Ian Ritchie. The Herald Business HQ. Friday 5th July 2024
In the last few weeks, graphics chip manufacturer Nvidia briefly became the world’s most valuable company when it exceeded a market value of $3.3 trillion, overtaking the other tech giants of Microsoft, Apple, Alphabet and Amazon.
For many years, Nvidea has been the leading designer and supplier of graphics processing units (GPU) chips, which enable personal computers and games consoles to generate and display extraordinary detailed, realistic 3D moving images.
Although this was a significant market for the company it didn’t attract a huge valuation. Eight years ago, Nvidea was worth less than 1% of today’s value.
The company’s fortunes began to take off when it emerged that its processors were suitable for mining Bitcoins.
And then, the massively parallel processing of GPU’s large data sets inherent in the creation of complex 3D images also turned out to be useful in enabling artificial intelligence (AI) processing. As a result, Nvidea has also become the leader of this fast-growing emerging field. It was smart enough to quickly adapt its business model to exploiting this task and is now selling systems for AI processing to data centres at over $1 million each.
It has been often said that during the Californian gold rush people that made most of the money were not the gold prospectors themselves but instead the traders who sold them the picks and shovels.
Nvidea finds itself today in the envious position of selling most of the shovels enabling today’s AI “gold rush”.
And we are certainly in the middle of a gold rush – where nobody is yet making any money out of AI, but everybody believes that there will be massive returns to be made once the new business models settle and people begin to deploy AI systems in earnest, and in volume, to real tasks.
It’s a bit reminiscent of the “dotcom” boom of the late 1990s when people were working out how to make revenue out of the newly emerging World Wide Web. Initially, companies developing web applications were valued very highly but once the dust settled and the winners and losers became more obvious there was an inevitable crash.
Which begs the question: is the massive valuation of today’s Nvidea justified based on its prospects?
Sandeep Gupta, an analyst at Barclays, argued earlier this year that Nvidia’s large market share would be hard to maintain given the increasing number of rivals and questioned how Nvidia’s customers would monetise their AI software.
Although they are currently the undisputed leader in AI processing, with 80% of the market, others are snapping at their heels. Many competitors and commentators argue that the GPU was designed in the 1990s for 3D gaming acceleration and is not optimised for AI tasks, carrying unnecessary overheads in its processing.
Glenn O’Donnell, senior vice president and analyst with Forrester Research says, “It’s burning power and using circuitry that’s not really necessary”.
Certainly, managing today’s AI tasks using GPUs is relatively expensive – it costs at least $200 million to build an AI large language model (LLM) model and it is estimated that the computer processing required to develop the answer for each individual query costs around $0.36, putting a relatively heavy overhead on anybody designing an AI business activity.
Leading established computer chip companies such as Intel (with its Gaudi3 processor) and AMD (with its Radeon GPU) have developed architectures they claim are suited for AI tasks, and several new companies claim to be developing significantly better ways to process LLMs.
Silicon Valley based start-ups such as SambaNova, Cerebras, Groq, and xAI, are all developing new dedicated AI processors and attracting huge backing.
Elon Musk’s company, xAI, just raised a Series B funding round of $6 billion this May from seasoned Silicon Valley investors Andreessen Horowitz, Sequoia Capital, and others. Here in the UK, Bristol based GraphCore is providing a ‘Software as a Service’ solution offering access to their AI servers in the cloud.
At the University of Edinburgh, a group led by Professor Themis Prodromakis is developing a hugely innovative new AI processing architecture based on memristors, a non-digital processing system which they claim will be able to deliver two orders of magnitude more efficiently than existing GPUs, also reducing the price per AI calculation to around 1% of today’s cost.
Of course, it is unlikely that Nvidea, which generated $14.88 billion in profit in its most recent quarter, will be easily beaten. It will undoubtedly be undertaking massive amounts of R&D to ensure that it can hold on to its market lead, and it also now has the resources to easily identify and acquire new emerging technology companies.
It's all still to play for.
AI is undoubtedly coming and will transform our economy, but it would be brave person to predict who will finally prevail.
Commenti