no simplemente sobre un sistema encendido. Dada la vertiginosa demanda del mercado, ofrecer a los clientes capacidad validada con rapidez es crucial para nuestro negocio; por eso estamos implementando Clockwork.io en todos nuestros clústeres”.
Las nuevas funcionalidades de TorchPass ponen la protección de las cargas de trabajo en manos de los equipos de plataforma
Clockwork.io ha ampliado las capacidades de TorchPass más allá de la migración de GPU, incorporando dos funciones de las que los equipos de plataforma carecían hasta ahora: la posibilidad de generar ellos mismos una instantánea de todo un trabajo distribuido y puntos de control de la aplicación lo suficientemente rápidos como para ejecutarse en segundo plano.
En el caso del entrenamiento, las instantáneas de la plataforma guardan el estado de ejecución de un trabajo en curso en todos sus nodos, permitiendo así restaurarlo tras una interrupción. Los equipos de plataforma e ingenieros de infraestructura de IA implementan esta solución para cargas de trabajo compatibles sin necesidad de que los propietarios de las aplicaciones modifiquen su código o añadan lógica de puntos de control. Los equipos empresariales pueden proteger los trabajos de entrenamiento en toda su flota mediante un único mecanismo, y los proveedores de nube pueden proteger los trabajos de los clientes cuyo código no controlan. Cuando los equipos implementan puntos de control a nivel de aplicación, la rapidez de los puntos de control de TorchPass permite realizarlos con mayor frecuencia; de este modo, se pierde menos progreso y se repite menos trabajo de cómputo tras un fallo.
En cuanto a la inferencia, los modelos de gran tamaño se ejecutan en dos o más servidores, por lo que un fallo en un enlace puede inutilizar una réplica completa e interrumpir una sesión de usuario o una tarea de agente en plena ejecución. LinkPass permite que dichas réplicas multiservidor sigan operativas a pesar de los fallos en los enlaces.
El aprendizaje por refuerzo depende tanto del entrenamiento como de la inferencia. Varias copias del modelo generan secuencias de ejecución, el sistema de entrenamiento aprende de ellas y los pesos actualizados deben llegar a las copias antes de que estas puedan generar resultados con la versión más reciente. Los puntos de control de aplicación de TorchPass transfieren los pesos actualizados a las réplicas de ejecución con mayor rapidez, mientras que LinkPass mantiene dichas réplicas en funcionamiento a pesar de los fallos en los enlaces.
“La tolerancia a fallos en clústeres solía ser un problema exclusivo del entrenamiento; ahora es también un problema de inferencia”, afirmó Dylan Patel, fundador, consejero delegado y analista principal de SemiAnalysis, empresa cuyas clasificaciones ClusterMAX evalúan a los proveedores de servicios en la nube basados en GPU. “En nuestras pruebas ClusterMAX, TorchPass reduce la pérdida de rendimiento efectivo durante el entrenamiento del 14 % a menos del 3 % en un proveedor neocloud con calificación “oro”. El aprendizaje por refuerzo (RL) vincula ambos procesos: las réplicas de inferencia generan simulaciones, el sistema de entrenamiento aprende de ellas y los pesos actualizados regresan a las réplicas. Clockwork.io permite que las réplicas sigan operando a pesar de las fluctuaciones de enlace y los fallos de red. Sus puntos de control ultrarrápidos aceleran la transferencia de pesos de vuelta a la flota de inferencia, evitando así que el proceso se detenga en cualquiera de las dos direcciones. La solución debe implementarse mediante una capa de tolerancia a fallos común para el entrenamiento, la inferencia y el aprendizaje por refuerzo”.
Los equipos de plataformas empresariales y los proveedores de servicios en la nube pueden ponerse en contacto con Clockwork.io para evaluar el software en sus cargas de trabajo o explorar oportunidades de colaboración.
La ronda de financiación y su destino
La ronda fue coliderada por Premji Invest, Wing Venture Capital y Seligman Ventures, con la participación de los inversores actuales NEA y e& Capital. Con esto, la financiación total de Clockwork.io alcanza los 73 millones de dólares. La empresa utilizará el capital para acelerar el despliegue de su conjunto de herramientas de tolerancia a fallos —aplicable a las fases de entrenamiento, inferencia y aprendizaje por refuerzo—, ampliar su adopción empresarial y escalar la entrega a través de socios en la nube.
“Lo único que escala a la perfección es la falta de fiabilidad: basta con instalar suficientes GPU en una máquina para que siempre haya algo fallando”, afirmó Greg Papadopoulos, socio de capital riesgo de NEA. “El método tradicional —detener la tarea y recargar un punto de control— carece de sentido a la escala actual. Clockwork considera el fallo como el estado habitual: TorchPass migra en tiempo real el entrenamiento desde una GPU que presenta errores y ahora permite capturar el estado de toda una tarea en ejecución sin necesidad de modificar el código. Ya está ahorrando decenas de miles de horas de GPU al mes. Apostamos por Clockwork.io en 2021 y estamos encantados de seguir apoyándolos mientras definen la capa de rendimiento de los clústeres de IA”.
Clockwork.io es pionera en Software-Driven AI Fabrics™, una capa programable situada entre el hardware y la carga de trabajo que permite que los clústeres de GPU sean observables, tolerantes a fallos y aprovechen al máximo su capacidad, independientemente del acelerador, la red o la nube utilizados. Las cargas de trabajo de IA requieren que todo el clúster funcione como una única máquina; sin embargo, los fallos y los cuellos de botella dejan las GPU inactivas. La plataforma FleetLens de Clockwork.io recupera esa capacidad perdida: gracias a una telemetría con precisión de nanosegundos, identifica exactamente qué GPU, nodo o enlace está ralentizando o bloqueando una tarea y valida los clústeres antes de su puesta en marcha. Además, mediante LinkPass y TorchPass, garantiza la continuidad de las cargas de trabajo —desde el entrenamiento hasta la inferencia— incluso ante fallos o degradaciones de la infraestructura. Según pruebas comparativas independientes realizadas por SemiAnalysis, TorchPass supera a los principales marcos de trabajo de código abierto y a los métodos tradicionales de punto de control y reinicio en cuanto a la velocidad de recuperación ante fallos. Empresas como LinkedIn, Together AI, WhiteFiber, Wells Fargo, Nebius, NScale y DCAI confían en Clockwork.io para impulsar su infraestructura de IA. Más información en www.clockwork.io.
Cette startup technologique permet aux particuliers et aux organisations de protéger leurs données, leurs œuvres et leur image, d’en octroyer des licences et d’en tirer une rémunération, tout en les mettant en relation avec la demande croissante de données humaines du monde réel dans les domaines de l’IA, de la recherche et des médias.
