Last updated: April 2026 – This article has been reviewed and updated with the latest information.

NVIDIA has made its most aggressive move yet to secure the future of AI data center connectivity, committing a combined $4 billion in strategic investments to silicon photonics leaders Lumentum Holdings and Coherent Corp. Announced in early March 2026, the twin deals – each worth $2 billion in growth equity plus multibillion-dollar purchase commitments – signal a fundamental shift in how the semiconductor giant views the bottleneck problem threatening to slow the AI infrastructure buildout.

The investments come at a critical inflection point for the AI industry. As GPU clusters scale from thousands to hundreds of thousands of chips, the copper-based electrical interconnects that link them are hitting physical limits in bandwidth, latency, and power consumption. NVIDIA’s bet on silicon photonics – using light instead of electrons to move data between processors – represents a strategic pivot that could reshape the $45.8 billion data center networking market and define the architecture of next-generation AI factories.

Inside NVIDIA’s $4 Billion Silicon Photonics Strategy

NVIDIA’s investment strategy is notable for its dual-supplier approach. Rather than backing a single vendor, the company split its $4 billion commitment equally between Lumentum Holdings (NASDAQ: LITE) and Coherent Corp (NYSE: COHR), investing $2 billion in each. Both deals include not just equity stakes but multibillion-dollar purchase commitments for advanced laser components, future capacity access rights, and support for new U.S.-based fabrication facilities.

Jensen Huang, NVIDIA’s founder and CEO, framed the investments in ambitious terms: “Together with Lumentum, NVIDIA is advancing the world’s most sophisticated silicon photonics to build the next generation of gigawatt-scale AI factories.” The statement reveals the scale NVIDIA envisions – AI data centers consuming a gigawatt or more of power, requiring optical interconnects that can move petabytes of data per second with minimal energy overhead.

Michael Hurlston, Lumentum’s CEO, confirmed the scope of the partnership: “This multiyear strategic agreement reflects our shared commitment to advancing the optics technologies that will power the next generation of AI infrastructure.” Both agreements are structured as nonexclusive arrangements, meaning Lumentum and Coherent remain free to supply other customers – a strategic choice that suggests NVIDIA wants to grow the entire silicon photonics ecosystem rather than lock up supply.

The investments will fund new U.S.-based manufacturing capacity, R&D expansion, and the scaling of production lines to meet what NVIDIA projects will be massive demand from hyperscalers and enterprise AI adopters over the next three to five years. This domestic manufacturing focus aligns with broader industry trends toward reshoring critical semiconductor supply chains, particularly after the disruptions of 2024-2025.

Why Silicon Photonics Is the AI Infrastructure Bottleneck Solution

To understand why NVIDIA is making a $4 billion bet on optics, you need to understand the bandwidth bottleneck threatening AI scaling. Modern AI training clusters – particularly those running NVIDIA’s Blackwell and upcoming Rubin architectures – connect thousands of GPUs through high-speed networks. Today, most of these connections rely on electrical copper interconnects and pluggable optical transceivers, both of which face fundamental physical limitations at the speeds and scales AI demands.

Traditional 1.6 Tbps optical transceivers consume approximately 30 watts each, with the digital signal processor (DSP) alone accounting for more than 15 watts – over half the total power draw. In a large AI cluster with thousands of interconnects, this adds up to megawatts of power consumed just moving data between GPUs, before any computation happens. NVIDIA’s co-packaged silicon photonics approach eliminates the external DSP entirely, delivering what the company claims is 3.5x lower power consumption compared to traditional pluggable transceivers.

Co-packaged silicon photonics works by integrating photonic components directly onto or adjacent to the switch silicon, reducing the signal path from inches to millimeters. This dramatically cuts latency, improves energy efficiency, and increases bandwidth density. NVIDIA demonstrated this technology at GTC 2025, integrating silicon photonics with its Quantum and Spectrum switch ICs using cutting-edge 200G SerDes technology.

“The shift from pluggable optics to co-packaged silicon photonics is analogous to the transition from discrete components to integrated circuits in the 1960s,” said Dr. Keren Bergman, Professor of Electrical Engineering at Columbia University and a leading researcher in photonic interconnects. “It’s not just an incremental improvement – it fundamentally changes what’s possible in terms of data center scale and efficiency.”

