Blain’s Morning Porridge 18th April 2024: AI Bubble Bursting? Or just reaching its expansion limit?
“Ford, there’s an infinite number of monkeys outside who want to talk to us about this script for Hamlet they’ve written.”
The AI bubble is running out of steam… which may be a very apt metaphor. Value in AI is driven by supply scarcity, which will be addressed by competition and capacity. The next issue is how AI innovation is over-exaggerated – we will require much more powerful computational tools to create next level Intelligence.
Major crisis in the Blain household averted this morning. Our Quooka boiling hot water tap went on the blink! The standby kettle is on the boat. I was all set to boil water in a pan for She-who-is-Mrs-Blain’s cup of tea… but awoken by the commotion, she came down, stepped in, went under the sink, flipped some hidden magical button off and on again. It worked. She said nothing, but gave me that look: you are an idiot, but I tolerate you because you worship me. It is not easy being married to a domestic goddess.
Back in the markets…
If you haven’t read The Hitchhikers Guide to The Galaxy, then I recommend you should. It may be 45 years old, but read it and be elevated by its wisdom on why stuff doesn’t work. Pay particular attention to the chapter on Deep Thought, the computer that gave us the answer 42 to “the meaning of life, the universe and everything”, and how it designed its own successor – the Earth – to figure out what the actual question was.
Has the bottom finally fallen out the stock market? When traders stop talking upside, but ask questions about gold, bitcoin adoption and who is going to buy Treasuries, then it’s clear why last week’s wobble has turned into a decline on the back of higher for longer rates, Fed comments, Israel and Iran, plus a rising sense of “what if we’re wrong about AI?”. Nvidia and the SOX (The Philadelphia Chip Index – which measures the price of the 30 largest US listed chip-makers) are in -10% correction territory.
Is the current dip a buy opportunity or a sell warning? I would remind readers its only two weeks till “Sell in May and run, run away..” Aside from the fact AI hopes and expectations have been driving the market, there are two aspects to the AI problem.
The first is supply and the current market incumbents:
I suspect the Funboy Tree of Nvidia, Meta and either Amazon or Microsoft (depends how I feel in the morning), will shortly become the Gruesome Twosome as Nvidia takes a reality punch to the kidneys…
The thing about AI is it’s a supply driven market. It’s all about how many ultra-expensive, AI suitable, parallel-processing Graphic Processing Units (“GPU”), Field Programmable Gate Arrays (“FPGA”), and accelerator Application-Specific Integrated Circuits (ASIC), chips are available to meet the burgeoning AI demand. (Someone is bound to ask why I missed NPUs, but I don’t know everything..)
When Meta, Alphabet, Amazon, Apple, Microsoft, or a dozen other firms hoping to associate and integrate AI into their stock price, brag about their AI skill-sets and unique growth multipliers on the back of AI, (or Jamie Dimon tells us JPM is fully integrating AI to make them even better at extracting fees from clients) they are entirely at the mercy of the AI supply chain gatekeepers.
Nvidia dominates the AI hardware and software design with control over 80% of the GPU market, while TSMC manufactures 90% of these AI chips – including the ultra-expensive H100/A100 processors required to run the AI models from datacentres. You can get a jet from Boeing faster than delivery of a couple of thousand AI chips – as clients of mine have discovered.
This morning TSMC produced its expected Q1 earnings beat, confirming AI chip demand remains strong, and that despite some manufacturing glitches its net income is up 8.9% from a year ago. It’s about to start producing 2nm chips which will be even more powerful and energy efficient. With 60% plus of global foundry revenues, TSMC is lightyears ahead of second largest chip firm, Samsung at 14%.
However, such dominance may not last forever. Nvidia faces competition. Some of the big AI chip users – including Microsoft, Alphabet and Amazon – are looking to design their own chips, while the UXL Foundation is developing an open-source AI Chip design to directly compete with Nvidia. (Meta is also going open-source.) Other chip firms are seeking to enter the AI market by designing new AI chipsets for mobiles and laptops.
TSMC commands chip supply – but will come under pressure as the strategic resource chip capacity represents becomes apparent to governments – witness the $52 bln in subsidies being paid out under the CHIPS and Science Act to attract chip production away from strategically vulnerable Taiwan to the US within a few years – although much of it has gone to TSMC! Samsung has big plans to build out its production, while Japan Advanced Semiconductor Manufacturing (JASM) will be bringing new capacity online this year.
The Chip Supply Chain will remain a highly competitive space – yet history shows Chip fabricators are not massive alpha generators because of the costs and complexity of their production and infrastructure process. (I love the story about the UK politician on a visit to Taiwan who asked how long it would take to build a Foundry in the North-East. She was shocked when her host said 20-years to train, educate and refine the human capital, and 2 years to build the fab.)
Chip making is a commodity business – a complex difficult one, but that gives the sector fundamental value in the modern economy.
The second issue are the hopes for AI that have driven anything AI associated to frankly unsustainable levels.
At present AI is basically the Large Language Models solving enormously complex jigsaws that collate, compare, and the backtest trillions of data points to come up with an answer that best approximates what is already accepted as the consensus answer – and do it all in seconds. It looks and feels like magic – but the reality is its just taking stored data from one chip’s memory and processing in another at speed. You can enhance the process with algorithms.
At present AI is somewhat ad-hoc and much of the growth-case is fuelled by over-imaginative marketers convinced by their own LLM tech writing the puff for them. I suspect all the noise following the launch of ChatGPT – which I have found to be inordinately useless for anything requiring more than a micro-gram of intellect – was premature.
Yet, we dumb chimpanzees are all terribly impressed and desperate for free bananas. It’s a variation on giving enough monkeys typewriters, then one of them is bound to write Hamlet. It is not intelligence, and whatever Elon Musk says about General Intelligence being around the corner – current artificial intelligence is not how the human mind works. That’s going to happen when memory and processing are combined in similar ways to the neurons in our brains – and even then we don’t really know where our consciousness resides… gosh, profound this early in the morning….
To think like human, and produce all the insights that go with that, will require a massive increase in computational capacity – and one question is whether current chip tech and the path we’ve gone down with Chip design will work. Current tech may all be terrible logical but may ultimately prove a limiting technology? Current chips may prove a mature technology, and remain tremendously useful just like railways remained even after the introduction of the autocar. (I did say in the intro the AI bubble was running out of steam!)
Market historians may like to recall just how destructive multiple railway financial crashes were in the 19th Century… I am wondering if the AI hype driven rally may already by played out, and the one to thing about is next generation quantum computing tech?
Out of time, and off to find a day job…
Bill Blain
Author of the Morning Porridge
Wind Shift Capital
www.windshift.capital