WILMINGTON, Delaware, 5 octobre 2026 — SignSplit PBC, une entreprise technologique en pleine croissance qui permet aux particuliers et aux organisations de protéger, de concéder sous licence et de partager leurs données, leurs œuvres et leur image, sort aujourd’hui du mode furtif avec un tour de table stratégique de 400 millions de dollars mené avec W Group, l’écosystème mondial de fintech et de technologie au service de plus de 40 millions d’utilisateurs. Cet engagement de financement, qui valorise SignSplit à un milliard de dollars, associe un apport en capitaux à un ensemble de ressources stratégiques sur plusieurs années, destinées à soutenir le déploiement mondial de SignSplit.
La prochaine génération d’IA s’appuiera de plus en plus sur les données, les connaissances, la créativité et l’expérience humaines issues du monde réel, dans des domaines allant des modèles avancés et de la robotique à la science, à la santé, aux médias et au divertissement. SignSplit permet aux particuliers et aux organisations de participer activement à cet avenir en protégeant, en concédant sous licence et en partageant leurs données, leurs œuvres et leur image, avec leur consentement, une provenance clairement établie et des conditions bien définies. SignSplit appelle cela des « données signées ».
Grâce à SignSplit, les entreprises spécialisées dans l’IA et la robotique, les chercheurs et d’autres organisations peuvent accéder à des bases de données et à des pools de recherche conçus pour répondre à des besoins spécifiques, tandis que les contributeurs fournissent des données issues du monde réel, leurs connaissances et leurs compétences, et reçoivent une rémunération en échange de leur participation. SignSplit intègre également une couche de vérification des données, du travail et de l’image, conçue pour permettre aux systèmes d’IA, aux plateformes de réseaux sociaux et à d’autres services numériques d’identifier les contenus signés et d’accéder aux informations associées concernant leur provenance, le consentement et les conditions d’utilisation.
« SignSplit propose des solutions pour un monde où la frontière entre l’expérience humaine et la technologie devient de plus en plus floue », déclare Alessandro Monterosso, cofondateur et CEO de SignSplit. « À mesure que l’IA s’implante davantage dans le monde réel, les connaissances, la créativité, les compétences et l’identité humaines seront de plus en plus liées à la manière dont la technologie évolue et interagit avec le monde qui nous entoure. Nous pensons que SignSplit et l’utilisation de données signées peuvent servir de pont entre la technologie et le monde réel, en associant la contribution humaine à la protection, à la participation et à la rémunération à mesure que la technologie progresse. »
« SignSplit établit une nouvelle norme en matière d’utilisation des données, des travaux et de l’image des particuliers et des organisations », souligne Volodymyr Nosov, fondateur et président de W Group. « Nous sommes convaincus que tous les acteurs sérieux du secteur de l’IA et de l’économie des données finiront par s’appuyer sur l’infrastructure de SignSplit. L’ampleur de notre engagement reflète à la fois notre conviction que nous en sommes encore aux prémices de cette évolution, et l’importance que nous accordons à ces solutions pour tous les secteurs d’activité qui s’appuient sur les données humaines. »
SignSplit a été fondé en 2024 par Glib Denisov, cofondateur, président exécutif et directeur des produits. « Lorsque nous avons lancé SignSplit en 2024, nous visions là où nous pensions que l’IA se dirigeait, pas là où elle se trouvait à l’époque », déclare Glib Denisov. « Nous avons constaté que la prochaine génération d’IA aurait besoin de bien plus que de simples données supplémentaires : elle aurait besoin de données issues du monde réel, de meilleure qualité, dont la provenance, les autorisations et les droits seraient clairement établis. »
À propos de SignSplit
SignSplit PBC est une entreprise technologique qui met en place l’infrastructure nécessaire à l’économie des données signées. Fondé en 2024 et constitué en tant que Delaware Public Benefit Corporation, SignSplit met en relation des particuliers et des institutions avec des entreprises spécialisées dans l’intelligence artificielle et la robotique, des chercheurs et des plateformes médiatiques grâce à sa plateforme de données signées. L’entreprise fournit une infrastructure permettant de protéger, d’octroyer des licences et de partager des données humaines issues du monde réel ainsi que d’autres formes de contributions humaines, avec un consentement, une provenance clairement établie et des conditions bien définies.
Pour en savoir plus et vous inscrire à SignSplit, rendez-vous sur signsplit.com.
À propos de W Group
W Group est un écosystème mondial d’entreprises qui façonnent l’avenir grâce à la technologie, à l’innovation et à des produits ambitieux. Aujourd’hui, W Group regroupe 11 entreprises, plus de 1 500 collaborateurs répartis sur 15 sites à travers le monde, et plus de 40 millions de clients dans 150 pays. Pour plus d’informations, rendez-vous sur https://www.w.group/.
Déclarations prospectives
Le présent communiqué de presse contient des déclarations prospectives concernant les projets de SignSplit, sa relation stratégique avec W Group et les évolutions attendues du marché. Ces déclarations reposent sur les prévisions actuelles et sont soumises à des risques et incertitudes ; les résultats réels peuvent s’en écarter sensiblement, et aucune des deux sociétés ne s’engage à les mettre à jour. Aucune information contenue dans le présent communiqué de presse ne constitue une offre ou une sollicitation concernant un titre ou un actif numérique.
New book by Hetz Ventures Managing Partner and former IDF intelligence officer maps actionable proposals for Israel’s defense, energy, education, governance and Jewish future
NEW YORK, Oct. 5, 2026 — Israel has repeatedly achieved what once seemed impossible. In Moonshots for Israel: 12 Big Ideas for a Small Nation, published today by Wicked Son Books, venture capitalist Judah Taub argues that Israel must move beyond incremental fixes and apply its entrepreneurial ingenuity to the nation’s most consequential challenges.
Published by Wicked Son Books / October 5, 2026 / Available at Amazon and wherever books are sold
Judah Taub, Venture Capitalist and author of Moonshots for Israel: 12 Big Ideas for a Small Nation
With a foreword by Start-Up Nation co-author and podcaster Dan Senor, Moonshots for Israel presents 12 ambitious proposals spanning defense, diplomacy, energy, education, healthcare, governance, demography and Israel-Diaspora relations. Each represents a genuine step change, carries the potential to create approximately $100 billion in value for Israel—more than $1 trillion collectively—and has been stress-tested with experts in the field.
“The State of Israel itself was a moonshot: an improbable idea made real through courage, ingenuity and relentless execution,” Taub said. “We are at a turning point. The human capital ready for meaningful action is enormous, and the desire to leap ahead is only growing. These are not proposals for academic discussion, but for action—each capable of dramatically changing the trajectory of Israel and the world.”