The Data Center Networking Market at a Crossroads

NVIDIA’s silicon photonics investments land in a data center networking market undergoing explosive growth driven almost entirely by AI workloads. According to industry research, the global data center networking technologies market – encompassing optical interconnects, high-performance switches, and SmartNICs – reached $45.8 billion in 2025 and is projected to grow to $103 billion by 2030, representing a compound annual growth rate of 17.6%.

The optical transceiver segment alone was valued at $8.42 billion in 2025, with projections pointing toward $9.15 billion in 2026 and accelerating growth as 800G and 1.6T transceivers become standard in AI-optimized data centers. Silicon photonics is expected to capture an increasingly large share of this market as co-packaged solutions prove their advantages in power efficiency and bandwidth density.

Market Segment	2025 Value	2030 Projection	CAGR	Primary Growth Driver
Data Center Networking (Total)	$45.8 billion	$103 billion	17.6%	AI cluster scaling
Optical Transceivers	$8.42 billion	$18.5 billion (est.)	17.0%	800G/1.6T adoption
Silicon Photonics ICs	$3.2 billion	$9.8 billion (est.)	25.1%	Co-packaged optics
High-Speed Ethernet Switches	$12.4 billion	$28 billion (est.)	17.7%	AI/ML workload networks
Optical Fiber & Cables	$7.8 billion	$15.2 billion (est.)	14.3%	Hyperscaler expansion
SmartNICs & DPUs	$4.1 billion	$11 billion (est.)	21.8%	Offload & security

The acceleration in spending is staggering. Big Tech companies collectively committed over $700 billion to AI infrastructure in 2025-2026, with networking representing approximately 15-20% of total data center capital expenditure. As AI clusters grow larger, the networking share of capex is expected to increase, making silicon photonics an increasingly critical investment category.

Competitive Landscape: Who Controls the AI Optics Supply Chain

NVIDIA’s investments make it the most aggressive investor in silicon photonics among the major AI chip companies, but it is far from the only player eyeing this market. The competitive landscape for AI data center optics is heating up rapidly, with established semiconductor giants, pure-play optics companies, and ambitious startups all vying for position.

Broadcom remains the dominant force in data center networking silicon, with AI semiconductor revenue reaching $20 billion in fiscal 2025 – up 65% year-over-year – and an AI-related order backlog exceeding $73 billion. Broadcom’s Tomahawk and Jericho switch families are the industry standard, and the company has been investing in its own optical connectivity solutions. In Q1 fiscal 2026, Broadcom’s AI revenue is expected to hit $8.2 billion, doubling year-over-year.

Intel has been a pioneer in silicon photonics research, having invested in the technology for over a decade. Intel’s silicon photonics products serve hyperscaler customers, and the company’s fabrication expertise gives it a natural advantage in manufacturing photonic integrated circuits at scale. However, Intel’s recent financial challenges and strategic restructuring have raised questions about its ability to invest aggressively in this space.

Cisco, with total FY 2025 revenue of $56.7 billion, remains the largest networking equipment company globally. Cisco’s optical networking portfolio includes transceivers, switches, and routers, and the company has been steadily integrating AI-optimized features across its product line. Cisco’s advantage lies in its massive installed base and enterprise relationships, though it has been slower than NVIDIA to pursue co-packaged silicon photonics.

“NVIDIA’s strategy of investing in optics suppliers rather than building its own is smart but risky,” noted Samik Chatterjee, analyst at JPMorgan. “It secures supply without the capital intensity of vertical integration, but it also means NVIDIA is dependent on partners executing flawlessly on a technology that is still maturing at scale.”

The Technical Architecture of Next-Generation AI Factories

NVIDIA’s vision for gigawatt-scale AI factories represents a dramatic departure from today’s data center architectures. Current AI training clusters typically connect thousands of GPUs using a combination of NVLink for intra-node communication and InfiniBand or Ethernet for inter-node networking. Silicon photonics promises to unify and simplify this architecture by replacing electrical interconnects at multiple levels of the network hierarchy.

At the chip-to-chip level, co-packaged silicon photonics enables direct optical connections between GPU packages, eliminating the need for separate transceiver modules. This approach uses advanced 200G SerDes technology to convert electrical signals to optical at the edge of the chip package, reducing signal path length from inches to millimeters and cutting latency by up to 50% compared to traditional pluggable optics.

At the rack-to-rack level, silicon photonics enables dense wavelength division multiplexing (DWDM), allowing dozens of data channels to travel simultaneously over a single optical fiber. This is critical for AI clusters that may span multiple buildings or even campuses, where copper interconnects simply cannot maintain the required bandwidth over longer distances.

