.png)
Blood Meridian: AI Ride Into The Dystopian Sunset
Barricade your windows and lock up your doors, because doomsday is coming. Dario and Altman, two individuals with a rivalry the likes of Bird and Magic, have found rare agreement on quite an alarming subject: that at the current pace of AI development, the upheaval will be disastrous. So, they’re calling for a slowdown in model development. Investors remain bullish on AI, but what now when the wraith in their risk models actually comes to life, and the AI race actually does slow down?
Illustration by The Geostrata
OpenAI raised $122 billion in March, but its financial model indicates that this money could be all dried up by 2028. OpenAI was planning an initial public offering for autumn 2026, but postponed, citing fear of the AI race and calling for a slowdown. Is this simply because they suffer losses so egregious?
THE CHIPMAKER DILEMMA
The stakeholders are convinced that eventually, AI will become valuable. The question is, will it become valuable before it becomes expensive? Data Centers, Electricity, Chips, Network Infrastructure– it upends the traditional logic that bigger infrastructure lowers unit costs. Frontier AI, on the other hand, demands great scale for greater capability, which in turn demands great capital, and further capability to justify said capital.
While Dario and Sam worry, Jensen Huang of NVIDIA says it’s rubbish. Is this because this slowdown affects chipmakers in a much more dire manner than it does hyperscalers? If the AI race slows down, the hyperscalers have enough on their plate to clear through, and don’t necessarily need more data centers to generate revenue.
SEMICONDUCTOR SELL-OFF
The Philadelphia semiconductor index in the past week has dropped 6%, largely attributed to OpenAI’s model training pause. In fact, many AI hardware stocks plummeted– SK Hynix (SKHY) fell 6.20%, SanDisk (SNDK) dropped 6.05%, and Micron Technology (MU) fell 4.22%. Qualcomm (QCOM) plunged 6.71%, Intel (INTC) dropped 6.18%, AMD (AMD) fell 5.37%, Marvell Technology (MRVL) dropped 4.81%, Corning (GLW) fell 4.50%, while META has had a massive September surge.
But this isn’t a death certificate on AI just yet. Way back in July, Semiconductor prices seemed to even then be getting kind of clobbered, while memory prices were rising. Hyperscalers were locking in long-term contracts, and analysts were still describing shortages. As the possibility of a slowdown is getting more and more real, semiconductors and hyperscalers are moving in virtually opposite directions.
Nvidia, Intel, AMD, and Marvell fell, while Meta, Alphabet and Microsoft rose, and it seems like finally, the market is catching up.
CASTLE WITH NO MOAT?
Despite all the whistleblowing, there has been no actual moratorium on slowing the AI race. Is this because there is no moat? No profitable unit economics? It’s not a technological moat, since models can be replicated and even made local. It can’t be data, since synthetic and public data are proliferating, as are better models.
Perhaps capital, but the only companies capable of generating that much are already in the race. Maybe the product itself is not the moat, so much as distribution, proprietary workflows, customer relationships, infrastructure, or the ability to embed AI into products people already use.
MUTUALLY ENSURED EXPENDITURE
So why not simply slow down? It’s the prisoner’s dilemma. Say there are companies A and B. Assume both A and B are in a race, both spending enormous amounts of money. If both slow down, both save money and perhaps allow the technology to develop at a more manageable pace. But if one slows while the other races, the company that slows risks handing its competitor the upper hand. And this, to these companies who are all neck-and-neck, is a matter of collapse.
At recent valuations, OpenAI is at $500Bn, and Anthropic at $380Bn. And maybe that’s where this look-the-other-way allegiance that Dario and Sam have started to develop with each other all stems from. Maybe the speed at which the AI race is going is, in fact, dangerous, but maybe this is a red herring to distract one another from the fact that there is no profit margin in sight; no, they cannot stop. The problem is not merely convincing companies that slowing down is desirable. It is convincing them that their competitors will slow down too.
COMMODIFIED INTELLIGENCE
That’s why this concept of a slowdown is so confusing, because it’s economically disastrous. There are financial ecosystems built around the belief that AI is going to be heavy. Semiconductor manufacturers, memory producers, data-centre operators, energy suppliers, networking companies and hyperscalers all bank on the fact that models will require more and more infrastructure as they develop. But what if that’s just not the case? What if the future is not another breakthrough in efficient intelligence but, in fact, making it simply more affordable and ubiquitous?
This brings us back to the market trends of Semiconductors and Hyperscalers. Markets seem to be wondering the same thing– who makes the infrastructure, and who actually monetizes it? The AI industry asks investors to believe enormous investment will one day result in enormous revenue. But if models get cheaper, more interchangeable, and more ubiquitous, the value becomes difficult for any one individual company to cling to.
CAPEX TO CASHFLOW
This isn’t a problem exclusively attributed to OpenAI and their genius financial model that requires the resources of a small nation. Big Tech is in hot water too. Alphabet, in July, despite reporting what would ordinarily be an absurdly successful quarter, became cash flow negative as AI capital expenditure went big. Its purchase commitments went over $800Bn, and capital expenditure is set to rise again in 2027. Wall Street, which has been cheering at every hundred dollars thrown in the general direction of a data center, has finally started to wonder, hey, when’s this coming back?
And now the uncomfortable question of revenue. It’s no mean number, or a pyramid scheme where everyone is buying someone else’s GPUS; with consumers and businesses driving it to $120Bn annually, what was virtually zero a few years back. Now we must wonder: what is the proportion to which revenue must reach to make the egregious amount of money being poured into AI infrastructure to finally become expenditure and not remain investment?
TAXPAYER OR VENTURE CAPITALIST?
Perhaps the most dire casualty of this slowdown is not the NASDAQ, but the American taxpayer. The AI boom does not sell chatbots so much as it sells investment, tax revenue, and eventual economic growth. There are no free lunches, and someone’s going to have to pay for the infrastructure.
The American government doesn’t collect tax on Nvidia’s market capitalization, or the potential of a model. It collects revenue. Companies spending billions on data centers are only economic stimuli if they generate productive value that justifies said expense. And if it doesn’t, America has very expensive graveyards for could-be’s, and very dirty drinking water.
Now the numbers seem to look even stranger, with just how much money is being funnelled into this industry. Tax incentives, energy infrastructure, semiconductor subsidies, data-centre construction, research spending and the public infrastructure required to keep the industry afloat all run on the assumption that AI will generate enormous returns. The question that emerges is what happens when the projected tax base never materialises at the scale required to compensate for the investment. The gap between AI’s paper valuation and performance is staggering, and the estimate gets more extreme.
By 2030, the industry is geared to spend $2 trillion on AI, with only a $1.2 trillion return. This results in an $800 billion gap, which economists call a perpetual J-Curve.
Maybe in the end it’s not about OpenAI or NVIDIA at all, but about the proletariat plight that seems to infiltrate everywhere and never fully resolve. It might just be about whether AI's productivity gains are large enough to generate the tax revenue needed to justify the enormous economic infrastructure being built around it.
BY DIYA
TEAM GEOSTRATA
Comments