Saw this graph earlier used to suggest there isn't an AI bubble - that increasing margins suggest we've some way to go before being overvalued. My counterpoint would be that it charts the export of low-margin industries like manufacturing from the US and the rise of high-margin software & service companies to their current place of complete domination of the S&P. And that doesn't mean we're not in an AI bubble - the amount of money being spent in capex on GPUs and data-centres won't be covered in this cycle or the next few GPU cycles. META et al are saying their bought GPUs will last 5-6 years, but that's more for accounting than any real statement on how long the hardware will actually stay relevant. The key is how their data-centres are built now - you really only have to build a building and parking lot once, and if build everything inside it so that everything from hardware to power supply to cooling can be upgraded easily, then your next capex cycle could be significantly cheaper. The current generation of GPUs and racks are power-hungry and will simultaneously become much more energy efficient and computationally powerful AND the software will become much more efficient, and that will allow a much longer lifespan of the data-centres themselves with minimal upgrades. BUT it still won't even be close enough for Nvidia's customer's to recoup their spending. Yes, what we're going to be asking of AI going forward is massive. The jump from basic inference to reasoning AI required 100x in terms of hardware and energy usage simply because there is diminishing returns in the quality of response from AI the longer it is required to think, or the more powerful the hardware used (or both). But people (and companies) aren't going to want to spend 100x when they use 100x the resources. There is an obvious conflict in the AI hardware industry which on one hand is saying "we're going to make so much money" whilst also saying "this is going to save ordinary companies so much money on staffing costs", both attacking the same pie, and ultimately competition will mean shrinking margins for the AI hardware companies. The similarities with the dotcom bubble crash should be obvious. Think of the Internet, an invention that would within 15 years of the dotcom bubble crash become utterly dominant - you couldn't think of doing business without the internet in 2015, nevermind now, but in the year 2000 the froth led to massive capex overspending and no way to recoup that spending for well over a decade. The timescales between the dotcom bubble burst and recovery and the coming AI bubble burst to recovery may even be about the same. What is interesting is that even it you backed dotcom bubble companies that survived and became the dominant force in the tech industry, like Microsoft and Apple, the drawdowns were massive. In current denominators, Microsoft fell $60 to $19 (-68%), Apple $1.34 to 24c (-82%), Intel $76 to $13 (-83%), Amazon $5.64 to 28c (-95%). Everyone before the dotcom bubble burst thought this kind of drop was impossible. Your options then were selling at a loss, bag-holding for ~14 years, or averaging down. Cisco was central to it all. They had concentrated sectoral risk - they were funded mostly by huge capex from a single sector of industry (telecoms). When revenue growth didn't match capex and didn't justify further expansion the telcos slashed capex at once and the demand shock was near-total. Cisco (like Nvidia) still had some Enterprise and Governmental customers and that is why it survived, but the stock collapsed 86%. People were watching Cisco earnings, but it was the Telcos earnings that they should have been eyeing like a hawk - that was the signal. Nvidia is much the same - everyone keeps watching Nvidia earnings and their suppliers to see if the bubble is continuing, but it's their customers (MS, Google, Amazon, Meta, Tesla et al), who you should be watching - and not their general revenues or even their capex spending, but their conversion of AI capex to into hard revenue. But that's not the only reason Nvidia's customers might cut their AI Capex spending - watch too for a US recession. If the US enters recession, and I think it is not far away looking at job numbers and housing, then spending on AI hardware will slow rapidly and dramatically - if it happens, that demand shock would utterly floor Nvidia as it did Cisco. My advice if you're already fully invested in tech is to look at your R:R - how big do you think Nvidia et al can get from here before the next recession ? Risking all your profit plus all your book for the least likely gain (compared to the risk of your previous gains) is poor R:R - picking up pennies on a railroad. But if you don't want to sell a stock that has been good to you - timing the top is impossible of course - at least take some sort of defensive action even if it slices some of your future profit. Maybe hedge your shares with puts, maybe switch to deep ITM calls to reduce your capital invested in AI by ~70%, or make sure you're trimming so you have ammunition to fire if the AI bubble does burst. If the AI bubble *doesn't* pop, you still catch the updraft with most of your cash that you think the most expensive companies in the world still have ahead of them, but you've diversified your portfolio for safety; conversely if the AI bubble *does* pop you're not left nursing a 70%+ drawdown and you've got plenty of ammunition throw afterwards at growing industries or average down once you've decided which companies are going to survive.