At the system level, NVIDIA envisions tightly integrated “AI factories” where compute (GPUs), networking (switches and optical interconnects), and power delivery are designed as a single system. The company’s recent partnerships with power companies AES, Constellation, Invenergy, NextEra, Nscale Energy & Power, and Vistra to build AI factories that integrate electricity generation, storage, and grid coordination demonstrate this holistic approach.

“The future AI data center is not a building full of servers – it’s a manufacturing facility for intelligence,” said Bob Wheeler, Principal Analyst at The Linley Group. “Silicon photonics is the connective tissue that makes it possible to scale these facilities beyond what copper interconnects can support.”

Power Efficiency: The Hidden Crisis Driving Photonics Adoption

The power consumption challenge in AI data centers is driving silicon photonics adoption as urgently as bandwidth demands. Networking infrastructure – including switches, transceivers, cables, and cooling for these components – can account for 15-25% of total data center power consumption in AI-optimized facilities. As GPU power draws continue to climb (NVIDIA’s B200 GPU draws 1,000 watts, and the next-generation Rubin platform is expected to push even higher), reducing networking power consumption becomes critical to staying within facility power budgets.

The math is straightforward but sobering. A large AI training cluster with 100,000 GPUs might require 200,000 or more high-speed optical connections. At 30 watts per traditional 1.6 Tbps transceiver, networking alone would consume 6 megawatts – enough to power roughly 5,000 homes. NVIDIA’s co-packaged silicon photonics, at approximately 8-9 watts per equivalent connection, would reduce this to under 2 megawatts, freeing up 4 megawatts for additional compute capacity.

This power savings compounds at gigawatt scale. A gigawatt AI factory – the kind Jensen Huang has been describing – would contain millions of optical connections. The difference between traditional and co-packaged photonics could translate to hundreds of megawatts in power savings, worth tens of millions of dollars annually in electricity costs alone.

The AI data center power crisis has already prompted unprecedented investment in energy infrastructure. Hyperscalers are signing multi-billion-dollar power purchase agreements, investing in nuclear energy, and even exploring on-site power generation. Silicon photonics won’t solve the power crisis alone, but it addresses one of the most immediate and controllable sources of power waste in AI infrastructure.

Lumentum and Coherent: Profiling the Silicon Photonics Leaders

The two companies receiving NVIDIA’s $4 billion combined investment bring complementary strengths to the partnership. Understanding their capabilities helps explain why NVIDIA chose a dual-supplier strategy.

Lumentum Holdings (NASDAQ: LITE) is a pure-play optics and advanced manufacturing company headquartered in San Jose, California. The company’s core competencies include vertical-cavity surface-emitting lasers (VCSELs), edge-emitting lasers, and advanced photonic integrated circuits. Lumentum has historically served the telecommunications and 3D sensing markets, but has been rapidly pivoting toward AI data center applications. The NVIDIA investment will fund a new U.S.-based fabrication facility dedicated to producing next-generation laser components for silicon photonics.

Coherent Corp (NYSE: COHR), formed through the 2022 merger of II-VI Incorporated and the original Coherent, is a diversified photonics and electronics company. Coherent offers one of the broadest portfolios in the industry, spanning lasers, optical components, networking equipment, and materials science. The company’s dominance in the multi-mode optical transceiver market – particularly its advanced 200G and 400G solutions – positions it as a critical supplier for current and next-generation AI data center networking.

Both companies have seen their strategic importance rise dramatically as AI infrastructure spending has accelerated. The NVIDIA investments provide them with the capital to expand manufacturing capacity at a pace that would have been difficult to fund through operating cash flows alone, particularly given the scale of investment required for advanced photonics fabrication.

Market Impact: What This Means for AI Infrastructure Stocks

NVIDIA’s $4 billion silicon photonics commitment has ripple effects throughout the AI infrastructure investment landscape. The deal validates silicon photonics as a critical technology layer in the AI stack, potentially creating a new category of “AI optics” investments that complements the existing focus on GPUs, memory, and power.

“This is a defining moment for the optical networking industry,” said Stacy Rasgon, Senior Semiconductor Analyst at Bernstein Research. “NVIDIA doesn’t make $4 billion bets lightly. This signals that the company sees optical interconnects as fundamental to its roadmap – not a nice-to-have, but a must-have for scaling beyond current architectures.”