Taub’s moonshots include:
The Abraham Canal: A sea-to-sea commercial corridor linking the Red Sea and Mediterranean, positioning Israel as an indispensable regional crossroads.
The Eighth Battlefield: A new model for the information war, using real-time data, sentiment analysis and measurable outcomes instead of conventional public diplomacy.
An Iron Dome for Defense Exports: A strategy to expand Israel’s defense industry until embargoes become economically and strategically costly for the countries considering them.
A Fund to Reverse Jewish Demographic Decline: A plan to reduce the cost of observant Jewish life (“the Jewish tax”) and strengthen Diaspora communities for generations to come.
From Water Scarcity to Water Superpower: A plan to export desalinated Israeli water across the Middle East, transforming scarcity into economic opportunity, regional cooperation and strategic influence.
Cheap Energy, Strong Hand: A strategy to turn Israel’s natural gas, solar resources and technology sector into lasting energy abundance and geopolitical leverage.
Taub is founder and managing partner of Hetz Ventures, an Israeli early-stage venture-capital firm focused on cybersecurity, data and AI infrastructure, with nearly $500 million under management. A former IDF intelligence officer who received the IDF Creative Thinking Award, he serves in the reserves at the intersection of AI and national security and lectures at Wharton, Yale and Harvard Business School on creative thinking, problem-solving and AI-driven innovation. His writing has appeared in The Wall Street Journal, Semafor, Newsweek and Haaretz.
Moonshots for Israel builds on Taub’s bestselling Hebrew book, HaPotential HaYisraeli, and applies at a national scale the “local maximum” framework introduced in his first English-language book, How to Move Up When the Only Way Is Down.
President Isaac Herzog has publicly cited and discussed ideas from Moonshots for Israel, calling Israel’s unrealized potential “tremendous” and saying, “It is crucial to catch this wave.”
Publication Information
Title:Moonshots for Israel: 12 Big Ideas for a Small Nation Author: Judah Taub Foreword: Dan Senor Publication date: October 5, 2026 Publisher: Wicked Son Books ISBN: 979-8904670238 Formats: Hardcover, trade paperback, eBook and audiobook Availability: Wherever books are sold
The investment firm’s Founder and CEO, Louis Hernandez, Jr., will also share his global perspective on how leaders can navigate the rapidly evolving digital landscape by delivering a keynote speech and driving discussions during a closed-door credit union executive roundtable.
BOCA RATON, Fla., Oct. 5, 2026 — Black Dragon Capital℠ (“Black Dragon”) — a multi-phased global investment partner which focuses on advanced technology that addresses the most significant faced by the communities it serves — will have a strong presence at the upcoming Money 20/20, happening from Oct. 18-21, 2026, in Las Vegas.
The firm will be supporting key portfolio companies in attendance and speaking at multiple events. Founder and CEO, Louis Hernandez, Jr., will participate in a private credit union executive roundtable hosted by Mitchell Stankovic & Associates and joined by some of the most recognized leaders in the credit union and banking industries. Black Dragon’s leadership team will be available for meetings with entrepreneurs and partners throughout the event.
Black Dragon Capital℠ was founded over a decade ago by recognized financial technology leaders focused on bringing innovative technologies to high-growth industry segments affected by digital disruption. The firm is led by a collaborative team which combines deep operational experience with a focus on community impact and experience supporting the growth and development of companies. Its Founder and CEO, Louis Hernandez, Jr., also founded For A Bright Future Foundation, a 501(c)(3) nonprofit dedicated to supporting the needs of underserved children through education, healthcare, the arts, and youth leadership development.
The firm has launched the Black Dragon Capital℠ FinTech Limited Partner Advisory Board, an organization comprised of recognized credit union and CUSO investment partners which focuses on collaboration, innovation and community engagement between innovative, future-forward credit unions and CUSOs in the industry and the intense operating focus and domain expertise of Black Dragon Capital. The collaboration seeks to build successful FinTech organizations that solve the most significant issues in the industry, creating sustainable long-term businesses, and support sustainable growth and innovation within the credit union and FinTech ecosystem.
Gaining Perspective from Tech Visionary Louis Hernandez, Jr. During this year’s Money 20/20, Mitchell Stankovic & Associates will be hosting a closed-door credit union executive roundtable which intends to move beyond a credit union-only conversation by examining how leaders are making decisions as financial services, technology, capital, and consumer expectations continue to converge.
Louis Hernandez, Jr., Founder, Managing Partner, CEO, and major investor of Black Dragon Capital℠ will be delivering a keynote speech before joining Sue Mitchell, CEO of Mitchell Stankovic & Associates, and Scarlett Sieber, EVP of Money20/20 and Informa, in facilitating the discussion at this event.
Mr. Hernandez plays a key role in shaping Black Dragon’s investments and direction, particularly in financial technology, sports and media technology, and digital commerce. He has over 30 years of founding and leading multiple companies that have transformed the media, finance, and commerce industries, including Open Solutions, Grass Valley and Avid. He has also founded several companies including the sole founder and original investor in Payveris, a money movement platform.
Having been inducted into the Innovator Leadership Hall of Fame by Banking Technology News and recently awarded the Visionary Leadership Award by the International Trade Association for Broadcast and Media during the 2025 IABMTV Impact1 Awards, he is a sought-after speaker and advisor. Mr. Hernandez is also a prolific author with several published works. He has written several books including Too Small to Fail (2010), Saving the American Dream (2013), The Storyteller’s Dilemma (2017), and his latest book, Digital Tsunami (2025).
Mr. Hernandez’ leadership at Black Dragon Capital℠ has influenced the firm and its portfolio companies to embrace a unique, operationally led approach which combines deep domain expertise with the ability to transform and grow companies. Like most of the leadership team Mr. Hernandez has been not simply a participant, but the founder, and leader of several technology companies, and has advised boards of directors of some of the most recognizable companies in the world, such as HSBC, Infosys, and Edison.
“I am extremely honored to join Sue Mitchell, Scarlett Sieber, and the credit union leaders attending this executive round table. Digital transformation is reshaping financial services at an unprecedented pace, making the ability to innovate and adapt more critical than ever. At Black Dragon Capital, we have built our approach around identifying and advancing technologies that transform businesses, strengthen financial institutions and position them for long-term growth. As financial services, technology, capital, and consumer expectations continue to converge, we believe that our collaboration with the leaders of the credit union industry can unlock new technologies that can positively impact the community, support innovation and business development, and remain sustainable for years to come,” said Louis Hernandez, Jr., Founder, Managing Partner, and CEO of Black Dragon Capital℠.
