There is no shortage of commentary in markets about whether today\u2019s AI boom looks like the dot-com era or not. Optimists argue that we are in a transformational moment. Skeptics warn this is classic \u201cthis time is different\u201d thinking. <\/p>\n
We are not dogmatic about either side. We do not think the best strategy is to blindly chase the hype, or to bet against it, and hope to nail the timing. Instead, we recommend focusing on the parts of the theme that can benefit from the current investment cycle with less sensitivity to the speculative long-term outcome of the AI race. <\/p>\n
Below, we respond to a few of the most common talking points we hear and outline how investors might think about both the bullish and bearish cases to build more rational strategies.<\/p>\n
Here are the popular arguments and statements we hear:<\/p>\n
(1) Bullish: The adoption speed of AI is unprecedented.<\/p>\n
We often hear that the uptake of new AI technologies is staggering and much faster than past foundational technologies, such as the internet or electricity. The claim is that growth is, and will remain, phenomenal. One version of this argument is: \u201cOpenAI already has around 800 million users worldwide.\u201d<\/p>\n
Response:<\/p>\n
This framing is a little too convenient. New technologies that are layered on top of existing infrastructure always scale faster than the infrastructure they sit on. Floppy disks spread faster than mainframe computers. Nokia became a global brand far faster than Alexander Graham Bell\u2019s telephone. That does not automatically mean the returns to early investors are guaranteed. There is no investment argument here. It is just a rudimentary observation.<\/p>\n
Fast adoption, by itself, does not tell you who will make the money. In previous technology waves, investing in the early developers did not necessarily pay off. Take the leading search engines or web browsers of the 1990s, for example. There is still enormous uncertainty around who will ultimately capture the cash flows, yet current valuations are starting to behave as if the winners are already locked in.<\/p>\n
There is also a bear interpretation of fast adoption. When a technological innovation spreads quickly, the next wave of innovation or open-source alternatives will likely spread quickly as well. That caps pricing power and compresses margins. Think of a pharmaceutical industry with no patent protection.<\/p>\n
Current large language model technology is already moving toward open source. The real bottleneck now is not the code; it is computing capacity, the decreasing learning rate from additional data, and the electricity supply. This is why the largest Tech players are racing to secure that capital-intensive layer first. They are spending heavily up front to gain first-mover advantage in infrastructure that requires massive sunk costs, which, almost by nature, creates an over-investment risk.<\/p>\n
For investors, the lesson is not necessarily \u201cbuy the headline incumbents.\u201d It is, \u201cfollow the bottlenecks.\u201d Focus on the supply chain and infrastructure that the entire ecosystem will depend on, rather than assuming that the current platform leaders will extract all the economics.<\/p>\n
(2) Bearish: \u201cSpaghetti charts\u201d of circular financing prove we are in a dangerous bubble.<\/p>\n
Another common claim is that these circular ownership and financing flows show excessive leverage through opaque structures. The comparison is to the late 1990s, when companies with little or no profit were receiving cheap funding.<\/p>\n
Response:<\/p>\n
This is not a clean analogy. What we are seeing today is mostly large bond deals, not speculative equity raises for profitless start-ups. Debt financing, in this context, actually signals a healthier risk profile than 1990s-style equity hype.<\/p>\n
These bonds are often backed by the cash-generating power of mega-cap Tech firms with very strong balance sheets. When a company like Nvidia supports financing for what is likely to become one of its largest future customers, it can look like it is plugging an extension cord back into itself. But it is underwriting risk for a massive client, and in effect, investing in its own growth as well.<\/p>\n
Yes, there is euphoria about AI growth. Yes, margins at companies like Nvidia could end up tighter than the market currently assumes as a result of these financing arrangements, but it is still not the same as the 1990s model of throwing equity at loss-making ventures with no path to cash flow.<\/p>\n
