From Tokyo: somewhere between the transistor and the photon, the next chapter of artificial intelligence is being written. The fight over how data moves inside a computer may not make headlines in the same way as model releases or chip announcements, but it is shaping the economics and energy footprint of every AI system on earth. This week, that fight got considerably more expensive.
On Tuesday, Nvidia-backed startup Ayar Labs raised $500 million to accelerate mass production of its co-packaged optics technology, a move that came just one day after Nvidia announced it would commit $4 billion to photonic networking providers Coherent and Lumentum ($2 billion each) to scale manufacturing capacity. The timing is not coincidental. It reflects a broader industry conviction that copper wiring, the backbone of chip-to-chip communication for decades, is running out of road.
The funding round brings Ayar's total capital raised to $870 million and values the company at $3.75 billion. The round was led by Neuberger Berman with participation from ARK Invest, Insight Partners, the Qatar Investment Authority, Sequoia Global Equities, 1789 Capital, AMD, Alchip, MediaTek, and Nvidia. The breadth of that investor list, spanning sovereign wealth funds, growth equity firms, and rival chip companies, tells its own story about the perceived commercial urgency.
The core problem Ayar is trying to solve is well understood in engineering circles, even if it rarely surfaces in public debate. Ayar's TeraPHY chiplets offer an alternative to copper for chip-to-chip communications, capable of supporting higher bandwidths over longer distances. Above 800 Gbps, copper interconnects are limited to a couple of metres and often require additional components to keep error rates in check; higher-speed copper links, like those found in Nvidia's NVL72 systems, are therefore typically confined to a single rack.
Pluggable optics can span longer distances, but at the cost of higher power consumption and latency. By integrating TeraPHY chiplets directly into a GPU or accelerator, Ayar says its designs can support significantly higher bandwidth while consuming a fraction of the power required by pluggable alternatives. In practical terms, that distinction matters enormously. Data transmission in legacy AI server architectures can consume a substantial share of total system power, and as cluster sizes grow, those energy costs compound quickly.
Ayar CTO Vladimir Stojanovic spelled out the ambition directly. "We want to be able to scale up to 10,000 GPU dies connected in a scale-up domain, while keeping the rack power and power density to around 100kW," he told The Register in an interview late last year. For context, one reference design developed in collaboration with Alchip uses eight of Ayar's next-generation TeraPHY chiplets, with the company claiming it can support more than 200 terabits per second of aggregate bandwidth per package, roughly five times the bandwidth available on Nvidia's Rubin GPUs.
Ayar says the new funds will be used to scale high-volume production and test capacity, expand global operations including at a new office in Hsinchu, Taiwan, strengthen ecosystem partnerships, and accelerate deployment of its CPO solution. Hsinchu sits at the heart of the global semiconductor supply chain, and the decision to plant a flag there signals Ayar's intent to embed itself directly into the manufacturing networks that will determine who wins this race.
The competitive picture is worth acknowledging honestly. Ayar is not alone. Rivals such as Celestial AI raised $250 million in March 2025, while established photonics firms Lumentum and Coherent now carry significant Nvidia backing of their own. Lightmatter, another CPO-focused startup, has also launched a photonic interposer product in a similar vein to Ayar's TeraPHY. And in February, Marvell Technology completed its roughly $3.25 billion acquisition of Celestial AI, consolidating another corner of the photonics market. The field is crowding quickly, and not every startup will survive the transition from prototype to mass production.
There is also a legitimate question about timing and execution risk. Silicon photonics has historically struggled with manufacturing precision, heat management, and packaging challenges. Ayar Labs says it has cleared those hurdles and is preparing for high-volume production, a claim investors appear willing to back with substantial capital. Whether the technology performs at scale in commercial deployments, not just in controlled reference designs, remains to be proven. Venture capital enthusiasm and engineering reality do not always converge on the same schedule.
From an Australian perspective, these developments carry real weight. Australian data centre operators, cloud computing customers, and the government's own digital infrastructure ambitions are all downstream of the technological choices being made right now in Silicon Valley and Hsinchu. Nvidia has described co-packaged optics not as an optional enhancement but as a structural requirement for future AI data centres. If that assessment proves correct, the economics of AI infrastructure globally, including in Australia, will shift in ways that reward those who engage early with these supply chains and standards.
What Australian observers often miss about the Indo-Pacific technology ecosystem is how much of this infrastructure story is being written in Taiwan. The concentration of advanced semiconductor design and packaging expertise in Hsinchu and its surrounding region means that decisions about AI's physical architecture are being made within a geopolitical context that matters deeply to Canberra. Australia's Department of Defence and trade policymakers would be wise to track the supply chain dependencies now being formed, before they become structural constraints.
The pragmatic read on all this is that silicon photonics is graduating from research curiosity to commercial infrastructure. The capital, the partnerships, and the strategic alignment between chip giants and startups all point in the same direction. Whether Ayar Labs specifically emerges as the dominant player matters less than the broader shift it represents: the age of copper-limited AI is ending, and the transition to optical interconnects will reshape data centres, energy consumption, and semiconductor supply chains for years to come. The companies, and countries, paying attention now will be better placed when that transition arrives in force. Australia Bureau of Statistics data already shows the nation's digital economy growing as a share of GDP; ensuring that growth is built on resilient, future-proof infrastructure is a question of both economic competitiveness and national interest.