NVIDIA's growth has taken a nosedive.
We're not talking about a slight dip here—its annual growth has plummeted from 265% all the way down to 75%. Meanwhile, a company you might not have heard much about—Micron—has seen its growth surge from a staggering -57% to a remarkable 200%.
These two trends are set to cross in Q4 2024.
In the gold rush of AI, the most profitable tool is shifting from GPUs to memory.
This shift is easy to understand. Between 2023 and the first half of 2024, everyone was scrambling for NVIDIA's GPUs—after all, you need graphics cards just to get into the game of training large models. During that period, NVIDIA was churning out quarterly growth of over 250%, like a money-printing machine.
But here’s the catch.
What happens after the models are trained? As AI transitions from the "training era" to the "inference era"—which is when it truly starts serving users at scale—the bottleneck changes. Running inference isn’t just about computing power; it’s all about bandwidth and memory.
Think about it: a large model can have hundreds of GBs of parameters that need to be crammed into high-bandwidth memory (HBM) for real-time responses. No matter how powerful the GPU, if the memory can't keep up, it’s still going to bottleneck.
It’s like building a highway: no matter how wide the road is, if there are only two toll booths, you’re still going to hit traffic.
Micron happens to be the one building those "toll booth windows." It’s one of the core suppliers of HBM globally, and when demand for AI inference skyrocketed, its orders took off too.
To be honest, NVIDIA's 75% growth would be explosive in any other industry, but the capital market is all about trends—dropping from 250% to 75% changes the narrative.
On the flip side, Micron’s jump from negative growth to 200% is the real story that has the market buzzing.
However, let’s temper our excitement a bit: Micron's massive surge is largely due to a low base effect. It dropped over 50% in 2023, so rebounding now looks impressive. We’ll need to see if that 200% can sustain over several quarters.
But the trend doesn’t lie—the profit center of the AI supply chain is moving from upstream computing power to midstream infrastructure.
The next beneficiaries may not be chip manufacturers, but those producing cooling systems, power supplies, and data centers.
The gold miners keep changing hands.
The truly smart money is always on the lookout for the next shovel.
Top comments (0)