Why Investing In AI Is A Great Idea & Why Investing In AI Is A Terrible Idea
The dawn of mainstream AI is among us because we are already seeing subjects like ChatGPT become dinner table talk as if it is some kind of new gizmo or gadget that everybody simply must try. Over the next few months, we are going to see the innovations of artificial intelligence carve a chasm of innovation through the business landscape so deep that there won’t be a single sector it doesn’t touch. We will see the exponential curve of improvements where the tech seems to monumentally leap over its previous generations faster than we have seen any technology before it. With these new technologies come new opportunities and new business models that also bring with them opportunities to invest.
If you haven’t already now is the time to invest in AI but not in the normal monetary investment sense. Sure you could have a look to see who is selling shares in some new company that is promising to be a game changer in AI or you could invest your money in a more established company that hosts the servers the technology runs on but these are not where I’m putting my investment.
Is AI the next dot-com bubble?
If you turn the clock back 25 years everything was moving online and the promise of “everything dot-com” was happening. There was a lot of money moving around but the sad reality of the dot com era was it was a hyped investment bubble and the majority of the thousands of companies didn’t turn out that well. In 1996 there were 739 IPOs of US firms and in hindsight, the odds of picking a winner sure were against you. Of all of the companies that had an IPO between 1995–2000 the only two that are noteworthy are eBay & Amazon. At the time Amazon was mostly promoted as selling books online and the larger vision of logistics was never publicised & “cloud computing” was a term that was only coined in 2006.
So if we turn the clock forward to today and start our hunt to find the next Amazon, where do we look? Well the problem is AI is an ever-changing beast, the king of the castle is the one that can stand on the top but for the time being, we are going to see new and better kings be crowned every month. It simply doesn’t make sense to invest in mid to late-stage companies that have a “unique solution” because the likelihood of a competitor releasing a superior product is only a matter of time. If you’re lucky a competitor will acquire you for some intellectual property that you have or to access the talent you have in-house but the more likely scenario is that they never saw you as competition and they will just leave them behind in their dust.
The trend right now with generative text is to go to the OpenAI website directly and use Chat GPT, the moment there is an easier solution people will shift to using that, and soon we will see it built into most of our other applications and websites. We are already seeing a shift to using things like Notion or Bing or even websites like GPTBoss.com. Picking a winner this early is a needle in a haystack.
What about selling shovels?
The saying “sell shovels during a gold rush” came from the California gold rush of the 1850s and you could argue that it is a sound option and it still works in some sectors, but I’m not investing in hardware. A couple of years ago during the crypto rush, NVIDIA (the graphics card maker) went from $50 in Nov 2019 to over $300 in Nov 2021 because everybody wanted graphics cards to mine for crypto. AI needs a lot of computing and the more mainstream it becomes logic dictates that we are going to need more hardware. Investing in TSM, Intel, Qualcomm, Broadcom, Micron, NVIDIA & AMD seems logical if there is a massive increase in specialist hardware sales but which one do you pick and will it trump +50m graphics cards sold mostly for crypto?
The hardware vendors are going to create a lot of new innovative hardware to optimise for many different AI & ML use cases but I don’t see the value of most of these locked up in a server room somewhere on the cloud. Making the fastest CPU, GPU or TPU is simply a race to the bottom, the moment your competitor comes out with a new product you need to adjust your price to remain competitive. Tensor Processing Units take a long time to architect and build and are only good for a single application because they are ASIC (Application-Specific Integrated Circuits).
There will be value at the edge where we use the hardware inside our smart cars or process things like medical imaging in real-time but cloud service providers constantly upgrading tech because some new hardware is released is a race to the bottom. Sure these hardware vendors could come out with the fastest bit of hardware and the next month another vendor undercuts them, but what is more, likely is the demand for the raw CPU or GPU power to drop significantly. It is projected that it costs Microsoft $30m per month to run ChatGPT, which is a drop in the bucket for a company that has a net profit of $72b, but the applications are going to get smaller.
Right now the data sets these models are trained on are massive, ChatGPT has 175 billion parameters taking up 45TB of text data, but over time the system requirements could come down significantly. We have already seen different training models output similar to ChatGPT with 1/10th the system resources meaning it may be possible very soon to run these applications on existing hardware. Developing more effective systems is the future of AI, if you can create an algorithm that runs 10x faster or can use 10x fewer resources with the same output you instantly become the new benchmark.
So where to invest?
The best thing to invest is your time, invest your time into understanding the landscape and looking at how AI can be implemented. A lot of companies will be hyping what they are trying to build but the truth of the matter is that AI can now be implemented into businesses within a very short period of time. In fact, you can now use ChatGPT to create a program that allows you to use ChatGPT inside that program. It’s not quite “the robot building the robot” but more like telling the robot to do something and it understands you.
The recently released API (Application Programming Interface) from OpenAI now allows people to rapidly implement ChatGPT into other programs. This means it is easier than ever to create custom software using AI tailored for individual use cases or to build it into existing custom software. The smart money is on the companies that are going to be doing the AI integrations with small to medium-sized companies, demand is going to skyrocket and it will be a while before the big players integrate AI into their slow tech solutions.
But if you have money burning a hole in your pocket, where do you put it? You will have better odds investing in people; finding the clever ones around you, the problem solvers, and the ones that think differently. Most people who have money to invest already understand business so partnering up with an entrepreneur and being an angel investor makes a lot of sense. There are a lot of businesses out there that can benefit from integrating AI into their workflow, you never know, you could help spawn the next Datacom or Seequent.