Exploring Industry-Transforming FinTech Collaborations with Black Dragon Capital℠ In line with the firm’s dedication to address the unique needs of the credit union industry, the Black Dragon Capital℠ leadership team will also be present at Money 20/20 to discuss strategic partnerships with the intention to build successful FinTech organizations that can solve the most significant issues in the industry, supporting the development of sustainable businesses, and support innovation and sustainable growth within the industry.
1 – Mr. Hernandez was nominated for these awards and was not required to pay any nomination, award fee or any other fees to receive these accolades.
About Black Dragon Capital℠ Black Dragon Capital℠, founded over a decade ago by recognized financial technology leaders, is a multi-phased investment firm focused on disruptive technologies in high growth industry segments that strengthen economic stability within communities. The firm is led by a collaborative team with a combination of intense operating experience, community impact focus, and a proven track record of supporting companies through periods of growth and development.
About the Black Dragon Capital℠ FinTech Limited Partner Advisory Board The FinTech Limited Partner Advisory Board (LPAB) is an organization comprised of recognized credit union and CUSO investment partners. The Advisory Board provides input on industry issues, feedback on operational and technical areas with potential and current investments and provides strategic advice and direction on all Black Dragon Capital’s investments for Fintech strategies. The Advisory Board’s feedback incorporates insight from Black Dragon’s Global Advisory Group, thought leadership and guidance on key events.
About For A Bright Future Foundation
Louis Hernandez Jr.’s Foundation For A Bright Future is a 501(c)(3) not-for-profit organization dedicated to supporting the needs of underrepresented and underprivileged children through education, healthcare, the arts, and youth leadership development. Our initiatives provide equal opportunity for all children to have the tools and opportunities to fulfill their life goals and become constructive members of our global community. Learn more at forabrightfuturefoundation.org
Media Contact or Inquiries: Viviana Lauschus Marketing Manager, Black Dragon Capital [email protected]
Aren Wong Social Media Manager, Black Dragon Capital [email protected]
IMPORTANT DISCLOSURE
Black Dragon Capital Investment Management, LLC is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration with the SEC does not imply a certain level of skill or training.
LOS ANGELES, Oct. 5, 2026 — Xavier Ngwala and Xander Ngwala, who brand themselves as the “Silicon Beach Brothers,” are the co-founders of Noteese (pronounced “Notice”), a Silicon Beach-inspired, AI-powered platform designed to help people capture, preserve, and communicate their achievements.
Harvard Business School Foundry – Xavier Ngwala (left) and Xander Ngwala (right)
Noteese Mobile View – Xander Ngwala (left) and Xavier Ngwala (right)
Based in Los Angeles, the brothers are developing Noteese while building a presence within the Silicon Beach entrepreneurial and technology ecosystem. Their work recently reached an important milestone with their acceptance into Harvard Business School Foundry for Noteese. Noteese was created around a simple idea: achievements should not disappear. The platform is designed to help users document accomplishments, connect supporting evidence to those achievements, and build a lasting Achievement Record that can be used when future opportunities arise.
For students and young professionals, those accomplishments may include academics, leadership, athletics, certifications, internships, volunteer service, entrepreneurship, creative work, and other milestones that are often scattered across documents, photographs, emails, and different platforms.
“Noteese is about helping people preserve the value of what they have accomplished and making those achievements easier to use when opportunities arise,” said the Ngwala brothers.
Acceptance into Harvard Business School Foundry gives Xavier and Xander an opportunity to continue refining the platform, testing ideas, learning from other founders, and strengthening their experience in technology, product development, and entrepreneurship.
For the brothers, the milestone is also part of a broader effort to establish Noteese within the Los Angeles startup community and continue building a platform intended to help users turn accomplishments into future opportunities.
The Silicon Beach Brothers are still early in their journey, but their focus is clear: continue developing Noteese, grow its presence within the Silicon Beach ecosystem, and create technology that helps people better recognize and communicate what they have achieved.
About Noteese Noteese is an AI-powered achievement platform designed to help users capture, organize, preserve, and communicate their accomplishments through a lasting Achievement Record.
Media Contact Nakia T. Whitley-Ngwala Genet Press Phone: (888) 900-9071 Email: [email protected]
LinkedIn verhindert monatlich Zehntausende von GPU-Stunden an Ausfallzeiten; neue TorchPass-Innovationen sichern den Fortschritt bei KI-Workloads ohne Codeänderungen und beschleunigen das verstärkende Lernen.
PALO ALTO, Kalifornien, 5. Oktober 2026 — Clockwork.io, dessen fehlertolerante Software dafür sorgt, dass KI-Trainings, verstärktes Lernen und Inferenz-Workloads auch bei Infrastrukturausfällen weiterlaufen, gab heute eine neue Finanzierung in Höhe von 31 Millionen US-Dollar, den produktiven Einsatz bei LinkedIn und Together AI sowie die erweiterte Nutzung durch WhiteFiber bekannt.
New TorchPass innovations preserve AI workload progress without code changes and speed reinforcement learning.
Das Unternehmen stellte außerdem zwei neue Funktionen für seine TorchPass-Lösung vor, die jeweils den Status eines laufenden verteilten KI-Jobs erfassen. Multi-Node-Plattform-Snapshots – eine Branchenneuheit für das Training – speichern einen gesamten laufenden Job über alle Knoten hinweg ohne Änderungen am Trainingscode und bewahren ihn für die Wiederherstellung auf. Schnelle, asynchrone Anwendungs-Checkpoints, die im Hintergrund während der Ausführung des Jobs erstellt werden, beschleunigen das verstärkende Lernen. Sie liefern aktualisierte Modellgewichte an die Inferenz-Replikate, die „Rollouts” generieren – also die Beispiele, aus denen das Modell lernt –, sodass diese Replikate weniger Zeit damit verbringen, zu warten oder mit einem veralteten Modell zu arbeiten.
Fehlertoleranz ist zu einer Voraussetzung für KI in großem Maßstab geworden
Große verteilte KI-Workloads können sich über Tausende von GPUs erstrecken, die synchron bleiben müssen: Eine ausgefallene GPU, eine unterbrochene Verbindung oder ein eingefrorener Server kann den gesamten Job zum Stillstand bringen. Meta berichtete von unerwarteten Unterbrechungen im Durchschnitt von etwa einer alle drei Stunden während eines 54-tägigen Zeitraums des Llama-3-Trainings auf 16.384 GPUs.
