Linque: Shaping the Future of Programmable Photonics

As AI models grow larger, the actual bottleneck is shifting from compute to bandwidth. Data centers these days are struggling to keep up with the massive volumes of data that need to move between chips. Energy use and latency are increasing as information travels through layers of electronic infrastructure that were not designed for workloads at this scale.

Photonics offers a fundamentally different approach. Using light instead of electricity to transmit and process information enables higher bandwidth, lower latency, and greater energy efficiency. The next frontier lies in making these photonic systems programmable and tightly integrated with electronics, allowing data to be routed and manipulated directly in the optical domain.

Linque was founded in 2021 by Samarth Vadia (CEO), Victor Funk (VP R&D), and Jonathan Förste (CTO) to advance this vision. The company is developing photonic hardware for cloud and AI applications. Its first product, RISE, is an all-optical circuit switch that eliminates optical-electrical-optical conversions. In the long term, Linque aims to build the world’s largest programmable photonic chips, which perform information processing directly in light.

What sets Linque apart is its approach to addressing one of the field’s core challenges: optical loss. In addition, Linque’s advances in architecture and system integration enable larger and more capable photonic processors that can scale efficiently.

The company is currently working with selected customers ahead of RISE’s planned launch in H1 2026. Until now, Linque has received €4.5M in non-dilutive funding. 

Learn more about the future of programmable photonics from our interview with Linque co-founder Samarth Vadia:

Note: The terms “photonic” and “optical” describe the same underlying technology, using light to transmit or process information, and are used interchangeably in this article.

What Is Your Background and What Inspired You To Start Linque?

I studied physics at IIT Delhi and spent several summers in Europe for research internships, which sparked my interest in exploring science internationally. After a year in Boston, I moved to Munich for my PhD, joining a Marie Curie project between LMU Munich and attocube, a German hidden champion in nanotechnology and quantum instrumentation. My research focused on quantum optics at the intersection of semiconductors and quantum physics.

Editor’s note: Quantum optics studies how light interacts with matter at the quantum level, including phenomena such as single-photon generation and entanglement.

When I finished my PhD in 2021, I wanted to stay close to this field but work in a more entrepreneurial setting. Around that time, breakthroughs like ChatGPT marked a turning point in how AI impacts our lives. It became clear that new compute paradigms were needed, primarily to address the growing energy consumption of AI systems.

Around the same time, my co-founders, Victor Funk and Jonathan Förste, whom I had known since our PhD days, shared the same conviction that photonics would play a major role in the future of computing. We decided to start Linque together to bring that vision to life.

How Is Linque Advancing Photonic Hardware Across the Semiconductor Value Chain?

Photonics can transform computing by utilizing light to transmit and process information with higher bandwidth and lower energy consumption compared to traditional electronic systems. This belongs to the realm of linear computation, which is both energy-efficient and faster, thanks to the rapid encoding and decoding of information. It also opens the door to a wide range of use cases.

Editor’s note: Linear computation involves mathematical operations that scale proportionally with input signals, in contrast to nonlinear computation, where outputs depend on more complex interactions such as amplification or thresholding.

Broadly speaking, we design and build hardware for cloud and AI applications with photonics at their core. Our model is fabless, and we aim to deliver a complete solution across the semiconductor and photonics value chain. We handle chip design in-house, collaborate with foundries for manufacturing, and work with OSAT partners for testing and assembly. From there, we provide fully integrated hardware systems to customers, combining photonic and electronic components into a unified solution.

Editor’s note: Fabless means a company designs its own chips but does not own a fabrication plant. Instead, manufacturing is outsourced to foundries.

Editor’s note: OSAT partners stands for “Outsourced Semiconductor Assembly and Test” companies, which package and test chips after a foundry manufactures them.

What Is Linque’s First Product and How Does It Translate Your Photonic Architecture Into a Near-Term Commercial Application?

Before commercialising our programmable all-optical chip, we are releasing our beachhead product called RISE, an all-optical circuit switch for the data center and hyperscaler networks. 

Editor’s note: A switch directs data traffic between network ports or nodes, ensuring that information reaches the correct destination. A circuit switch establishes a dedicated optical path between two nodes for the entire duration of a data transfer. (See below: In contrast, a packet switch sends data in small packets that are each routed independently through the network.)

