Key Takeaways
- CamGraPhIC, the graphene photonics arm of Italy’s 2D Photonics Group, has secured €211 million in state-backed funding approved by the European Commission to industrialise its optical interconnect platform.
- The capital will finance a pilot manufacturing facility near Milan and accelerate deployment of graphene-based Optical I/O technology into AI accelerators, high‑performance computing systems, and advanced data centres.
- CamGraPhIC aims to boost bandwidth density while cutting latency and power consumption versus leading silicon photonics, directly tackling AI’s internal data‑movement bottleneck.
- The grant sits alongside earlier equity raises, including a €25 million Series A round backed by CDP Venture Capital, NATO Innovation Fund, Sony Innovation Fund and others.
Quick recap
In a landmark deeptech funding move, CamGraPhIC, the graphene photonics subsidiary of 2D Photonics Group, has been approved for a €211 million grant by the Italian state under the European Commission’s State Aid Framework to tackle AI’s crippling data interconnect bottleneck.
The company will use the funds to industrialise its graphene‑based optical input/output platform and build a pilot manufacturing facility near Milan, targeting next‑generation AI accelerators and data centres. The funding was first highlighted publicly via a post from EU‑Startups on X, which flagged it as one of Italy’s largest deeptech packages to date.
Graphene optics to move data at AI scale
According to 2D Photonics, today’s largest AI systems are increasingly constrained not by raw compute, but by how quickly and efficiently data can move between chips and accelerator nodes. Conventional electrical links and even state‑of‑the‑art silicon photonics are struggling with rising power draw, thermal load, and latency as model sizes and cluster densities scale up.
CamGraPhIC’s answer is a graphene‑based Optical I/O platform that replaces slower, power‑hungry interconnects with high‑bandwidth optical links designed to sit much closer to the processor. Graphene’s electronic and optical properties enable a higher bandwidth density at significantly lower energy per bit than leading silicon photonics, promising to reset the ceiling on AI cluster performance and efficiency rather than providing incremental gains.
The €211 million facility‑level funding will support device qualification, early production runs and eventual transfer of the technology into high‑volume foundry processes, creating a pipeline from lab‑scale research to commercial deployment in AI accelerators, high‑performance computing and advanced data centres. This comes on top of equity financings such as CamGraPhIC’s €25 million Series A, co‑led by CDP Venture Capital, NATO Innovation Fund and Sony Innovation Fund, to scale its graphene photonics transceivers.
Why this move matters now?
AI data centres are hitting power and cooling ceilings, making interconnect efficiency a strategic choke point for both hyperscalers and semiconductor vendors. As cluster sizes grow, every incremental watt spent on moving data between GPUs or custom accelerators erodes the economics of training and serving large models. This has opened a window for new optical interconnect players that can offer better bandwidth‑per‑watt than traditional approaches.
CamGraPhIC is entering a competitive but fast‑forming niche alongside companies working on advanced optical and energy‑efficient infrastructure for AI data centres, such as C2i, which recently raised USD 15 million for a grid‑to‑GPU power‑optimisation platform. The backing of the European Commission and Italy’s state‑aid framework signals a desire to build sovereign capability in AI hardware plumbing rather than relying solely on foreign chip and optics vendors. If successful, the Milan facility could become a reference point for European optical interconnect manufacturing at scale.
Competitive optics and interconnect snapshot
Below is a simplified, AI‑oriented comparison using representative data and typical positioning in this emerging segment. Figures are indicative and based on public descriptions of focus and scale, rather than formal product price sheets.
| Feature/Metric | CamGraPhIC (graphene Optical I/O) | C2i (grid‑to‑GPU infra) | Typical silicon photonics vendor |
| Primary focus | On‑package optical interconnect for AI/HPC data movement. | Power and efficiency optimisation for AI data centres. | Optical links for data centre networking and short‑reach interconnects. |
| Context window (AI use) | Targets intra‑cluster bandwidth for very large AI models (multi‑thousand GPU scale, effectively “long” context at hardware level). | Indirect; improves uptime and power budgets for large AI clusters. | Suited to current‑generation cluster scales with incremental bandwidth gains. |
| Pricing per 1M “tokens”* | Aims to reduce effective interconnect cost per unit of data moved via lower energy per bit; pricing not yet public. | Reduces cost per inference/training run via power savings; service pricing undisclosed. | Mature but higher energy‑per‑bit profile; cost tied to existing optics supply chains. |
| Multimodal support | Hardware‑level: agnostic, supports any AI model type relying on fast data movement (text, vision, multimodal). | Software and power‑infra‑level; supports any workload running in the data centre. | Generally workload‑agnostic, optimised for network traffic patterns more than model type. |
| Agentic capabilities | None at software level; enables more complex agentic systems by lifting bandwidth and latency limits inside clusters. | None directly; supports always‑on, power‑efficient agentic services. | None; provides transport for AI traffic without workload intelligence. |
| Funding scale | €211 million state‑aid grant plus earlier €25 million Series A. | USD 15 million venture round. | Large incumbents with diversified revenue, typically hundreds of millions+ annually. |
From a strategic angle, CamGraPhIC looks strongest where bandwidth density and energy per bit are the main constraints, particularly for next‑generation AI accelerators and tightly coupled HPC systems. By contrast, C2i appears better positioned for operators focused on grid‑to‑rack efficiency rather than on‑package data movement, while traditional silicon photonics vendors still dominate mature, general‑purpose optical networking but may struggle to match graphene’s promised performance‑per‑watt if CamGraPhIC executes on its roadmap.
Bayelsa Watch’s Takeaway
In my experience, hardware shifts like this often matter more to AI’s long‑term trajectory than yet another model release, and this €211 million package has that feel. I think this is a big deal because it treats AI’s data bottleneck as an infrastructure problem to be engineered away, not a cost of doing business.
If CamGraPhIC can turn graphene Optical I/O into a reliable, manufacturable product, it could quietly reprice the economics of training and serving frontier‑scale models in Europe and beyond. I generally prefer to see public capital used where private markets hesitate, and backing a capital‑intensive optical fab near Milan fits that logic, though execution risk remains high. For now, I see this as net bullish for European AI hardware sovereignty and for users who ultimately benefit from cheaper, faster AI services riding on more efficient interconnects.
