Skip to main content

$NVDA: When You are The Moat

NVIDIA had their earnings call yesterday for the quarter ending Dec'23. Markets were muted in anticipation. As expected, the S&P 500 rose by 2.5% on the back of a strong performance and pipeline.

The day after, NVIDIA stock rallied to all time high of $800. This gave the company a market cap of USD 2 Tn, surpassing Alphabet, Inc., and becoming the fourth largest listed company in the world by market value. 

For perspective consider this - the single day gain of USD 277Bn was bigger than the largest listed company in India - the world's 4th biggest equity market, and by an estimate its market cap was now larger than the entire SENSEX of India.

Who knew? Perhaps not even Berkshire Hathaway. (See share holding pattern in the links below).

One of the simplest reasons for the meteoric rise of NVIDIA is, as Warren Buffet once famously said about resilient businesses, that NVIDIA provides a moat to the the software firms for their business of developing and productising AI and, specifically, GenAI. 

Imagine a very, very large Excel sheet, where every cell is linked with other cells using some simple formula. That is, if you make a change in any one cell, the cascading effect will be seen across the entire grid where every cell would require a re-computation. Such a computation need not be intensive at a cell level, but the sheer quantity (ie the number of cells require re-computation concurrently) can be overwhelming even for the fastest of Intel CPUs. 

AI model training and data retrieval is similar. It requires computation that is very concurrent in nature. A large Language Model (LLM) for a GenAI such as ChatGPT typically has billions of parameters, where change in each parameter can affect all other parameters.  

Unprecedented resources are required for a task such as this. A large portion of those resources are GPUs - specialised computer chips designed to handle 'concurrent' calculation.  NVIDIA is the current market leader in developing and supplying such GPUs to the resource hungry AI companies.

Apart from the availability of very large quality dataset and algorithm development itself, this resource intensive nature of developing and maintaining LLMs raises barriers of entry in the market. A very important force in the industry. 

This is the moat.

Now, in May last year, an internal memo from Google leaked. It got famous by the meme- "We do not have a moat, and neither do OpenAI". The conversation of the memo centred around the resource availability of large dataset and the algorithm development at Google and OpenAI, and how Open Source models are catching up. "While our models still hold a slight edge in terms of quality, the gap is closing surprisingly quickly." lamented the memo.

So, when software doesn't hold as a moat, it has to default to hardware.

Therefore, NVIDIA.

But there is a problem.

This moat has a price-tag. (which moat doesn't? Well, those that do not, and are shrouded in casual ambiguity become your competitive advantage!)

It was perhaps this moat building that took Sam Altman to Saudi Arabia, and the rumours that followed about the trillion dollar investments towards chip development. True or not, the threat perception of proprietary GenAI such as ChatGPT and Gemini from the Open sourced and funded models such as Llama3 by Meta are high.

To cement its entrench position, NVIDIA is said to be in conversation with Alphabet, Amazon, Meta, Microsoft and OpenAI to build custom chips for their respective positions. It is not difficult to predict how this might play out. There can be only one Wintel-like alliance. Meta has announced an intent to purchase upto 350,000 NVIDIA H100 GPUs, taking the total stockpile to 600,000 GPUs. At a discounted rate of USD25,000 a piece, this is more than USD 15 Bn for the boxes alone.

As an analyst observed, "The people who made the most money in the Gold Rush of mid-1800s were the ones providing the tools to get the job done, and not those hunting for the precious metal. NVIDIA is effectively playing the same role today in this tech revolution." 

Therefore, what the moat is guarding is an entirely separate issue. 

In the next post we will look into the hype, the Concentration of power, the de-centralization and true democratisation of AI.

Did you read anything interesting this week on AI? Would love to know. Drop in a comment!

 

Moat and the Gold Rush
A GPU moat guarding the castle of shovel-makers during the Gold Rush of mid-1800s.

References and Further reading:

Comments

Post a Comment

Popular posts from this blog

The Independent Directors at OpenAI

Sam Altman was the CEO and  Greg Brockman  was the chairman of the board  at OpenAI.org, the parent company that is listed as a not-for-profit organization in the US u/s 501(C)(3).   On 17 Nov 2023 both of them were fired by the Independent Directors of the board. This post talks about the 4-day drama that ensued at the back of these events, focusing on the role of Independent Directors. (Try here for a related earlier post.) One year ago the company launched the ChatGPT, the Large Language Model, that rose to prominence with its Generative AI capabilities (“GPT” or Generative Pre-trained Transformer) and human-like response and interactive interface (“Chat”). At launch ChatGPT was based on based on GPT-3.5 series. The launch took the internet by storm as Microsoft unveiled its commercial partnership with the firm, and its global marketing machine geared into action.  To accommodate for this new profit-making "partnership" endeavor, the firm came up with anothe...
Post a Comment
Read more

OpenAI and the Network Effect (ft. Md Rafi and Ola Krutrim)

"Who is the greatest Bollywood singer of all times?" I typed into chat.krutrim.com It listed seven, but missed Mohammad Rafi.  Horrified, I followed up, "Why is Mohammad Rafi not in this list?"  And it missed the context, replying, "Mohammad Rafi is not in the list because the list you are referring to is not provided." With a deep sigh, it reminded me of Altman's India visit June last year. Someone asked him if India should invest in building a Foundational model (assuming funding and talent is not as issue). And he replied , "it would be hopeless to compete with us on training foundation models.. you shouldn’t try”. Try they will, and they should. The world's fourth(?) largest economy has pockets of deep pockets that can sustain the demands of developing a resource hungry technology such at Foundational LLMs. But distribution, diffusion and monetisation remains challenging, when chatGPT, Copilot and Gemini in Indic languages are just an App ...
Post a Comment
Read more