Artificial intelligence is no longer just a software story. It is becoming a capital-spending story big enough to rival the great buildouts of the modern era, with one new forecast saying the money going into AI could end up costing more than the railways built in both the U.S. and the U.K., and then some when the internet is added on top. The new warning lands as Wall Street is still trying to price a surge in spending from the biggest technology companies, even as investors start to ask how much infrastructure the AI boom can absorb before returns get harder to defend.
The latest read-through comes from a MarketWatch report that says an accounting giant expects the AI buildout to outstrip the cost of railways in both countries, while also surpassing the internet’s infrastructure bill. The pitch is simple: this wave is different because of its scale and its spending growth. That framing matters because it shifts the AI debate away from chatbots, chips and product releases, and toward something more old-fashioned and more expensive — roads, wires, servers, power and the physical system needed to keep the machines running.
The comparison to railroads is not accidental. It is one of the classic tests for whether a technology craze is becoming an economy-wide investment cycle. Goldman Sachs Global Institute has already argued that AI capex is on a path large enough to stand out against several famous U.S. investment eras. Goldman estimates AI capital expenditure will reach about 2.4% of U.S. GDP in 2026. That would put it above the peaks seen in the highway buildout, the space race, the 1990s telecom boom and the shale revolution. Goldman says that level is also similar to 1930 electrification, when spending reached 2.6% of GDP.
That comparison alone is striking. But Goldman’s longer view is even more aggressive. In its “Tracking Trillions” report, dated May 1, 2026, the firm projects AI capex could climb to 4.4% of GDP or higher by 2030. Goldman says the only historical U.S. period that exceeded that projection was the 1840s U.K. “Railway Mania,” when railway capital deployment hit 7.3% of GDP in a single year, according to Andrew Odlyzko at the University of Minnesota. That is the kind of benchmark that turns a sector story into a macro one.
The practical meaning is that AI infrastructure spending is no longer being treated as a side expense tucked inside earnings calls. It is becoming a central investment plan for the companies building the platforms. That creates a feedback loop: the more promising the AI race appears, the more money gets poured into data centers, networking, chips and power. But the more money that goes in, the more investors need proof that the return will eventually justify it.
The scale of the spending is not theoretical. The Wall Street Journal reported that four U.S. tech giants — Microsoft, Meta, Amazon and Alphabet — are planning up to $670 billion in AI infrastructure spending in 2026 alone. That is a staggering number even before one tries to separate what is truly incremental AI investment from broader cloud and data-center spending. It also helps explain why the market keeps treating every clue about capex as a real earnings event, not just a balance-sheet footnote.
Still, the spending picture is not perfectly clean. The evidence available shows conflicting 2026 estimates for the same four companies, including a separate commentary that puts the figure at $725 billion, up 77% from $410 billion. Because those numbers cannot be reconciled from the accessible material, the safer takeaway is not the exact total. The safer takeaway is that the range itself is enormous, and large enough to dominate the economics of the largest U.S. tech firms.
That matters for investors because the market is no longer just rewarding AI narrative. It is also discounting the cost of building the narrative into reality. If Microsoft, Meta, Amazon and Alphabet keep pushing harder on infrastructure, then AI becomes a real capital cycle with all the usual consequences: heavier depreciation, more power demand, more dependence on supply chains and more pressure to show revenue that matches the spend.
The broader market reaction to this exact story was not clearly verified in the accessible sources, so there is no clean tape-reading to hang on the headline. But the underlying tension is obvious. Investors have loved the idea that AI can drive a new era of productivity and platform growth. What they have not yet fully solved is how to value a race that demands massive upfront spending before the payoff is known.
That is why the railroad comparison resonates. Railways changed the economy, but they also destroyed capital along the way when too much track was laid too fast. The same logic applies here, even if the tools are newer. AI infrastructure can be transformative and still be a poor short-term trade if spending outruns monetization. The companies doing the building may have the cash flow to fund the race, but shareholders still have to live with the dilution of returns while the system scales.
There is also a macro angle. Goldman’s 2.4% of GDP estimate for 2026 and 4.4% projection for 2030 suggest AI is becoming a visible share of national investment, not just corporate strategy. That means the boom is starting to matter beyond the tech sector. It touches power generation, construction, semiconductors, networking and industrial capacity. In other words, the AI buildout is becoming a piece of the real economy.
The next concrete milestone is Goldman Sachs Global Institute’s May 1, 2026 “Tracking Trillions” report, which gives the clearest forward projection currently available in the evidence. That report points to the possibility that AI capex keeps climbing into 2030 rather than peaking early. If that path holds, the market will have to decide whether today’s spending is the foundation of a durable productivity cycle or the expensive first leg of an infrastructure binge.
For now, the message from the numbers is blunt. The AI race is already big enough to compare with electrification, the highway buildout and the telecom boom, and Goldman says it could move even higher. The MarketWatch story adds the most dramatic comparison of all: railways in the U.S. and the U.K., plus the internet, and still not enough to capture the full scale of the spending. That is not a normal software cycle. It is a capital era, and investors are only beginning to price the cost of entry.
