Turn an article from #英伟达#: NVIDIA is no longer just a company but an AI empire.
Many still see NVIDIA as a chip-selling company, but I increasingly believe this understanding is outdated. When I break down this investment map, a clearer logic begins to emerge: — this isn't just investing, it's "orchestrating the entire AI demand chain."
If you only look at the surface, you might think it's diversified investment, but if you connect all the nodes, you'll find they all point to the same endpoint: GPU demand.
(1) First, look at the bottom layer. $Intel INTC$ $Synopsys SNPS$
This is the chip design and manufacturing capability, essentially ensuring that the entire semiconductor ecosystem won't become a bottleneck. You can think of it as: even if competition intensifies in the future, the fundamental design ability remains within the system.
(2) Next, a layer above, is networking and optical interconnection. $Lumentum LITE$ $COHR $MRVL
These companies solve the same problem — how data flows at high speed between AI clusters. Without this layer, even the strongest GPU would be bandwidth-limited. In other words, this layer determines whether "computing power can be truly utilized."
(3) Then, the most critical layer — compute distribution. $CRWV $NBIS and GPU cloud providers like Nscale, Lambda, essentially turn NVIDIA's GPUs into "rentable resources" rather than hardware sold once. This step is crucial because it changes the business model: from selling devices → to a continuous leasing ecosystem.
(4) Next is the model layer. OpenAI, xAI, Mistral, Cohere, Perplexity
This layer is the real black hole for "computing consumption." All model companies are fundamentally the biggest consumers of GPUs. You'll notice an interesting structure: NVIDIA isn't investing in "one winner," but in "all potential winners." No matter who wins, GPUs will be consumed.
(5) Moving up, are applications and frontier scenarios. Autonomous driving (Wayve), robotics (Figure AI), AI in pharmaceuticals ($LLY, $RXRX), even nuclear fusion (Commonwealth Fusion). These may seem scattered, but fundamentally, there's only one: — they turn AI from "software" into "real-world demand."
Once AI enters the physical world, the demand for computing power no longer grows linearly but exponentially. This is also why NVIDIA's layout includes "seemingly unrelated" tracks. Because it bets not on an industry, but on the "demand explosion point."
So now, returning to a more core question: why is NVIDIA investing so broadly? Because it is doing something very few companies can — simultaneously controlling the supply side (chips + networking + compute) and laying out the demand side (models + applications + scenarios).
What does this mean?
It means it’s not just participating in the AI cycle but "amplifying the AI cycle." Many see AI as a competition, but NVIDIA's approach is more like: — not betting on who wins, but ensuring everyone must use my chips.
Looking at this map from this perspective reveals a deeper structure: every investment reinforces a closed loop:
Investment → Demand growth → GPU consumption → Revenue increase → Reinvestment — this is the true flywheel.
Many still see NVIDIA as a chip-selling company, but I increasingly believe this understanding is outdated. When I break down this investment map, a clearer logic begins to emerge: — this isn't just investing, it's "orchestrating the entire AI demand chain."
If you only look at the surface, you might think it's diversified investment, but if you connect all the nodes, you'll find they all point to the same endpoint: GPU demand.
(1) First, look at the bottom layer. $Intel INTC$ $Synopsys SNPS$
This is the chip design and manufacturing capability, essentially ensuring that the entire semiconductor ecosystem won't become a bottleneck. You can think of it as: even if competition intensifies in the future, the fundamental design ability remains within the system.
(2) Next, a layer above, is networking and optical interconnection. $Lumentum LITE$ $COHR $MRVL
These companies solve the same problem — how data flows at high speed between AI clusters. Without this layer, even the strongest GPU would be bandwidth-limited. In other words, this layer determines whether "computing power can be truly utilized."
(3) Then, the most critical layer — compute distribution. $CRWV $NBIS and GPU cloud providers like Nscale, Lambda, essentially turn NVIDIA's GPUs into "rentable resources" rather than hardware sold once. This step is crucial because it changes the business model: from selling devices → to a continuous leasing ecosystem.
(4) Next is the model layer. OpenAI, xAI, Mistral, Cohere, Perplexity
This layer is the real black hole for "computing consumption." All model companies are fundamentally the biggest consumers of GPUs. You'll notice an interesting structure: NVIDIA isn't investing in "one winner," but in "all potential winners." No matter who wins, GPUs will be consumed.
(5) Moving up, are applications and frontier scenarios. Autonomous driving (Wayve), robotics (Figure AI), AI in pharmaceuticals ($LLY, $RXRX), even nuclear fusion (Commonwealth Fusion). These may seem scattered, but fundamentally, there's only one: — they turn AI from "software" into "real-world demand."
Once AI enters the physical world, the demand for computing power no longer grows linearly but exponentially. This is also why NVIDIA's layout includes "seemingly unrelated" tracks. Because it bets not on an industry, but on the "demand explosion point."
So now, returning to a more core question: why is NVIDIA investing so broadly? Because it is doing something very few companies can — simultaneously controlling the supply side (chips + networking + compute) and laying out the demand side (models + applications + scenarios).
What does this mean?
It means it’s not just participating in the AI cycle but "amplifying the AI cycle." Many see AI as a competition, but NVIDIA's approach is more like: — not betting on who wins, but ensuring everyone must use my chips.
Looking at this map from this perspective reveals a deeper structure: every investment reinforces a closed loop:
Investment → Demand growth → GPU consumption → Revenue increase → Reinvestment — this is the true flywheel.
