![\"A](\"https:\/\/www.newsbeep.com\/il\/wp-content\/uploads\/2025\/10\/image_605147551.jpg\")
<\/p>\n<\/p>\n
Jonathan Kitchen\/DigitalVision via Getty Images<\/p>\n<\/p>\n
You have likely heard by now about the circular funding<\/a> of AI capex.<\/p>\n Advanced Micro Devices (AMD<\/a>) issuing warrants to OpenAI<\/a> in exchange for OpenAI buying AMD chips Nvidia (NVDA<\/a>) investing $100B in OpenAI as OpenAI deploys much of the funding into buying NVDA chips Oracle (ORCL<\/a>) buying NVDA chips to service a big contract with OpenAI NVDA potentially investing in XAI XAI investing in NVDA chips <\/p>\n Bloomberg sums it up quite nicely into a graphic.<\/p>\n Bloomberg<\/p>\n<\/p>\n Basically, it is the same capital sloshing around between these companies, creating what appears to be investment of unprecedented scale.<\/p>\n Bears point to this circle as being reminiscent of Cisco’s (CSCO<\/a>) vendor financing during the internet dot-com bubble of 1999, in which Cisco would finance the dot-com companies’ purchases of its equipment\/services.<\/p>\n We all know how disastrous that wound up.<\/p>\n AI bulls dismiss this analogy by pointing to the vast difference between the megacaps of the AI world compared to the startups of the dot-com era. Surely these massive companies are better able to make good on their end of the deals.<\/p>\n I am unconvinced that companies being larger offers any sort of protection. Beyond the Cisco analogy, there are plenty of other instances in which vendor financing has not worked out well.<\/p>\n Plug Power (PLUG<\/a>) provided a form of vendor financing to Amazon (AMZN<\/a>) to sell it hydrogen.<\/p>\n CNBC reports<\/a> on Plug’s deal with Amazon:<\/p>\n “Amazon said Plug will supply 10,950 tons of green hydrogen per year for its transportation and building operations starting in 2025. Amazon expects Plug to provide enough green hydrogen to power 30,000 forklifts or 800 long-haul trucks. As part of the deal, Plug said<\/a> it granted the e-commerce giant a warrant to buy up to 16 million shares at an exercise price of $22.9841 for the first 9 million shares. Amazon agreed to spend $2.1 billion on Plug products over the seven-year term of the contract in order for its warrants to fully vest.”<\/p>\n If PLUG’s share price had held up, the profit from these warrants would have paid for a substantial portion of AMZN’s hydrogen purchase. Today, however, it seems the warrants are worth close to nothing.<\/p>\n SA<\/p>\n<\/p>\n This was just a drop in the bucket for AMZN, but it is just an example of how vendor financing can fail even when the recipient is extremely solvent.<\/p>\n Some of the problem with the AI infrastructure boom is certainly that these deals are circular, but another portion of the problem is that the buyers of the hype product cannot afford what they are buying.<\/p>\n It felt a lot more sustainable when it was Microsoft and Google using their massive cash flows to finance their buildout of AI infrastructure.<\/p>\n It feels a lot less sustainable now that the billions upon billions of capex have expanded to companies that do not have enough cash flow to pay for it or, in some cases, don’t really have any cashflow.<\/p>\n Oracle has dipped into negative cashflow on its AI capex OpenAI has more capex than cashflow CoreWeave and XAI cannot fuel their capex with cashflow <\/p>\n To me this is reminiscent of the housing bubble circa 2005. Homes were being sold to people who absolutely could not afford them.<\/p>\n The home purchases were financed with what were called NINJA loans or (no income, no job, no assets) at very high loan-to-value.<\/p>\n Just as the banks were underwriting on the prospect that maybe those individuals would get a job, the vendor financing of AI infrastructure sales to these companies is predicated on the idea that they will someday become profitable.<\/p>\n I think that is a much more nebulous concept than it initially appears. The financial success of companies like NVDA or Constellation (CEG<\/a>) is already here. These companies have grown earnings by extraordinary percentages, which I think has given a false impression that AI is already profitable.