Where this does become a real market risk is further down the chain of financial markets, if ever. The danger is not simply \u201cCompany A lends to Company B, who buys from Company A.\u201d The danger is when those bonds get sliced, tranched, and re-levered in the broader financial system, creating synthetic exposures and circular liabilities. That is when you start to resemble pre-2008 style financial engineering. We are far from that scenario.<\/p>\n
On the other hand, one should take note of deals such as Meta\u2019s financing of its massive Louisiana data center, where Meta shifted debt out of its balance sheet to an SPV created by Morgan Stanley and funded by Blue Owl and Pimco, alongside Meta. <\/p>\n
In short, Tech-to-Tech financing alone is not yet a systemic risk. It starts looking systemic if everyone else piles in with leverage to get indirect exposure to those same cash flows. Investors should focus less on the optics of \u201ccircular customer-supplier relationships,\u201d which are not new (think Korean chaebol or Japan\u2019s keiretsu networks, where firms cross-hold and backstop each other), and more on whether structured products are building hidden leverage on top of those bonds.<\/p>\n
(3) Bullish: AI will deliver huge productivity gains and cost savings for businesses.<\/p>\n
The bullish view is straightforward: companies will automate more work, cut costs, and boost margins.<\/p>\n
Response:<\/p>\n
Maybe. It is plausible. But it is also not entirely new. Office-task automation has been progressing for years through code that stitches together multiple applications. A lot of this is increasingly done with open-source tools, which is why languages like Python have exploded in use.<\/p>\n
The real cost savings will come from firms automating their own internal processes using their own data and intellectual property. That leans toward custom deployment, not generic off-the-shelf large language models. Yes, AI will help teams write code faster or generate the next slide deck faster. But the real value will be in tailoring workflows and decision-making tools to proprietary data.<\/p>\n
That means the idea that current incumbents will extract most of the economic rents from every company\u2019s cost-cutting is optimistic, and likely unrealistic. We are already seeing models like DeepSeek that let firms build internal, closed-loop AI solutions. As that trend continues, businesses will try to internalize the gains rather than pay high rents to an external platform.<\/p>\n
(4) Bearish: The choke point will be power and hardware. Growth projections will fail because the grid cannot keep up.<\/p>\n
Response:<\/p>\n
This is a serious risk, and it could become one of the defining constraints on the AI buildout. The physical bottlenecks are already visible. Power demand from data centers is surging faster than incremental generation capacity. Yes, there is ongoing investment in energy infrastructure, but it is struggling to keep pace.<\/p>\n
For example, media reports note that Kinder Morgan, a major natural gas pipeline operator, has a multibillion-dollar backlog of projects with committed customers. At the same time, drilling activity in places like the Permian Basin is accelerating to supply more natural gas for electricity generation. Nuclear and large-scale hydro are possible alternatives, but they take a long time to build. Renewables are not only politically out of fashion but also incapable of creating the required energy intensity in a reliable way. In the meantime, natural gas becomes the swing fuel.<\/p>\n
This creates a new problem: AI buildout needs cheap, reliable electricity, but so do the producers trying to supply the energy. Both are trying to secure priority on the same grid capacity. Layer on geopolitical strains such as sanctions on Russian energy, plus growing U.S. energy exports, and it is not hard to imagine upward pressure on energy prices as resource constraints start to bind on that front.<\/p>\n
If energy prices rise and grid access becomes a queue at the same time, the economics of the current \u201cinfinite data center\u201d story start to bend. But if energy prices do not rise, the investment boom in places like the Permian Basin could be pushed away from its break-even assumptions.<\/p>\n
That tension between AI enthusiasm and energy-sector profitability is going to be an important theme for investors as early as next year, in our view. It will help gauge whether current AI growth expectations are sustainable, and where the next investable themes may emerge.<\/p>\n
We recommend being focused on the upstream suppliers of materials, equipment, and financing. Stock screens for higher margin, higher free cash flow names in that space will continue to perform as the investment boom will likely strengthen profit growth in the upstream supply chain.<\/p>\n