Die übliche Reaktion besteht darin, einen Checkpoint – eine gespeicherte Kopie des Fortschritts des Jobs – neu zu laden. Die Wiederherstellung kann bis zu 90 Minuten dauern, lässt funktionsfähige GPUs im Leerlauf warten und erfordert, dass der Job die seit diesem Checkpoint abgeschlossenen Arbeitsschritte wiederholt. Kunden zahlen für im Leerlauf befindliche GPUs und wiederholte Berechnungen, und die Fertigstellung der Modelle dauert länger. Mit zunehmender Größe der Jobs gefährdet jeder Neustart mehr GPU-Zeit.
In dieser Größenordnung ist es eine Anforderung an die Infrastruktur, dass nützliche Arbeit trotz Ausfällen weiterläuft. Clockwork.io erfüllt diese Anforderung mit einer Suite zur Fehlertoleranz, die Plattformteams als Schicht zwischen der Hardware und der Arbeitslast bereitstellen. LinkPass leitet den Datenverkehr um eine ausgefallene Verbindung herum um, sodass der Job den Fehler gar nicht bemerkt. TorchPass verlagert die Arbeit von einer ausfallenden GPU auf eine funktionsfähige, sodass das Training fortgesetzt wird, anstatt zurückgesetzt zu werden. Beide Lösungen sind bereits im Einsatz. Die heute angekündigten neuen Plattform-Snapshots von TorchPass erfassen den Zustand eines laufenden verteilten Jobs, sodass der gesamte Job wiederhergestellt werden kann, wenn ein Ausfall zu groß ist, um ihn zu umgehen.
„Ausfälle sind im KI-Maßstab unvermeidlich. Der Verlust von Stunden nützlicher Arbeit durch diese Ausfälle sollte es jedoch nicht sein”, sagte Suresh Vasudevan, CEO von Clockwork.io. „Fehlertoleranz ist ein Goodput-Multiplikator: Sie sorgt dafür, dass GPUs weiterhin nützliche Arbeit verrichten, anstatt auf die Wiederherstellung zu warten oder bereits erledigte Arbeit zu wiederholen. Wir haben unsere Software gemeinsam mit Unternehmen und Cloud-Anbietern entwickelt, die einige der größten GPU-Flotten betreiben, sodass sie die Ausfälle bewältigt, mit denen diese tatsächlich konfrontiert sind. Dieser Schutz gehört in die Infrastruktur, auf die sich Unternehmen und Cloud-Anbieter täglich verlassen.”
Die Nutzung breitet sich in Unternehmen, bei Hyperscalern und Neoclouds aus
Unternehmen, die ihre eigenen GPU-Flotten betreiben, Hyperscaler und Neoclouds setzen Clockwork.io aus demselben Grund ein: Ein größerer Teil ihrer GPU-Stunden fließt in nützliche Arbeit.
LinkedIn hat die LinkPass-Netzwerkfehlertoleranz in seiner gesamten KI-Infrastrukturflotte implementiert und verhindert damit jeden Monat Ausfallzeiten im Umfang von Zehntausenden von GPU-Stunden.
„Bei einer KI-Infrastruktur dieser Größenordnung sollte ein einzelnes Netzwerkproblem niemals funktionierende GPUs außer Betrieb setzen oder laufende Workloads unterbrechen. Vor der Einführung von Clockwork.io konnte ein einziger Ausfall einer InfiniBand-Netzwerkkarte einen Server mit acht GPUs außer Betrieb setzen, während ein Ausfall eines Switch-Ports einen zweiten Server lahmlegen und die Auswirkungen auf 16 GPUs verdoppeln konnte”, sagte Raghu Hiremagalur, SVP und CTO Infrastructure bei LinkedIn. „Clockwork.io hat dazu beigetragen, dieses Betriebsmodell zu transformieren. Die Netzwerk-Fehlertoleranztechnologie leitet den Datenverkehr automatisch auf intakte Pfade um, sodass Jobs unterbrechungsfrei weiterlaufen können, während Fehler an Verbindungen, optischen Komponenten, Kabeln oder Netzwerkkarten behoben werden. Insgesamt verhindert Clockwork.io in unserer gesamten Flotte jeden Monat Ausfallzeiten im Umfang von Zehntausenden von GPU-Stunden. Indem Clockwork.io ehemals störende Betriebsvorfälle in beherrschbare Wartungsereignisse umwandelt, hat es dazu beigetragen, die Infrastrukturauslastung und die betriebliche Effizienz zu verbessern.”
Together AI bringt TorchPass als Service auf seinen GPU-Clustern auf den Markt. Auf der PyTorch-Konferenz werden die beiden Unternehmen einen Live-Trainingsjob mit mehreren Knoten demonstrieren, der trotz simulierter Netzwerk- und GPU-Ausfälle ohne Neustart fortgesetzt wird.
„Unsere Kunden bewerten uns anhand des Goodput, also des Anteils ihrer GPU-Stunden, der das Modell tatsächlich voranbringt”, sagte Pavneet Ahluwalia, Produktleiter bei Together AI. „Node Repair erkennt bereits Fehler und stellt automatisch Ersatzkapazität bereit. TorchPass und LinkPass von Clockwork.io bauen auf dieser Grundlage auf und sind darauf ausgelegt, Jobs trotz GPU-Ausfällen und Verbindungsfehlern weiterlaufen zu lassen und so den Fortschritt zu bewahren. Wir bringen sie als nächste Ebene der Ausfallsicherheit auf der Plattform auf den Markt.”
WhiteFiber (NASDAQ: WYFI), ein bestehender Kunde, weitet den Einsatz der Clockwork.io-Software auf seine wachsende globale GPU-as-a-Service-Infrastruktur aus.
„Es ist entscheidend, die Zuverlässigkeit eines Clusters vor der Inbetriebnahme auf Herz und Nieren zu prüfen, denn ein Kunde, der einen versteckten Fabric-Fehler erbt, zahlt später dafür in Form von fehlgeschlagenen Jobs und verlorenen GPU-Stunden”, sagte Tom Sanfilippo, Chief Technology Officer bei WhiteFiber. „Optiken mit Grenzwerten, falsch konfigurierte Netzwerkkarten und Verbindungen, die zwar einen Basistest bestehen, aber unter Last an Leistung verlieren, können durchrutschen. Die automatisierte Flottenprüfung von Clockwork.io validiert jede Verbindung und jeden Knoten auf einmal, lokalisiert Fehler innerhalb von Minuten und ermöglicht es uns, diese vor der Abnahme zu beheben. Wir nehmen Cluster schneller in Betrieb, und der erste Trainingslauf eines Kunden läuft auf einer Fabric, die durchgängig validiert wurde – und nicht nur eingeschaltet ist. Angesichts der rasant wachsenden Marktnachfrage ist es für unser Geschäft entscheidend, validierte Kapazitäten schnell an Kunden bereitzustellen, und deshalb erweitern wir den Einsatz von Clockwork.io auf alle unsere Cluster.”