<\/p>\n These are labeled as AI stocks, so the market might see their profitability as AI being profitable, but they are not actually a representation of AI making money. If we examine the industry structure, I think it’s clear that most of the profits are in the suppliers of AI infrastructure rather than the AI itself. Having strong suppliers such as NVDA is actually a bad thing for AI profitability.<\/p>\n For complex situations like this, it can be helpful to turn to classic business strategy models like Porter’s 5 forces.<\/p>\n 2MC<\/p>\n<\/p>\n Each of these factors impacts the long-run margins of the industry. Let’s briefly discuss their implications for the profitability of AI.<\/p>\n So far, AI itself has made very little money, at least compared to the amount of capital invested in it. Things like customer subscriptions to ChatGPT represent true AI revenues.<\/p>\n The lion’s share of the money made so far is by the suppliers<\/p>\n Chips Data centers Electricity providers <\/p>\n In each category there are relatively few suppliers.<\/p>\n Chips are mostly NVDA and Broadcom (AVGO<\/a>) with a little bit of AMD and perhaps some others on the fringe that are trying to break in.<\/p>\n Data centers are either provided by the major data center companies like Equinix (EQIX<\/a>) and Digital Realty (DLR<\/a>), or the hyperscalers are vertically integrating by building their own.<\/p>\n Electricity suppliers are increasingly sparse the more near-term you want the electricity. In a 15-year window, one could build their own plants or contract to have them built. However, in the next 5 years there will be a clear shortage of power. One can attempt to buy the tiny bit of spare capacity from the grid or use power purchase agreements (PPAs) with concierge energy providers. On the time scale AI companies are looking to power up, they are really stuck dealing with the location specific utility and companies like Constellation or Clearway Energy (CWEN<\/a>), which are increasingly enjoying being able to charge remarkably high prices for power.<\/p>\n Getting back to Porter’s 5 forces, high negotiating power of suppliers is a bad thing for industry margins.<\/p>\n Substitution is an unclear force at this time because it is not yet known exactly what functions the AI will perform. The capacity for substitution will likely vary by function, but I don’t see it being a particularly threatening force to the industry.<\/p>\n New supply, however, is a major threat to AI margins. The more different platforms invest in building out their AI, the more competitive rivalry there will be in the industry. Each time there is a new announcement of a company investing 10s of billions into AI, the AI stocks seem to rise, which is baffling to me. More investment from competitors is unequivocally a bad thing for industry margins.<\/p>\n Finally, regarding buyer power, it is somewhat hard to read because sales of the AI itself are still in a fledgling state. Ultimately, buyer power will likely depend on how fungible the various AI functions are. If 5 different AI providers offer the same functionality, it will be quite easy for customers and potential customers to go with the lowest-price provider.<\/p>\n The holy grail that the AI companies seem to be seeking is AI fulfillment of a key job function that somehow cannot be replicated by peers. Perhaps there is room to do this through network externalities, software ecosystem lock-in, or other similar gating. For example, Google Sheets can do nearly everything that Excel can do, but because so many systems are integrated with Excel and so many workers were trained in Excel, it retains a strong customer base.<\/p>\n![\"bubble](\"https:\/\/www.newsbeep.com\/il\/wp-content\/uploads\/2025\/10\/1843741-17601230201037135.png\")
<\/p>\n![\"A](\"https:\/\/www.newsbeep.com\/il\/wp-content\/uploads\/2025\/10\/1843741-17601230201609244_origin.png\")
<\/a><\/p>\n![\"A](\"https:\/\/www.newsbeep.com\/il\/wp-content\/uploads\/2025\/10\/1843741-17601230200679278_origin.png\")
<\/a><\/p>\n
![\"A](\"https:\/\/www.newsbeep.com\/il\/wp-content\/uploads\/2025\/10\/1843741-17601230201476376_origin.png\")
<\/a><\/p>\n