#SecretHistory of 2000.
How we are about to see a repeat of 2000.
AI is a bubble in hardware. But AI has still not started.
What you see below is market cap of Cisco over last three decades. At the time of dot com boom, CISCO briefly became world's most valuable company.
As you can see it had market cap of nearly $387 B in March 2000.
Internet was kind of in play since 1995. But truly it took off in 2000 after companies sighed relief after Y2K.
CISCO was truly the connection to Internet and possibly only connection even though TCP/IP protocol on which the Internet runs is open.
When the gold rush started CISCO was the seller for the shovels and other tools needed by the prospectors. So, there was a bubble.
So, after reaching its zenith, in March 2000 Cisco started its descent. This is classic bubble going bust. As hardware manufacturer Cisco could produce cutting edge products and lowest cost only for some time before the competition was on its heel. I remember John Chambers being the darling of the press just like Jensen Huang of NVDA now.
Fast forward to 2024 we are seeing exactly same bubble in NVDA. The company briefly went to $3T and became world's most valuable company just like Cisco did in April 2000 and exactly for the same reasons.
NVDA has 88% monopoly in GPUs and likely positioned itself nicely because of two historical reasons.
The crypto currency and block chain was considered a fad by other players whereas NVDA took the bait. But then CUDA happened which allowed all software engineers access to parallel execution that speeded up tensor calculations which were how the machine learning, deep learning applications achieved speed.
AI has been there for 50 odd years. But the advent of transformers moved this away from NLP which had been stuck in the region of no progress for two decades and AI took off. This is where NVDA came in. CUDA was the horse for the NVDA cart. AI was the destination.
This is very similar to companies suddenly jumping on to the bandwagon of Internet immediately after Y2K. Everyone wanted to be there, and CISCO was the only horse cart. Today every company wants to use the AI for strategic advantage and cost cutting. And NVDA GPUs are till recently in high demand and there is bubble.
But just like what happened to Cisco, NVDA as a hardware play will lose its pole position.
Specifically, there are newer approaches. When usage of tensors became more ubiquitous, companies like Google built own custom tensor processing units called TPUs. Grok, I hear is fast because of its own custom chip like solution which they call LPU. So sooner or later NVDA will lose its uniqueness as other hardware companies catch up.
CISCO lost its uniqueness in approximately two years and dropped to 13 percent of its hay day value in September 2002. NVDA too will follow similar trajectory.
Please do not get me wrong. We have not even seen proverbial tip of iceberg for AI. I have seen most insane, proverbial earth shattering never before seen wave coming for AI. We will have companies whose valuations will far outstrip the $10T in this cycle. Just like the Internet in 2000 had hardware company Cisco reaching valuation of $387 B in 2000 and the later software plays using Internet easily broke this valuation figure with MSFT, META breaking $1T barrier with pure software play and AMZN and AAPL breaking it by mixed play with software being the reason to scale. We truly are at the very beginning of AI wave. But first bubble has to pop.