Neue TorchPass-Funktionen legen den Schutz von Workloads in die Hände der Plattformteams
Clockwork.io hat TorchPass über die GPU-Migration hinaus um zwei Funktionen erweitert, über die Plattformteams bisher nicht verfügten: einen Snapshot eines gesamten verteilten Jobs, den sie selbst erstellen können, und Anwendungs-Checkpoints, die schnell genug sind, um im Hintergrund ausgeführt zu werden.
Beim Training speichern Plattform-Snapshots den Ausführungsstatus eines laufenden Jobs über alle seine Knoten hinweg, sodass der Job nach einer Unterbrechung wiederhergestellt werden kann. Plattformteams und KI-Infrastrukturingenieure setzen dies für unterstützte Workloads ein, ohne darauf warten zu müssen, dass Anwendungsbesitzer ihren Code ändern oder Checkpoint-Logik hinzufügen. Unternehmensteams können Trainingsjobs in ihrer gesamten Flotte mit einem einzigen Mechanismus schützen, und Cloud-Anbieter können Kundenjobs schützen, über deren Code sie keine Kontrolle haben. Wenn Teams Checkpoints auf Anwendungsebene erstellen, können die schnellen Checkpoints von TorchPass häufiger erstellt werden, sodass nach einem Ausfall weniger Fortschritt verloren geht und weniger Rechenaufwand wiederholt werden muss.
Bei der Inferenz laufen große Modelle auf zwei oder mehr Servern, sodass eine einzige fehlerhafte Verbindung eine gesamte Replik lahmlegen und eine Benutzersitzung oder eine Agenten-Aufgabe mitten im Ablauf unterbrechen kann. LinkPass sorgt dafür, dass diese Repliken auf mehreren Servern auch bei Verbindungsausfällen weiterarbeiten.
Verstärkendes Lernen hängt sowohl vom Training als auch von der Inferenz ab. Kopien des Modells generieren Rollouts, der Trainer lernt daraus, und aktualisierte Gewichte müssen die Kopien erreichen, bevor diese mit der neuesten Version generieren können. Die Anwendungs-Checkpoints von TorchPass übertragen die aktualisierten Gewichte schneller an die Rollout-Replikate, während LinkPass dafür sorgt, dass diese Replikate auch bei Verbindungsausfällen weiterarbeiten.
„Die Fehlertoleranz von Clustern war früher ein Problem beim Training. Mittlerweile ist sie auch ein Problem bei der Inferenz”, sagte Dylan Patel, Gründer, CEO und Chefanalyst bei SemiAnalysis, dessen ClusterMAX-Bewertungen GPU-Cloud-Anbieter benchmarken. „In unserem ClusterMAX reduziert TorchPass den Goodput-Verlust beim Training für eine Neocloud mit Gold-Rating von 14 % auf unter 3 %. Reinforcement Learning (RL) verbindet beide Aspekte miteinander: Inferenz-Replikate generieren Rollouts, der Trainer lernt daraus, und die aktualisierten Gewichte werden an die Replikate zurückgesendet. Clockwork.io sorgt dafür, dass die Replikate auch bei Verbindungsausfällen und Netzwerkstörungen weiterlaufen. Seine extrem schnellen Checkpoints beschleunigen die Rückübertragung der Gewichte in die Rollout-Flotte, sodass der Lauf in keiner Richtung ins Stocken gerät. Eine einzige Fehlertoleranzschicht für Training, Inferenz und RL – genau hier muss die Lösung liegen.”
Teams von Unternehmensplattformen und Cloud-Anbieter können sich an Clockwork.io wenden, um die Software für ihre Workloads zu evaluieren oder Möglichkeiten für eine Partnerschaft zu erkunden.
Die Finanzierungsrunde und was damit finanziert wird
Die Runde wurde gemeinsam von Premji Invest, Wing Venture Capital und Seligman Ventures angeführt, unter Beteiligung der bestehenden Investoren NEA und e& Capital. Damit beläuft sich die Gesamtfinanzierung von Clockwork.io auf 73 Millionen US-Dollar. Das Unternehmen wird das Kapital nutzen, um die Einführung seiner Suite für Fehlertoleranz in den Bereichen Training, Inferenz und bestärkendes Lernen zu beschleunigen, die Akzeptanz in Unternehmen zu steigern und die Bereitstellung über Cloud-Partner auszuweiten.
„Das Einzige, was sich perfekt skalieren lässt, ist Unzuverlässigkeit: Steckt man genug GPUs in einen Rechner, fällt immer irgendetwas aus”, sagte Greg Papadopoulos, Venture Partner bei NEA. „Das alte Vorgehen – den Job stoppen und einen Checkpoint neu laden – macht bei den heutigen Größenordnungen keinen Sinn mehr. Clockwork behandelt Ausfälle als Normalzustand: TorchPass verlagert das Training live von einer ausgefallenen GPU und erstellt nun Snapshots eines gesamten laufenden Jobs ohne Codeänderungen. Dadurch werden bereits Zehntausende von GPU-Stunden pro Monat eingespart. Wir haben Clockwork.io erstmals 2021 finanziell unterstützt und freuen uns sehr, das Unternehmen weiterhin zu fördern, während es die Leistungsschicht des KI-Clusters definiert.”
Clockwork.io ist Vorreiter bei Software-Driven AI Fabrics™, einer programmierbaren Schicht zwischen Hardware und Workload, die GPU-Cluster beobachtbar, fehlertolerant und über jeden Beschleuniger, jedes Netzwerk und jede Cloud hinweg voll auslastet. KI-Workloads erfordern, dass der gesamte Cluster als eine einzige Maschine agiert, doch Ausfälle und Engpässe führen dazu, dass GPUs ungenutzt bleiben. Die FleetLens”-Plattform von Clockwork.io gewinnt diese verlorene Kapazität zurück: Nanosekundengenaue Telemetrie lokalisiert die GPU, den Knoten oder die Verbindung, die einen Job verlangsamt oder zum Stillstand bringt, validiert Cluster vor dem Start und sorgt mit „LinkPass” und „TorchPass” dafür, dass Workloads trotz Infrastrukturausfällen und Leistungseinbußen weiterlaufen – vom Training bis zur Inferenz. SemiAnalysis hat in unabhängigen Benchmarks festgestellt, dass TorchPass Ausfälle schneller behebt als Checkpoint-Restart und führende Open-Source-Frameworks. LinkedIn, Together AI, WhiteFiber, Wells Fargo, Nebius, NScale und DCAI vertrauen auf Clockwork.io als Grundlage für ihre KI-Infrastruktur. Weitere Informationen finden Sie unterwww.clockwork.io.
