NVIDIA and Corning to Build 3 US Factories for AI Optics

NVIDIA and glass technology giant Corning have announced a major partnership to build 3 new advanced manufacturing facilities in the United States, dedicated entirely to developing co-packaged optics (CPO) and optical communication technology for AI infrastructure. The joint venture, which spans sites in North Carolina and Texas, will create at least 3,000 jobs and marks one of the largest domestic investments in AI-related optical infrastructure to date.

The announcement sent Corning's stock surging 14%, while NVIDIA shares climbed nearly 3%, signaling strong investor confidence in the strategic direction of both companies.

Key Takeaways at a Glance

• 3 new factories will be built across North Carolina and Texas
• At least 3,000 jobs will be created through the partnership
• Corning's US optical communication manufacturing capacity will increase 10x
• US fiber optic production capacity will grow by more than 50%
• Corning issued $500 million in warrants to NVIDIA as part of a long-term partnership agreement
• NVIDIA received warrants to purchase up to 15 million shares of Corning common stock at $180 per share

Why Optical Communication Is the Next AI Bottleneck

As AI models grow exponentially in size and complexity, the traditional copper-based interconnects that link GPUs and other processors inside data centers are hitting fundamental physical limits. Heat generation, signal degradation, and power consumption all worsen as data transfer speeds increase through copper wiring.

Co-packaged optics represents a paradigm shift. Instead of routing electrical signals through copper traces and then converting them to light at the edge of a rack, CPO technology integrates optical components directly onto or adjacent to the processor package itself. This dramatically reduces latency, cuts power consumption, and enables data transfer rates that copper simply cannot match.

For NVIDIA, whose GPU clusters power the vast majority of the world's AI training workloads, solving the interconnect bottleneck is not optional — it is existential. The company's next-generation Blackwell and future architectures demand bandwidth densities that only optical solutions can deliver at scale. Partnering with Corning, the world's leading manufacturer of specialty glass and optical fiber, gives NVIDIA a vertically integrated supply chain for this critical component.

The $500 Million Warrant Deal Explained

Beyond the factory announcements, the financial structure of the partnership reveals how deeply both companies are committing to this relationship. Corning and NVIDIA signed a securities purchase agreement that includes 2 distinct warrant instruments.

First, NVIDIA received traditional warrants allowing it to purchase up to 15 million shares of Corning common stock at a strike price of $180.00 per share. Second, Corning issued pre-funded warrants for up to 3 million additional shares at a nominal exercise price of just $0.0001 per share — essentially giving NVIDIA an immediate equity stake.

The total value of the warrant package is approximately $500 million. While the companies did not disclose additional financial terms, the structure mirrors similar strategic partnerships in the semiconductor industry where a major customer takes an equity position in a key supplier to ensure priority access and aligned incentives.

• Traditional warrants: 15 million shares at $180/share — a bet on Corning's long-term stock appreciation
• Pre-funded warrants: 3 million shares at $0.0001/share — an immediate, near-free equity stake
• Strategic alignment: NVIDIA gains financial incentive to help Corning succeed
• Supply assurance: The equity relationship helps lock in manufacturing capacity for NVIDIA's needs

This type of arrangement is increasingly common in the AI supply chain. For comparison, similar warrant-based partnerships have been seen between cloud hyperscalers and chip manufacturers, though the scale of this deal — focused specifically on optical technology — is unprecedented.

Manufacturing Scale: A 10x Capacity Leap

The numbers behind the factory expansion are striking. Corning has committed to increasing its US-based optical communication manufacturing capacity by a factor of 10. Simultaneously, the company's domestic fiber optic production capacity will grow by more than 50%.

These figures reflect the enormous anticipated demand for optical interconnects as AI data centers scale from thousands to tens of thousands — and eventually hundreds of thousands — of GPUs per cluster. Current-generation data centers operated by companies like Microsoft, Google, Amazon, and Meta are already pushing the limits of electrical interconnect technology.

