Lumilens Inc., known for developing optical networking chips for artificial intelligence clusters, recently raised $900 million to expand its operations. The majority of the funds came from a Series C round of $700 million, which valued the two-year-old company at $5.51 billion.
Atreides Management, Bain Capital Ventures, Meritech, Seligman Ventures, and Spark Capital participated in this round, leading the financing. Qualcomm and over a dozen other investors also participated in the investment. The company is led by co-founder and CEO Ankur Singhal.
Historically, data centers have used copper wires to transmit data between processors. This method allows for lower speeds, but AI clusters exceed these limitations. Copper can only transmit data at a speed of about 1.6 terabits per second over a distance of approximately one meter, which limits operation within a single rack cabinet.
Lumilens' chips replace copper connections with fiber optic cables. Fiber optics allow for the transmission of more data over longer distances with less energy consumption, which is critical for companies building clusters of thousands of graphics processing units (GPUs).
Lumilens focuses on network scaling through two product lines. The first is Co-Packaged Optics (CPO), which integrates optics directly into the GPU. Due to its proximity to the processor, data travels a shorter distance, increasing energy efficiency. The company claims that CPO can connect thousands of GPUs.
The second line is Near-Packaged Optics (NPO), where optics are located near the GPU on the motherboard but not inside it. Although NPO is less efficient than CPO, it is easier to implement currently.
Singhal believes that it will take another two to three years for the large-scale volume deployment of Co-Packaged Optics. Lumilens first tested its chip in 2024, and the second next-generation chip is under testing and planned for commercial launch next year.
For network scaling, Lumilens produces chips for pluggable transceivers that attach to switches and connect entire rows of GPUs. This technology supports speeds of up to 800 gigabits per second, 1.6 terabits per second, and higher. Since optics are used, there are no distance limitations that make copper impractical outside of a rack cabinet. Singhal emphasized that companies need to transition to optical solutions to go beyond the rack cabinet.
All Lumilens chips are built on LumiCore—a common set of building blocks. It includes photonic components, mixed-signal integrated circuits (mixed-signal ICs), and interposers that serve as the base layer for optical and electronic elements. The company states that LumiCore reduces development time from several years to several months, and the interposer design simplifies assembly.
Lumilens has developed its own manufacturing process using robots and AI-based software to support scaling. The $900 million will be directed towards accelerating chip development and production. In the race among AI labs to create larger clusters, Lumilens is betting that optical communication will become the foundation of next-generation AI infrastructure.
Lumilens, an American startup specializing in optical interconnects for data centers, has raised over $700 million in a recent funding round, valuing the company at $5.5 billion. The San Jose, California-based company aims to replace some of the electrical connections used between servers with light-based systems. This information was exclusively reported by The Wall Street Journal.
The investment was led by Atreides Management, Bain Capital Ventures, Meritech, Seligman Ventures, and Spark Capital, according to the company itself. Since its founding two years ago, Lumilens has already raised over $900 million and begun shipping its equipment to a major cloud provider, whose name remains confidential.
This movement occurs amid growing demand for computational power dedicated to artificial intelligence. With the concentration of more advanced chips in data centers, the need to transfer information rapidly between processors has become a central challenge of current infrastructure.
The advancement of AI systems has intensified pressure on data centers, requiring the connection of large volumes of processors at higher speeds. Currently, many of these connections rely on copper cables, a solution recognized for its reliability and low cost, but which shows limitations when data volume and distances increase.
Lumilens is part of a group of companies seeking to solve this bottleneck through photonic technology. The proposal involves replacing electrical signals with light transmissions, using optical connections capable of transporting data via fibers.
Other corporations are also exploring this market. Lightmatter, established in 2017, has already received $850 million in investments and had a valuation of $4.4 billion in 2024. Previously, Celestial AI, founded in 2020, developed a similar methodology before being acquired by Marvell in February for $3.25 billion.
Despite the technical potential, the implementation of optical technology faces economic barriers. According to Ankur Singla, CEO of Lumilens, the necessary equipment has a significantly higher cost than copper-based alternatives.
The executive explained that the company's strategy is to reduce the cost disparity between the two models, making the replacement more practical. In addition to developing components, the startup also focuses on product manufacturing.
Singla also expressed the intention to keep production outside of China. This decision gained importance after a Reuters report indicated that the Trump administration intended to impose restrictions on the entry of certain Chinese components into data centers, including optical equipment.
Lumilens offers two main lines of solutions. One is for communication between different racks within the same data center, a process called 'scale out'. The other focuses on connecting chips within a single rack, in a configuration known as 'scale up'. The company's first commercial product is an optical transceiver, a device that converts electrical signals into light to enable data transmission via fiber optic cables. This equipment is already being supplied to a major cloud provider, which uses this technology alongside its own artificial intelligence chips.
Lumilens expects to capitalize on the progress of artificial intelligence to accelerate a transformation that the data center industry has awaited for years. Although optical technology was seen as an alternative to copper, its high price limited its large-scale adoption.
Multiplier has successfully raised $35 million in a Series B funding round. These funds are intended to scale its artificial intelligence-powered professional services platform. The General Partnership led this round, with existing investors Ribbit Capital and Lightspeed Venture Partners also participating.