LinkedIn prevents tens of thousands of GPU-hours of downtime monthly; new TorchPass innovations preserve AI workload progress without code changes and speed reinforcement learning.
PALO ALTO, Calif., Oct. 5, 2026 — Clockwork.io, whose fault-tolerance software keeps AI training, reinforcement learning and inference workloads running through infrastructure failures, today announced $31 million in new funding, production deployments at LinkedIn and Together AI, and expanded adoption by WhiteFiber.
New TorchPass innovations preserve AI workload progress without code changes and speed reinforcement learning.
The company also introduced two new capabilities for its TorchPass solution, each capturing the state of a running distributed AI job. Multi-node platform snapshots, an industry first for training, save an entire running job across every node without changes to the training code and preserve it for recovery. Fast, asynchronous application checkpoints, taken in the background while the job runs, accelerate reinforcement learning. They deliver updated model weights to the inference replicas that generate rollouts, the examples the model learns from, so those replicas spend less time waiting or working from a stale model.
Fault tolerance has become a requirement for AI at scale
Large distributed AI workloads can span thousands of GPUs that must stay in sync: one failed GPU, dropped link or frozen server can stall the entire job. Meta reported unexpected interruptions averaging roughly one every three hours during a 54-day period of Llama 3 training on 16,384 GPUs.
The typical response is to reload a checkpoint, a saved copy of the job’s progress. Recovery can take up to 90 minutes, leaves healthy GPUs waiting, and requires the job to repeat work completed since that checkpoint. Customers pay for idle GPUs and repeated computation, and models take longer to complete. As jobs grow, each restart puts more GPU time at risk.
At this scale, keeping useful work running through failures is an infrastructure requirement. Clockwork.io meets it with a fault-tolerance suite that platform teams deploy as a layer between the hardware and the workload. LinkPass reroutes traffic around a failed link so the job never sees the fault. TorchPass moves work from a failing GPU to a healthy one so training continues instead of rolling back. Both are in production. TorchPass’s new platform snapshots, announced today, capture the state of a running distributed job so the whole job can be restored when a failure is too large to migrate around.
“Failures are inevitable at AI scale. Losing hours of useful work to them should not be,” said Suresh Vasudevan, CEO of Clockwork.io. “Fault tolerance is a goodput multiplier: it keeps GPUs doing useful work instead of waiting for recovery or repeating work already done. We built our software alongside enterprises and cloud providers operating some of the largest GPU fleets, so it handles the failures they actually see. That protection belongs in the infrastructure enterprises and cloud providers rely on every day.”
Adoption expands across enterprises, hyperscalers and neoclouds
Enterprises running their own GPU fleets, hyperscalers and neoclouds are adopting Clockwork.io for the same reason: more of their GPU-hours go to useful work.
LinkedIn has deployed LinkPass network fault tolerance across its AI infrastructure fleet and prevents tens of thousands of GPU-hours of downtime each month.
“At AI infrastructure scale, a single network issue should never sideline healthy GPUs or interrupt running workloads. Before Clockwork.io, one InfiniBand NIC flap could remove an eight-GPU server from service, while a switch port flap could drain a second server, doubling the impact to 16 GPUs,” said Raghu Hiremagalur, SVP, CTO Infrastructure, LinkedIn. “Clockwork.io helped transform that operating model. Its network fault-tolerance technology automatically reroutes traffic onto healthy paths, allowing jobs to continue uninterrupted while link, optic, cable, or NIC faults are repaired. In aggregate, Clockwork.io prevents tens of thousands of GPU-hours of downtime per month across our fleet. By turning what were once disruptive operational incidents into manageable maintenance events, Clockwork.io has helped improve infrastructure utilization and operational efficiency.”
Together AI is bringing TorchPass to market as a service on its GPU Clusters. At the PyTorch Conference, the two companies will demonstrate a live multi-node training job continuing through injected network and GPU failures without restarting.
“Our customers grade us on goodput, the share of their GPU-hours that actually move the model forward,” said Pavneet Ahluwalia, Product Lead, Together AI. “Node repair already detects faults and provisions replacement capacity automatically. Clockwork.io’s TorchPass and LinkPass build on that foundation and are designed to keep jobs moving through GPU faults and link failures, preserving progress. We are bringing them to market as the next layer of resilience in the platform.”
WhiteFiber (NASDAQ: WYFI), an existing customer, is expanding its use of Clockwork.io software across its growing global GPU-as-a-service footprint.
“Pressure-testing a cluster’s reliability before it reaches production is critical, because a customer who inherits a hidden fabric fault pays for it later in failed jobs and lost GPU-hours,” said Tom Sanfilippo, Chief Technology Officer, WhiteFiber. “Marginal optics, misconfigured NICs, and links that pass a basic test but degrade under load can slip through. Clockwork.io’s automated fleet audit validates every link and node at once, localizes faults in minutes, and lets us correct them before acceptance. We bring clusters up faster, and a customer’s first training run lands on a fabric validated end-to-end, not just powered on. With market demand growing as rapidly as it is, getting validated capacity to customers quickly is critical to our business, and it is why we are expanding Clockwork.io across our clusters.”
New TorchPass capabilities put workload protection in platform teams’ hands
Clockwork.io extended TorchPass beyond GPU migration with two capabilities platform teams have not had: a snapshot of a whole distributed job that they can take themselves, and application checkpoints fast enough to run in the background.
For training, platform snapshots save a running job’s execution state across all of its nodes so the job can be restored after an interruption. Platform teams and AI infrastructure engineers deploy it for supported workloads without waiting for application owners to modify their code or add checkpointing logic. Enterprise teams can protect training jobs across their fleet with one mechanism, and cloud providers can protect customer jobs whose code they do not control. Where teams checkpoint at the application level, TorchPass’s fast checkpoints can be taken more often, so less progress is lost and less computation repeated after a failure.
For inference, large models run across two or more servers, so one bad link can take down a whole replica and cut off a user session or agent task mid-stream. LinkPass keeps those multi-server replicas serving through link failures.
Reinforcement learning depends on both training and inference. Copies of the model generate rollouts, the trainer learns from them, and updated weights must reach the copies before they can generate with the latest version. TorchPass’s application checkpoints carry the updated weights to the rollout replicas sooner, while LinkPass keeps those replicas serving through link failures.