Networking and switching are the initial applications we are pursuing because they align with what is currently possible at the chip level. The number of components and channels we can integrate today matches the requirements of an optical switch, which makes it the most practical use case to bring to customers first. 

Editor’s note: In this context, networking refers to connecting computing systems such as servers, accelerators, and storage devices through high-speed communication links, while switching focuses on directing data traffic between those connections within a network.

It also supports native multiplexing and intrinsic network monitoring, which strengthen security and enable intelligent resource allocation for high-priority workloads.

Editor’s note: Native multiplexing means combining multiple optical signals (by wavelength or spatial mode) into a single channel to increase total data throughput.

Editor’s note: Intrinsic network monitoring refers to measuring signal quality and network status directly within the optical layer, without converting data back into electrical form.

By avoiding optical-electrical-optical (OEO) conversions and keeping the signal entirely in the optical domain, RISE provides clear benefits: latency in the nanosecond-to-microsecond range, lower power consumption, and bandwidth throughput that can scale beyond terabits per second.

RISE is planned to hit the market in H1 2026. Ahead of the launch, we are already working with selected customers to validate its feasibility in real environments. Our most recent system includes three separate tape-outs of ASICs with heterogeneous integration, a reconfiguration time of 10 microseconds, over 100 fibers, and more than 1000 electrical I/Os. These are milestones that illustrate how we are scaling the technology step by step toward production.

Editor’s note: Tape-out marks the point when a chip design is finalized and sent to a foundry for fabrication.

Editor’s note: ASIC stands for “application-specific integrated circuit”, a custom-designed chip optimized for a particular function.

Editor’s note: Reconfiguration refers to changing which optical connections inside the switch are active, effectively redirecting light paths between servers. The reconfiguration time indicates how quickly the switch can update these connections and reorganize the network’s structure in real time.

Editor’s note: I/O refers to input and output connections, the channels that allow data to move into and out of a chip or between different subsystems.

Most Switches Rely on Conversions Between Optical and Electrical Domains. How Does Linque’s All-Optical Architecture Avoid Them?

In today’s data centers, most switches are packet-based, following standards such as Ethernet or InfiniBand. Each port typically includes a pluggable transceiver that converts incoming optical signals into electrical ones. The switching and routing then take place in the electronic domain before the signal is converted back into light for transmission.

Editor’s note: Packet-based switches handle data divided into small packets that are independently routed through a network. (See above: In contrast, circuit-based switches establish a continuous path between two points for the entire duration of a data transfer.)

Editor’s note: Ethernet and InfiniBand are communication protocols widely used in data centers for transferring information between servers and storage systems.

Editor’s note: A port is a physical interface on a switch or device where data enters or exits, typically through an optical fiber or electrical connector.

Our approach removes this OEO conversion step. The fiber connects directly to the photonic chip, where routing takes place entirely in the optical domain. The data never enters the electronic domain. Electronics are still needed for control and programming, but the signal itself stays optical end-to-end. This brings clear benefits: Latency is significantly reduced, and energy consumption decreases since data conversion is a significant source of overhead. 

Our approach uses circuit switching rather than packet switching. That means we do not inspect every packet individually. Instead, the network operator knows where the data is flowing between nodes and can dynamically redefine the circuits or network topology very quickly, although not on a packet-by-packet basis.

Editor’s note: Network topology refers to the arrangement of connections between nodes in a network, such as the organization and interconnection of servers, switches, and links.

What Is the Long-Term Product Vision You’re Building Toward?

Traditionally, photonics has primarily been utilized in transceivers, where its function is limited to converting data between optical and electronic signals. 

Editor’s note: Transceivers in photonics convert data between optical and electrical domains (while transceivers in electronics handle purely electronic signal transmission).

Our focus is on developing an all-optical chip that can handle part of the information processing directly in the optical domain. What we are building are larger “programmable” chips that go beyond simple conversion and perform real processing in the optical domain.

Building a truly large photonic chip requires integrating a very high number of these active components on the same chip. For context, a conventional optical transceiver typically has just one modulator and one detector. To enable functions like switching, data communication, or information processing, you need many modulators working together in sequence. That is the key enabler for unlocking new use cases.

Editor’s note: Modulators control how light carries information. They adjust properties such as intensity or phase to encode data in optical signals used for communication or processing.