The choice of North Carolina and Texas as manufacturing locations is strategic. North Carolina is already home to a significant portion of Corning's US operations and boasts a skilled workforce in advanced manufacturing. Texas, meanwhile, has become a hub for semiconductor and data center investment, with companies including Samsung, Texas Instruments, and Tesla all expanding their presence in the state.

The 3,000 jobs created by these factories will span a range of roles, from advanced manufacturing technicians to optical engineers and research scientists. This job creation comes at a time when the US government is actively encouraging domestic semiconductor and AI infrastructure investment through initiatives like the CHIPS and Science Act.

The Broader AI Infrastructure Arms Race

Since OpenAI launched ChatGPT in late 2022, investment in AI infrastructure has reached historic levels. The race to build bigger, faster, and more efficient data centers has driven unprecedented demand for GPUs, networking equipment, cooling systems, and now optical interconnects.

NVIDIA sits at the center of this boom. The company's market capitalization has made it the world's most valuable semiconductor firm, and its H100 and B200 GPUs are the workhorses of modern AI training. But as NVIDIA's CEO Jensen Huang has repeatedly emphasized, the GPU is only one piece of the puzzle. The networking fabric that connects thousands of GPUs into a unified computing cluster is equally critical.

This is where CPO technology enters the picture. Current optical transceiver modules — the devices that convert electrical signals to light for transmission over fiber — are separate, pluggable components. They consume significant power and add latency. Co-packaged optics eliminates this separation by bringing the optical engine directly onto the switch or processor package.

• Current approach: Pluggable optical transceivers sit outside the chip package
• CPO approach: Optical engines are integrated directly with the processor
• Power savings: CPO can reduce interconnect power consumption by 30-50%
• Bandwidth gains: Enables multi-terabit-per-second links per port
• Density improvements: Frees up valuable rack space in data centers
• Latency reduction: Shorter signal paths mean faster data movement between GPUs

What This Means for the Industry

The NVIDIA-Corning partnership signals that optical interconnects are transitioning from a future technology to a present-day priority. For data center operators, this means planning for a fundamental shift in how computing clusters are designed and built.

For cloud providers and enterprises building AI infrastructure, the message is clear: the next generation of AI systems will require optical connectivity at a scale that today's supply chain is not yet equipped to deliver. The 10x capacity expansion Corning is undertaking reflects the magnitude of the gap between current supply and projected demand.

For investors, the deal validates the optical communication sector as a critical link in the AI value chain. Corning's 14% stock jump reflects the market's recognition that the company is positioning itself as an indispensable partner to the world's most important AI chipmaker.

For competitors like Broadcom, Marvell, and Intel — all of whom are developing their own CPO and optical interconnect solutions — the NVIDIA-Corning partnership raises the competitive stakes. NVIDIA's willingness to invest $500 million in equity warrants and co-develop manufacturing capacity suggests the company is building a proprietary optical supply chain that rivals may struggle to match.

Looking Ahead: The Road to Optical AI Data Centers

The factories announced this week will take time to build and ramp to full production, likely entering meaningful output in 2026 or 2027. This timeline aligns with NVIDIA's product roadmap, which is expected to introduce increasingly optical-centric networking solutions in its next-generation platforms.

The long-term vision is a data center where every GPU-to-GPU connection runs over optical fiber, with co-packaged optics eliminating the electrical bottleneck entirely. Achieving this vision requires not just new chips, but entirely new manufacturing capabilities for optical components at scale — exactly what the NVIDIA-Corning partnership aims to deliver.

As AI models continue to grow — with training runs now consuming tens of thousands of GPUs over weeks or months — the interconnect fabric becomes the defining constraint on system performance. The companies that solve this challenge will shape the next era of artificial intelligence infrastructure. With this week's announcement, NVIDIA and Corning have placed a very large bet that optical technology is the answer.