Artificial intelligence is driving record investment, with Microsoft, Amazon, Google, and Meta alone committing around $350 billion this year. Unlike during the internet bubble, capacity is not being built ahead of demand but in response to it — shortages of chips, power and facilities are the binding constraints. The boom will continue because the largest technology companies have no choice but to commit; if they do not, they risk becoming tomorrow’s lunch. And because AI is best understood as an evolutionary rather than a revolutionary step, the internet boom and its large-scale disruption are not the right blueprint for what lies ahead.
Nine of the ten largest companies in the S&P 500 are technology firms. In order of market value, these are Nvidia, Microsoft, Apple, Alphabet, Amazon, Meta, Broadcom, Tesla, and Oracle. Together, they account for around 40% of the index.
These firms are now deeply engaged in the artificial intelligence build-out that is propelling equity markets. The scale of investment to enable AI adoption is historic, although not unprecedented. This year, Microsoft, Amazon, Google, and Meta alone are committing close to $350 billion to data center construction worldwide.
Railroad, electricity and the internet
At present, US data center investment is running at roughly $200 billion a year, equivalent to about 0.7% of GDP. While sizeable, it remains below earlier episodes of transformative infrastructure spending.
During the great railroad expansion of the 19th century, annual investment reached between 2% and 3% of US GDP. The spread of electricity in the 1920s and 1930s required a similar level of utility expenditure. Even the internet and telecom boom at the turn of the millennium saw capital spending peak at about 1.2% of GDP.
By comparison, today’s AI-driven data center surge is smaller in relative terms but accelerating. Given the pace of adoption by both consumers and enterprises, it is reasonable to expect that investment will continue to rise.
Demand is outrunning supply
There have been worries about overinvestment in data centers, as happened during the internet boom of the late 1990s. At the height of the dot-com boom, vast sums were spent laying optical cable in anticipation of future traffic. Some estimates suggest that by the early 2000s, as much as 85% of fiber capacity in the US was unused.
The key difference is that this time firms are not building to meet future anticipated demand; rather, they cannot keep up with present demand. Amazon, Microsoft, and Google all report that demand for compute is running ahead of what they can supply, held back by bottlenecks in chips, power and physical facilities.
Surging demand for AI is epitomized by OpenAI, which reports having over 700 million weekly users — the fastest growth ever recorded by an application. OpenAI CFO Sarah Friar quipped at Goldman’s annual technology conference that every Monday, on its regular call with Microsoft, it reiterates the same request: more compute.
Some investors worry that present pricing is so far below the cost of compute that there may not be profitable business models for companies such as OpenAI. The company charges only $20 per month for ChatGPT, well below current compute costs. Like many growth firms, it is prepared to burn cash in pursuit of scale, on the assumption that profits can be pursued once the growth runway shortens.
AI costs will come down. Moore’s law — the doubling of transistors roughly every two years — has historically halved the cost of computing power, though it is an empirical trend rather than a law of physics.
Prisoner’s dilemma
The data center building contest among Microsoft, Amazon, Google, and Meta is driven not only by commercial motives but also by strategic considerations. In theory, firms would be better off moderating AI spending, but in practice, the fear of being outspent — and therefore outpaced — leaves none willing to pull back.
The result is overinvestment from a short-term perspective, but one that is rational in strategic terms. No technology leader wants to be remembered as the leader who missed the AI moment. This competitive trap may be the single biggest reason why the AI boom will continue.
Who captures value in technological transition?
Technological shifts create winners and losers. During the internet boom, companies selling the “picks and shovels” such as Cisco, were not the ultimate winners. Infrastructure providers like the telecom firms fared even worse. The real value accrued to the application layer — Microsoft for enterprise, Google and Amazon on the consumer front.
Many of the winners were not even present during the advent of the internet. The iPhone enabled new applications such as Facebook and Uber. Cloud computing allowed more disruption on the enterprise front with winners such as Salesforce, Adobe, and Microsoft.
AI is statistical learning
The AI transition will not follow the 1990s internet blueprint exactly. At its core, AI is statistical modeling applied to very large datasets. Roughly put, its recent progress stems from two forces: the use of graphics processors, which are highly efficient at matrix multiplication, and the vast datasets accumulated over three decades of internet use. Together, these have allowed models to become very sophisticated, as seen with ChatGPT.
Many practical problems can be characterized by statistical modeling. Which video is a user most likely to watch next? How should a driverless car react in a given situation? What is the likely trajectory of a drone, given its launch parameters? AI is well suited to such questions.
In the near term, the technology will be applied to the easiest and most profitable problems. That means more targeted advertising, more compelling social media feeds, and more efficient digital services. Over time, models will extend to more complex challenges, from robotics to autonomous systems.
AI is an evolutionary rather than a revolutionary step
The birth of the internet was a transformational event and destroyed entire industries — newspapers, video rental chains, and more — replacing them with technology firms. AI is evolutionary rather than revolutionary. Strong businesses are more likely to adapt than to be swept aside.
Uber may end up benefiting from robotaxis; SaaS companies such as Salesforce may leverage agentic AI rather than be replaced by easier AI coding; Google has already modified search to be AI-assisted, and Meta’s advertising has been strengthened by better targeting.
The incumbents of the internet era were video rental shops; the incumbents of the AI era are global technology giants that are themselves driving change. Google’s Gemini models top the Apple Store downloads, Microsoft holds a stake in OpenAI, and Amazon develops custom hardware through Annapurna Labs in partnership with Anthropic.
In the long run, only the paranoid survive
This does not mean today’s leaders are invulnerable. Intel once looked unassailable in the 1990s, capturing the bulk of PC industry profits alongside Microsoft, yet later faltered to Taiwan Semiconductor Manufacturing Company (TSMC). Strategic missteps and technological shifts can still reorder the hierarchy.
In the short run, AI will deliver incremental change. In the longer run, it may enable more imaginative advances in robotics, autonomy, and quantum computing. As the cliché goes, people tend to overestimate change in the short term and underestimate it in the long term.
Perhaps this simply reflects our difficulty in grasping the power of compounding: small incremental changes accumulate into wonders. As Albert Einstein is said to have quipped, “Compound interest is the eighth wonder of the world. He who understands it, earns it … he who doesn’t, pays it.”