The main technical challenge is that every component introduces optical loss. As light passes through a chain of components, the losses accumulate. You may start with a strong signal, but by the time it reaches the output, it can be weakened to the point where noise dominates. Overcoming this cumulative loss has been one of the significant limitations in the field, and solving it is a core technological problem we are addressing at Linque.

Editor’s note: Optical loss refers to the reduction in signal strength as light travels through optical components or fibers.

What Does “Programmable” Mean in the Context of Photonic Chips and What New Capabilities Does It Unlock?

With “programmable”, I mean the ability to manipulate information directly in the photonic domain. This is achieved using active components, such as modulators, including Mach-Zehnder or ring modulators.

Editor’s note: Mach-Zehnder modulators split and recombine light beams to control how light carries information. Ring modulators use tiny circular waveguides (the optical equivalent of wires in electronics) to change the light’s properties in a compact and energy-efficient way.

What Are the Main Technical Bottlenecks You Encounter?

We see our technical challenges in two main domains: system integration and scaling.

The first challenge is system integration, which exists on two levels:
(a) integration between domains, combining photonics and electronics, and
(b) integration on the chip level across different photonic materials.

Photonics offers clear advantages, but the future will depend on the convergence of optical and electronic technologies. The primary challenge lies in integrating optical and electronic chips to enable them to operate together effectively. 

In such systems, electronic driver circuits generate high-frequency radio signals that control photonic components like modulators and switches. 

At the same time, multiple chiplets must exchange data quickly and reliably, requiring precise coordination and high-performance communication links. Integration across these domains can be achieved through heterogeneous or advanced approaches such as 3D chip integration.

Editor’s note: Heterogeneous integration refers to combining different types of chips or materials, such as optical and electronic components, into a single package to enhance performance and reduce size.

Editor’s note: 3D chip integration stacks multiple layers of chips vertically and connects them with microscopic interconnects to shorten communication paths and increase density.

Editor’s note: Radio frequency signals are high-frequency electrical signals used to drive or synchronize optical components. They are generated by electronic driver circuits, which supply the voltage and timing needed to operate modulators and other active photonic devices.

Integration is also critical at the chip level across materials. While we use silicon photonics, light generation requires active materials such as indium phosphide. To achieve full functionality, photonic systems need to integrate multiple material platforms.

That is why we are working with partners in the semiconductor value chain (including the Fraunhofer Institutes in Germany) on scalable heterogeneous integration of III-V materials with silicon, and with collaborators in Taiwan to align these developments with CMOS manufacturing processes.

Editor’s note: Indium phosphide (InP) is a compound semiconductor that efficiently emits and detects light, making it a key material for photonic devices.

Editor’s note: III-V materials are semiconductors made from elements in groups III and V of the periodic table, such as indium phosphide (InP) or gallium arsenide (GaAs), often used for light generation.

Editor’s note: CMOS stands for “complementary metal–oxide–semiconductor”, the standard technology used in most modern electronic chips.

The second challenge is scaling programmable photonic chips to incorporate a very large number of components. Traditional transceivers operate with only a few, but advanced data processing on a chip requires tens of thousands to millions of components. Achieving this scale requires optimization across design, material, and geometry to enhance performance, minimize losses, and operate at high speeds. These losses correspond to the optical losses mentioned earlier, where signal strength decreases as light passes through many components.

How Do You Differentiate Yourself From Other Photonic Chip Companies?

There is strong competition in this space, with many players active across different areas. For us, differentiation lies in how we design and build large-scale programmable photonic chips, as well as our approach to system integration.

From the customer’s perspective, the advantage of using our technology is clear: we can deliver lower power consumption at the system level, reduced latency, and faster dynamic reconfiguration compared to alternative technologies.

What Advice Would You Give to Fellow Deep Tech Founders?

What has helped us a lot so far is the support of the founding and startup community. My advice is simple: ask for help. Reach out to fellow founders, to VCs, or to accelerators. That approach has been invaluable for us. Whenever questions arose (e.g., how to approach hiring, securing funding, or dealing with early-stage challenges), we often found that another founder, just a year or two ahead, had faced the same issues. Almost everyone we turned to was open, approachable, and willing to share their experience.

That kind of support has made a real difference, and I pay it forward whenever I can.