Company	Ticker	NVIDIA Investment	Key Technology	AI Revenue Exposure	Strategic Position
Lumentum	LITE	$2 billion	VCSELs, lasers, PICs	Growing rapidly	Pure-play optics, new fab
Coherent	COHR	$2 billion	Transceivers, materials	~30% of revenue	Broadest portfolio
Broadcom	AVGO	N/A (competitor)	Switch silicon, optics	$20B+ FY2025	Networking silicon leader
Intel	INTC	N/A (competitor)	Silicon photonics ICs	Moderate	Fab expertise, restructuring
Cisco	CSCO	N/A (competitor)	Networking equipment	Growing	Largest installed base
Marvell	MRVL	N/A (competitor)	DSPs, PAM4 optics	~35% of revenue	Custom silicon for cloud

The investments also highlight the increasing vertical integration of the AI supply chain. NVIDIA, once purely a chip designer, is now investing in optics, power infrastructure, networking software, and even AI model development. This expanding scope mirrors what we’ve seen from other Big Tech companies in their AI infrastructure spending, where the boundaries between chip design, system architecture, and infrastructure are blurring.

Historical Context: How AI Rewrote the Optical Networking Playbook

Silicon photonics has been a promising technology for decades, but AI has transformed it from a research curiosity into a strategic imperative. The history of optical interconnects in data centers follows a pattern familiar to many technologies: years of slow, incremental progress followed by a sudden demand shock that accelerates adoption by years or even decades.

Before 2023, optical interconnects in data centers primarily served the long-haul and metro networking markets – connecting data centers to each other across cities and continents. Inside data centers, copper interconnects handled most of the short-reach connections between servers and switches. The economics were simple: copper was cheaper and sufficient for the bandwidth requirements of traditional cloud workloads.

The AI training revolution changed this calculus entirely. When OpenAI trained GPT-4 in 2023, it used a cluster of roughly 25,000 NVIDIA A100 GPUs – a scale that pushed copper interconnects to their limits. By 2025, leading AI labs were building clusters of 100,000 or more GPUs, requiring bandwidth densities that copper simply cannot deliver efficiently. The physical limitations of copper – signal degradation over distance, electromagnetic interference, and heat generation – become acute at the speeds (400G and above) and densities required for AI workloads.

NVIDIA’s GTC 2025 co-packaged silicon photonics demonstration marked the technology’s transition from laboratory to commercial readiness. By March 2026, with the Lumentum and Coherent investments, NVIDIA has moved from demonstrating the technology to securing the supply chain needed to deploy it at scale. This rapid progression – from prototype to billion-dollar supply agreements in under a year – reflects the urgency of the AI infrastructure buildout.

The Bandwidth Bottleneck: Why Copper Can’t Keep Up with AI

Understanding the technical limitations driving the shift to silicon photonics requires examining the specific demands of modern AI workloads. Large language model training involves a technique called data parallelism, where training data is split across thousands of GPUs that must constantly synchronize their model weights. This synchronization – known as the “all-reduce” operation – generates enormous network traffic that is highly latency-sensitive.

A typical all-reduce operation on a 100,000-GPU cluster can generate petabits per second of aggregate network traffic. Every microsecond of network latency translates directly into wasted GPU cycles – expensive compute time where processors sit idle waiting for data. At current NVIDIA B200 GPU pricing, even a 1% reduction in GPU utilization due to network latency in a large cluster can cost millions of dollars annually in wasted compute.

Copper interconnects face three fundamental challenges at these scales. First, signal integrity degrades rapidly at speeds above 100 Gbps per lane, requiring power-hungry signal conditioning that adds latency. Second, copper cables generate significant heat, compounding the already severe thermal management challenges in dense GPU clusters. Third, copper is physically bulky compared to fiber optics, creating cable management challenges in facilities housing hundreds of thousands of connections.

Silicon photonics addresses all three challenges simultaneously. Optical signals maintain integrity over much longer distances without amplification, generate minimal heat, and fiber optic cables are dramatically thinner and lighter than copper alternatives. The co-packaged approach that NVIDIA is pursuing takes these advantages further by eliminating the electrical-to-optical conversion bottleneck at the transceiver, delivering what amounts to “native optical” communication between processors.

U.S. Manufacturing and Supply Chain Implications

A significant but underappreciated aspect of NVIDIA’s silicon photonics investments is their focus on U.S.-based manufacturing. Both the Lumentum and Coherent deals include provisions for building new domestic fabrication facilities, aligning with the broader semiconductor industry’s push to reshore critical manufacturing capabilities.

