As an engineer that worked at a leading cloud provider and one of the people responsible for designing a service that pushed through millions of queries that processed billions of tokens per day, I wanted to share a behind-the-curtain view on the technical and economic impact of the direction the AI industry - and technology as a whole - is heading.

History #

Starting with history, let’s talk about the dot-com bubble. At the time, it was a massive surge in investment into Internet-first companies - generating hundreds of billions of dollars in cash flow to these ideas with little to no proven value in the market. When the crash started, and inbound venture capital was no longer propping up unprofitable corporations, those that were unable to fund their own activities failed in quick succession.

In essence, the survival of companies was determined entirely by demonstrated ability to fill a market niche.

Venture Capital #

For those unfamiliar, when a startup’s life begins, it often requires external funding to get off the ground. Employees need to be paid, servers need to be purchased, and the general costs of running a business need to be accounted for. It can cost hundreds of thousands of dollars in salary alone to build out a “minimum viable product” over the course of a few months.

Minimum Viable Product #

“What is an MVP?” you may ask. An MVP is the first step in proving your startup has value. It is often defined as the absolute minimum set of features and functionality required to create a product that a customer would pay for.

It is often spoken of with religious fervor - your company lives and dies with your MVP. If investors don’t like your MVP, they aren’t going to give you more money to expand and scale. Initially, everything hinges on this concept and its execution.

After MVP #

Once your MVP has been built, it is natural to look to expand your startup by hiring more employees to handle additional tasks within the company, or to purchase additional assets necessary to make the company run. Computers and servers, for example, if you’re an Internet company. Cars if you’re a taxi service. Inventory for your retail stores. All of these things require more money.

While a company can grow by bringing on customers slowly with the MVP, this is often far slower than may be desired - either because the founders of the startup want to get rich quicker, or because there is risk that a competitor may steal business from you by simply spending more money to develop a competing product.

Rapid expansion requires funds. Therefore, most startups look to bring in additional capital by selling portions of the company’s shares to investors. This brings in the funding necessary to continue the company’s operation.

The Dot-Com Bubble #

When this rapid expansion can no longer be sustained due to a lack of runway (a term for how much money the company has, expressed in terms of months of operation it can sustain with its current rate of expenditure), the company has to cut costs. Often, it reaches the point where the company can no longer afford to operate, and must shut down.

During the dot-com bubble, companies that could not afford to operate met this fate. However, this did not necessarily mean that the idea was inherently a failure - simply that the execution was not able to be completed with the landscape of the time.

This last point is key: The company did not succeed at the time. The Internet didn’t simply disappear when the bubble popped. The companies that were not able to sustain themselves either closed their doors, were bought, or aggressively cut costs.

We have businesses today operating on the Internet that could never have been imagined in the late ’90s to early ’00s. For example, it would have been difficult to predict that everyone would have a portable communication device in their pocket with the ability to stream movies while on a plane, or take photos and videos of quality rivaling cinema cameras - then editing and sharing it all on that little device. In essence, the Internet never disappeared because of the dot-com bubble.

Current AI Investment #

AI companies today spend up to 80% of their capital on infrastructure. This means that if an AI startup raises ten million dollars, up to eight million will be spent on the chips that power their product. The money raised in early-stage companies is being spent in ways and amounts previously unheard of.

Investment is being performed by institutions large and small - from angel investors and incubators (organizations that will invest a comparitively small amount and provide support and mentorship to early-stage startups) putting $100,000 into ideas that aren’t yet fully formed, all the way to the hundreds of billions of dollars transacted by major players in the industry to guarantee their security in the market, such as Nvidia, Microsoft, Blackrock, and others.

But wait - major players securing the position in their market? That’s right. This is the kind of circular funding that nearly guarantees the potentiality of - at minimum - a future market correction.

Circular Investments #

Many of us have heard about the $100,000,000,000 deal by Nvidia backing OpenAI. OpenAI will be spending that money… on buying Nvidia’s chips. This is entirely circular, and has generated net-zero cash movement, but propped up the stock prices of both companies.

This trend is worrying. OpenAI isn’t the only example - Nvidia has also purchased what essentially amounts to all of CoreWeave’s spare, otherwise unsellable compute, guaranteeing profit for the company which purchases massive amounts of hardware from the chip-making giant.

Nvidia isn’t the only place where this occurring. Previous to the start of COVID, investments in seed-stage (very early stage, before that magical MVP) companies rarely were measured in millions of dollars. Now, we have companies regularly raising $10,000,000 seed rounds, simply because it is some new niche to which one can apply AI.

This money is then funneled up the chain and into major players and investment banks, which then is redistributed to other startups in the form of investments, which then goes to infrastructure providers, then again to new startups as investments, and the cycle continues. Much like how in 2008, when mortgages to default, it caused the collapse of a delicately-built house of cards based on circular investments, the AI industry is primed to collapse hard.

AI’s Role in the Modern World #

AI, even with all its faults, is here to stay. Much like the Internet in the 1990s and 2000s, a massive collapse did not kill the technology. AI fundamentally enables processes and capabilities at scales that would require massive numbers of humans otherwise, for incredibly small amounts of money. Modern Large Language Models (LLMs) can can produce hundreds of thousands of words in just a few minutes, at a cost of just $10 (as per OpenAI’s GPT-5 pricing). A skilled human with expertise in a given subject could charge thousands of dollars to produce the same output.

An additional benefit is that AI can be deployed and scaled horizontally (meaning more AI models running next to each other, otherwise known as in parallel) with very very little cost. To run more AI models simply requires hardware which can be easily rented from compute providers and made accessible in seconds. To hire an additional human to complete a task could take weeks. This massive “peakiness” allows rapid growth and decline in costs and revenue, tailored to what the market desires at a given time - or even what demand dictates across a given day.

There is no way to match the capabilities that AI provides with humans. It’s not about the best human in a field vs an AI model, it’s about the model being “good enough.” If you were to pay an average person at minimum wage to produce work about something they were previously unfamiliar with (say, writing, programming, data analysis, stock trading, painting, teaching, or even making critical business decisions), that would still have a cost of $7.25 per person per hour. If you were to provide $7.25 to OpenAI, that would produce you around 500,000 words of output on any subject you would like. In the same timeframe, a person could produce up to 6,000 words, if they were to write at the maximum reasonable typing speed of an average person, at 100 words per minute.

It is not that the AI models are amazing at their tasks and will replace skilled specialists. It’s that these models are good enough to perform a broad range of simple tasks, and that’s just based on the current state of the technology.

Even though a lot of uses of AI are less than perfect, and some are actively imperfect, there will be a large societal shift in what is possible to achieve in a given timeframe.

Why none of this matters #

Let’s jump back to the dot-com bubble. Now that it’s in the past by about 25 years, did the massive crash change the level of permeation that the Internet has into our lives? No. Not even slightly. We as a society are entirely reliant on the ability for communication to occur at long distances, and that is the fundamental proposition of the Internet as a whole. AI’s fundamental proposition is super-human speed to execute a given human-sized task.

In twenty-five years, if the market crashed tomorrow, AI will still be in use. The major players may change, there will be small, independent shifts in what we utilize and don’t utilize AI for, but that doesn’t mean it will disappear. The companies will change, the people will change, but fundamentally, the alteration of society that AI has already inflicted will not go away.

Much like the way railroads and industrialization took over the world in the 1800s, AI will not go away without a fight.