“Cluster fault tolerance used to be a training problem. It is now an inference problem too,” said Dylan Patel, Founder, CEO, and Chief Analyst at SemiAnalysis, whose ClusterMAX ratings benchmark GPU cloud providers. “In our ClusterMAX, TorchPass cuts training goodput loss from 14% to under 3% for a gold-rated neocloud. Reinforcement Learning (RL) ties the two together: inference replicas generate rollouts, the trainer learns from them, and the updated weights go back to the replicas. Clockwork.io keeps replicas serving through link flaps and network failures. Its extremely fast checkpoints accelerate weight transfer back into the rollout fleet, so neither direction stalls the run. One fault-tolerance layer under training, inference, and RL is where this has to be solved.”
Enterprise platform teams and cloud providers can contact Clockwork.io to evaluate the software for their workloads or explore partnership opportunities.
The round and what it funds
The round was co-led by Premji Invest, Wing Venture Capital, and Seligman Ventures, with participation from existing investors NEA and e& Capital. It brings Clockwork.io’s total funding to $73 million. The company will use the capital to accelerate the rollout of its fault-tolerance suite across training, inference and reinforcement learning, expand enterprise adoption, and scale delivery through cloud partners.
“The one thing that scales perfectly is unreliability: put enough GPUs in one machine and something is always failing,” said Greg Papadopoulos, Venture Partner, NEA. “The old playbook: stop the job, reload a checkpoint, makes no sense at today’s scale. Clockwork treats failure as the normal state: TorchPass migrates training off a failing GPU live, and now snapshots an entire running job with no code changes. It’s already saving tens of thousands of GPU-hours a month. We first backed Clockwork.io in 2021 and are thrilled to keep supporting them as they define the performance layer of the AI cluster.”
Clockwork.io pioneers Software-Driven AI Fabrics™, a programmable layer between hardware and workload that makes GPU clusters observable, fault-tolerant, and fully utilized across any accelerator, network, or cloud. AI workloads need the whole cluster to act as one machine, yet failures and bottlenecks idle GPUs. Clockwork.io’s FleetLens platform recovers that lost capacity: nanosecond-accurate telemetry pinpoints the GPU, node, or link slowing or stalling a job, validates clusters before launch, and, with LinkPass and TorchPass, keeps workloads running through infrastructure failures and degradations, from training to inference. SemiAnalysis has independently benchmarked TorchPass as recovering from failures faster than checkpoint-restart and leading open-source frameworks. LinkedIn, Together AI, WhiteFiber, Wells Fargo, Nebius, NScale, and DCAI trust Clockwork.io to power their AI infrastructure. Learn more atwww.clockwork.io.
The CAZ GP Stakes Fund Series Gives Investors Two Ways to Own a Stake in Leading Private Asset Managers
HOUSTON, Oct. 5, 2026 — CAZ Investments (“CAZ”), a Houston-based leading alternative investment firm, today announced the launch of the CAZ GP Stakes Growth Fund, a registered interval fund designed to give investors exposure to the long-term growth of some of the leading private asset managers.
Together, the CAZ GP Stakes Growth Fund and the existing CAZ GP Stakes Income Fund form the CAZ GP Stakes Fund Series, an approach CAZ calls “GP Stakes, Your Way.” Investors can pursue income, growth, or a combination of both, based on their objectives.
“We believe owning a piece of a private asset manager is one of the most attractive business models we have seen in our 25 years of investing,” said Christopher Zook, Founder and Chief Investment Officer of CAZ Investments. “We designed this fund series for our own personal capital, and now investors can customize their exposure to it. If your priority is income, the CAZ GP Stakes Income Fund was designed for you. If you are focused on long-term compounding, the CAZ GP Stakes Growth Fund was built for you. The fund series allows you to tailor your approach and select the combination that makes sense for you.”
Both funds are open to all investors regardless of accreditation status, with a minimum investment of $2,500. Each vehicle offers daily pricing and subscriptions, structured redemption through a quarterly repurchase program, a flat management fee with no performance fee, and 1099 tax reporting.
CAZ Investments works with a global network of wealth management firms. With the fund series, investment professionals can align each client’s allocation to their priorities, whether current income, long-term growth, or both.
CAZ was recognized by GP Stakes News as the GP Stake Investor of the Year for 2025, which reflects its position as a dominant participant in acquiring minority stakes in private asset management companies, commonly referred to as GP Stakes. CAZ has allocated over $8 billion to GP Stakes in the last 10 years and currently holds minority ownership in more than 120 private asset managers across Private Equity, Private Credit, and Private Real Estate.
CAZ Investments is a Houston-based alternative investment firm that thematically curates differentiated investment opportunities for its global network of investors. Founded in 2001, CAZ has grown to become one of the world’s top 100* allocators to private equity. With $12+ billion in assets currently deployed, CAZ provides exposure to asset classes such as GP Stakes, Professional Sports Ownership, Space and Defense, Energy, and more.
The CAZ global network consists of over 9,700 investors in all 50 states and 43 countries. The group includes more than 500 wealth management firms, investment advisors, family offices, and institutional investors, who value the innovative and differentiated opportunities that CAZ develops.
Investors should consider the investment objectives, risks, and charges and expenses of the Fund(s) before investing. The prospectus contains this and other information about the Fund(s) and should be read carefully before investing. The prospectus may be obtained at (855) 886-2307 or www.cazgpstakesfund.com.
The CAZ GP Stakes Income Fund and the CAZ GP Stakes Growth Fund are distributed by Ultimus Fund Distributors, LLC, Member FINRA/SIPC. CAZ Investment Registered Adviser, LLC is not affiliated with Ultimus Fund Distributors, LLC. Each Fund should be considered a speculative investment and entails substantial risks, and a prospective investor should invest only if it can sustain a complete loss of its investment.
Each Fund is newly formed and has limited operating history. Investing in Shares involves a high degree of risk.
Unlike many closed-end funds, the Shares are not listed on any securities exchange. To provide Shareholders with limited liquidity, each Fund intends to conduct repurchases of Shares in each quarter. No Shareholder will have the right to require a Fund to redeem its shares. There is no assurance that investors will be able to tender their shares when or in the amount desired.
A Fund’s distributions may be funded from unlimited amounts of offering proceeds or borrowings, which may constitute a return of capital and reduce the amount of capital available to the Fund for investment. Each Fund is non-diversified. Investing involves risk, including loss of principal. Past performance is no guarantee of future results. There is no guarantee that any investment strategy will achieve its objectives, generate profits or avoid losses.
The Fund invests in private funds which are subject to certain risks including those related to illiquidity, indirect fees, valuation, limited operating histories and limited information regarding underlying investments.