The CHIPS and Science Act, signed into law in 2022 and still deploying its $52.7 billion in semiconductor manufacturing incentives through 2026, has created a favorable environment for domestic photonics manufacturing investment. While most CHIPS Act funding has targeted traditional semiconductor fabrication, the advanced packaging and photonics segments are increasingly recognized as equally critical to the AI supply chain.

“Silicon photonics manufacturing has been concentrated in Asia, particularly Taiwan and China, which creates obvious supply chain vulnerabilities for a technology that will be foundational to AI infrastructure,” said Christopher Rolland, Senior Semiconductor Analyst at Susquehanna International Group. “NVIDIA’s investment in domestic manufacturing is as much about supply chain security as it is about securing supply.”

The geopolitical dimension adds urgency. As the U.S. tightens restrictions on AI chip exports to China – a policy landscape that has been evolving rapidly through 2025-2026 – ensuring domestic capacity for critical enabling technologies like silicon photonics becomes a national security consideration. NVIDIA’s complex relationship with China export controls makes securing U.S.-based supply chains even more strategically important.

Expert Analysis: Five Predictions for the Silicon Photonics Market

Based on NVIDIA’s investments and the current trajectory of AI infrastructure development, industry analysts and experts have outlined several predictions for how the silicon photonics market will evolve:

Prediction 1: Co-packaged optics will reach volume production by late 2027. NVIDIA’s investment timeline and the construction schedule for new fabrication facilities suggest that co-packaged silicon photonics will begin appearing in commercial AI systems within 18-24 months. Early deployments will likely target NVIDIA’s next-generation Rubin platform, which is expected to feature native optical interconnect support.

Prediction 2: The silicon photonics market will exceed $10 billion by 2029. Current estimates place the silicon photonics integrated circuit market at approximately $3.2 billion in 2025. With the combination of AI-driven demand, NVIDIA’s supply chain investments, and the technology’s inherent advantages in power efficiency, the market is positioned for roughly 25% annual growth over the next four years.

Prediction 3: At least two more major chip companies will make billion-dollar photonics investments by the end of 2026. NVIDIA’s move is likely to trigger competitive responses from AMD, which has been developing its own optical interconnect strategy, and potentially from cloud hyperscalers like Google and Amazon, which have the scale and motivation to invest directly in photonics technology.

Prediction 4: Optical interconnects will become the default for AI clusters exceeding 50,000 GPUs by 2028. As silicon photonics matures and costs decline through manufacturing scale, the economic crossover point – where optical becomes cheaper than copper on a total cost of ownership basis – will shift from ultra-large clusters to mid-scale deployments.

Prediction 5: Silicon photonics will drive a wave of M&A activity in the optical components sector. NVIDIA’s investments have highlighted the strategic value of photonics capabilities. Expect larger semiconductor and networking companies to acquire smaller photonics specialists, similar to the consolidation wave that reshaped the cybersecurity industry in 2025-2026.

Implications for the Broader AI Chip Ecosystem

NVIDIA’s silicon photonics strategy has implications that extend well beyond optics. The investments signal how the company views the future competitive landscape for AI infrastructure – and how it plans to maintain its dominant position as competitors like AMD, Intel, and custom chip designers at Google, Amazon, and Microsoft intensify their efforts.

By investing in optical interconnects, NVIDIA is essentially building a systems-level moat. Even if a competitor produces a GPU that matches Blackwell or Rubin in raw compute performance, that GPU would still need to plug into NVIDIA’s networking ecosystem – including its NVLink, InfiniBand, Spectrum Ethernet, and now silicon photonics interconnects – to deliver comparable system-level performance. This strategy mirrors what the company has done with CUDA in software: creating an ecosystem that is as important as the hardware itself.

The implications for Broadcom’s AI chip business are particularly interesting. Broadcom has built a massive business designing custom AI accelerators for hyperscalers, but its networking silicon – particularly its Tomahawk switch family – is used across the industry, including in NVIDIA-based systems. If NVIDIA’s co-packaged silicon photonics becomes tightly integrated with its switch silicon, it could create a more vertically integrated networking stack that reduces Broadcom’s addressable market.

For the broader AI chip market, NVIDIA’s photonics investments underscore a shift from chip-level competition to system-level competition. The companies that will win in AI infrastructure are those that can deliver complete, optimized systems – from silicon to software to interconnects to power – rather than individual components. This raises the bar for competitors and could accelerate industry consolidation.

