NVIDIA's Supply Chain Empire

Today, NVIDIA announced a $500 million investment in Corning. This funding will be used to expand Corning’s optical connectivity manufacturing capacity in the United States by ten times, increase fiber optic capacity by over 50%, and build three new factories in North Carolina and Texas. This capacity expansion includes the construction of three advanced manufacturing plants in North Carolina and Texas; the additional capacity will supply optical connectivity products for ultra-large data centers, supporting their large-scale deployment of NVIDIA’s accelerated computing clusters.

Relying on a long-term strategic partnership with NVIDIA, Corning has sold two warrants to NVIDIA:

1. A call warrant: exercisable at $180 per share, with a maximum subscription of 15 million Corning (GLW) shares;

2. A prepaid warrant: with an exercise price of only $0.0001 per share, with a maximum subscription of 3 million shares;

The total transaction value of the two warrants is $500 million.

Looking at NVIDIA’s investment landscape over the past two years, a phenomenon becomes apparent: NVIDIA is using real money to reshape the entire AI industry supply chain. Control over the industry chain has become the most difficult barrier to cross in the AI era.

NVIDIA’s Investment Landscape

Over the past two years, NVIDIA has made strategic moves across various segments of the AI industry chain. We have compiled publicly available information into the following diagram:

As shown in the diagram, NVIDIA’s investments cover everything from the most fundamental chip manufacturing to top-tier AI applications, nearly every key link has a presence.

Collaboration with Corning

Corning previously stated that it would only expand production after seeing customer prepayments. NVIDIA’s $500 million investment is actually a signal to the market: in the next 2-3 years, fiber optic demand for AI data centers will grow more than tenfold.

This is not just NVIDIA’s own demand, but the entire AI industry’s demand. However, by locking in capacity early, NVIDIA has achieved one thing: when competitors want to build large-scale AI clusters, they may find that fiber optic supplies are already fully booked.

Domestic conditions also reveal some clues. Since this year, fiber optic prices have risen dramatically, and prices are expected to continue increasing in the future.

Supply Chain Control

In a situation where the entire industry chain is experiencing shortages, supply chain management has become a huge moat.

NVIDIA’s ability to control the supply chain is unmatched by other competitors. This capability is reflected not only in technology but also in capital and strategy.

The core strategy of NVIDIA’s cooperation with Corning is “invest and lock in capacity early.” This is fundamentally different from traditional supply chain collaborations.

This reminds me of long-term contracts in the storage industry. The big difference between this wave of storage long-term agreements and the previous one is that many buyers want to participate in the original manufacturers’ capital expenditures. In other words, they want to bind themselves into a community of shared interests, rather than just signing a volume and price locking agreement like in the past.

Why this change? Because everyone realizes that in an era of tight capacity, relying solely on contracts is not enough. Only by investing real money can they ensure priority in the supply chain.

Speaking of supply chain locking, we must mention OpenAI.

Last week, there were reports that OpenAI encountered some issues. We analyzed that its main problem lies in product strength; in terms of actual productivity, there is indeed a gap compared to Claude. However, OpenAI’s biggest advantage still lies in supply chain capacity.

Last year, OtterMan frequently signed cooperation agreements with supply chain companies and made various big promises, essentially locking in capacity. This will be a significant advantage for OpenAI in the future.

Because in the AI industry, products can be iterated, and technology can be caught up with, but capacity takes time to build. When competitors want to scale up, they may find TSMC’s wafer capacity fully booked, data center power capacity insufficient, and fiber optic supplies unavailable.

This is the power of the supply chain.

Domestically, we understand that CSPs have also been actively stockpiling recently.

Therefore, managers of AI and semiconductor industry chains should consider that if certain materials might be seized by AI for capacity, it’s best to lock in those supplies early. Many products in the AI industry chain are no longer in the era of paying first and receiving goods later.