The acquired investments will be directed towards acquiring additional specialized firms that provide professional services. Furthermore, they will help the company grow its technological and operational teams in response to the increasing demand for AI-based business services.
Multiplier operates on a permanent holding company model. By acquiring existing enterprises, the company allows them to retain their names, management, and a significant degree of autonomy. Instead of replacing people with artificial intelligence, Multiplier creates AI-based products that enhance efficiency and the quality of customer service.
Founder and CEO Noah Pepper stated in a company blog post that AI should serve as a tool to empower working professionals, not replace them. He believes that AI tools allow specialists to dedicate more time to tasks requiring judgment by eliminating routine work.
Multiplier integrates its engineering teams directly into the acquired firms. The company has also hired over 30 technologists whose work, together with subject matter experts, leads to the creation of AI tools tailored to the needs of specific businesses. Since the companies using this technology are not exclusively owned by third-party software developers, Multiplier employees have access to client and supplier databases, which, according to the founders, means 'access to clients, products, and processes.'
The company's model is focused on industries where growth is limited by expertise. Many professional service firms rely heavily on experienced practitioners and manage information through disparate systems. Multiplier believes that AI can automate preparatory and routine tasks, leaving final decisions to qualified specialists.
Since its inception, the company has completed eight acquisitions, five of which were closed within the last year. It also has four other firms that have signed letters of intent. In total, the company's network serves clients in more than 10 countries.
The latest investment coincides with the appointment of a senior executive. This person will oversee finance, operations, partnerships, personnel, and the expanding San Francisco office. The company believes this approach ensures more effective scaling while preserving valuable client relationships and professional experience over time.
Allen Shim, former CFO of Slack, now holds the positions of President and Chief Financial Officer at Multiplier. The company plans to use the funds to effectively build a scalable technological base, team, and operational process architecture. Shim previously led the successful initial public offering of Slack before its acquisition by Salesforce.
According to The General Partnership, accounting and other professional services markets continue to face capacity constraints despite growing demand. The investment firm is convinced that Multiplier's approach allows trusted specialists to serve more clients without compromising service quality.
The new capital will primarily go towards additional acquisitions and further development of proprietary AI solutions. Multiplier intends to develop its specialized firms and improve its technological capabilities. The company's investment approach differs from traditional 'buy and build' strategies, where founders of acquired companies are completely replaced after the acquisitions. Instead, founders remain in leadership roles, receiving continuous support in the form of long-term capital and operational management.
Horizon3, a San Francisco-based company specializing in artificial intelligence-driven cybersecurity, has secured $250 million in Series E funding. The funds will be used to further expand its platform internationally across the globe. Following this round, the company's valuation exceeded $2 billion.
This valuation is more than triple the approximately $600 million achieved during the previous funding round a year ago. The investment is intended to support Horizon3's native cybersecurity expansion into international corporate markets and client operational activities.
The funding was led by NightDragon and NEA, with participation from existing and new global strategic investors. The company plans to accelerate innovation and strengthen proactive cybersecurity capabilities for enterprises facing increasingly complex, AI-driven threats.
The raised capital will be utilized for global sales development, marketing expansion, market growth, and initiatives to develop autonomous cybersecurity technologies worldwide. Horizon3 intends to increase its presence in Europe and enter the Singaporean and Australian markets through strategic international expansion.
The company will increase R&D investments to enhance its NodeZero platform and AI-based security agents. These autonomous agents are capable of effectively and automatically detecting and remediating vulnerabilities within enterprise IT systems.
The investment will also go towards strengthening the proactive AI security platform and accelerating the innovation process. This financing follows growing enterprise demand for protection against AI-driven cyber threats and evolving regulatory compliance requirements.
Snehal Antani and Anthony Pillitiere, former colleagues from JSOC, founded Horizon3 with the goal of automating penetration testing via NodeZero. NodeZero continuously attacks client networks using self-guided penetration testing technology to proactively detect cybersecurity vulnerabilities.
Unlike annual audits, the platform safely tests live production environments using fully automated continuous security testing capabilities. This helps enterprises identify attack vectors, prioritize security weaknesses, and quickly and efficiently implement remediation measures.
The company reported conducting over 310,000 security tests in client production environments globally. Unlike traditional testers, Horizon3 assesses the entire enterprise infrastructure while production systems continue to operate uninterrupted.
The platform also incorporates intelligent honeypots that help organizations detect active attackers targeting enterprise systems in real time. Currently, the platform is used by over 7,000 enterprises across sectors such as banking, healthcare, public administration, and Fortune 10 companies worldwide.
The rapid adoption of AI has changed the approach to how security issues arise. For instance, an attacker with the right automated approach can quickly identify weaknesses and initiate complex attacks faster than manual methods, which reinforces the automation of security assessment.
Horizon3 reported a year-over-year recurring revenue growth of approximately 120%, noting that its annual recurring revenue approached $100 million last year. The company serves between 7,200 and 7,300 clients globally, including managed service providers and some of the world's largest corporations.
The new funding will also support the development of AI training loops, allowing AI defense systems to continuously improve by analyzing new attack methodologies. This should support enterprises' efforts to move from simple vulnerability awareness to the ability to automatically mitigate the most severe security threats.
Given that new cyber threats emerge at 'machine speed' and with high sophistication, the company asserts that AI-based security testing is indispensable for enterprises seeking to protect critical assets and confidential data.