What NVIDIA’s Silicon Photonics Bet Means for AI’s Future

NVIDIA’s $4 billion silicon photonics investment is more than a supply chain play – it’s a statement about the future of AI infrastructure. By committing this level of capital to optical interconnect technology, NVIDIA is signaling that the next phase of AI scaling will be defined not just by faster chips but by the ability to connect those chips into increasingly massive, efficient systems.

The implications are profound. If silicon photonics delivers on its promise of 3.5x power reduction and dramatically lower latency, it could unlock AI system architectures that are simply impossible with copper interconnects. Clusters of millions of GPUs, spanning multiple buildings, communicating at the speed of light with minimal energy overhead – this is the vision that NVIDIA’s investment enables.

For the broader technology industry, the investment validates a thesis that has been building for years: the bottleneck to AI progress is shifting from compute to connectivity. The companies that solve the interconnect problem – whether through silicon photonics, advanced packaging, or novel network architectures – will be as important to the AI revolution as the chip designers themselves.

As Jensen Huang pursues his vision of gigawatt-scale AI factories, silicon photonics stands as the critical enabling technology that will determine whether that vision becomes reality. NVIDIA’s $4 billion bet suggests the company is confident it will – and is positioning itself to profit from every photon.

Frequently Asked Questions

What is silicon photonics and why does it matter for AI?

Silicon photonics is a technology that uses light (photons) instead of electrical signals (electrons) to transmit data between computer chips. It matters for AI because modern AI training clusters require thousands of GPUs to communicate at extremely high speeds. Traditional copper interconnects are hitting physical limits in bandwidth, latency, and power consumption at these scales. Silicon photonics offers 3.5x better power efficiency and dramatically lower latency, enabling the construction of larger, more efficient AI systems.

How much did NVIDIA invest in silicon photonics companies?

NVIDIA invested a total of $4 billion – $2 billion in Lumentum Holdings and $2 billion in Coherent Corp. Both investments include growth equity stakes plus multibillion-dollar purchase commitments for advanced laser components and future capacity access rights. The deals also fund new U.S.-based fabrication facilities for photonics manufacturing.

When will co-packaged silicon photonics be available in commercial AI systems?

Based on current investment timelines and manufacturing facility construction schedules, co-packaged silicon photonics is expected to reach volume production by late 2027. Early deployments will likely target NVIDIA’s next-generation Rubin GPU platform. Some prototype and limited-production systems may appear sooner for selected hyperscaler customers.

How does silicon photonics compare to traditional copper interconnects?

Silicon photonics offers several advantages over copper: approximately 3.5x lower power consumption per connection, significantly reduced latency (signal paths measured in millimeters vs. inches), higher bandwidth density over longer distances, reduced heat generation, and physically smaller cables. The main current disadvantage is cost, though this gap is narrowing as manufacturing scales and AI cluster sizes make the total cost of ownership comparison favorable for optics.

What companies compete with NVIDIA in silicon photonics?

Key competitors include Broadcom (dominant in networking switch silicon with $20 billion+ in AI semiconductor revenue), Intel (a decade-long silicon photonics researcher with fabrication expertise), Cisco (the largest networking equipment company with $56.7 billion in annual revenue), and Marvell (a growing player in custom silicon and PAM4 optics for cloud customers). Cloud hyperscalers like Google and Amazon are also developing internal optical interconnect capabilities.

Why did NVIDIA invest in two separate photonics companies instead of one?

NVIDIA’s dual-supplier approach serves multiple strategic purposes: it reduces supply chain risk by avoiding dependence on a single vendor, it uses complementary strengths (Lumentum’s laser expertise vs. Coherent’s broader portfolio including transceivers and materials), and it helps grow the overall silicon photonics ecosystem. Both agreements are nonexclusive, meaning the suppliers can also serve NVIDIA’s competitors, which further encourages ecosystem growth.

Related Coverage

• The AI Data Center Power Crisis: How Big Tech’s 125 GW Appetite Is Reshaping the US Energy Grid
• Big Tech’s $700 Billion AI Infrastructure Bet: Inside the 2026 Spending Race
• Broadcom’s AI Revenue Surges 106% to $8.4 Billion: Inside the Custom Chip Empire
• NVIDIA Blackwell GPU Pricing: B200, B300 and DGX Cost Breakdown
• Nvidia Restarts H200 Chip Sales to China: Inside the Policy Reversal
• NVIDIA’s Next-Gen Blackwell Ultra GPUs: What We Know So Far
• The $96 Billion Cybersecurity M&A Wave: How Industry Consolidation Is Reshaping Security