The argument in the US for letting Big Tech run faster is seductively simple. Artificial intelligence is the next great general-purpose technology, China is the principal competitor, and excessive regulation could cost America the race. The Trump administration has accepted much of this argument. The difficulty is that an AI race conducted on these terms may settle another contest first: who owns the infrastructure on which the digital economy will run.
Combined capital expenditure by Amazon, Alphabet, Meta and Microsoft was about $170 billion in the second quarter of 2026, 72 per cent higher than the year ago period. Industry numbers compiled by FactSet put the combined spending of Big Tech on data centres and chips during 2026 and 2027 at roughly $1.5 trillion. These numbers reveal something larger than an investment boom. Few potential competitors can buy admission to a business where hundreds of billions of dollars have become the entry fee.
READ | Regulation of Big Tech, AI will test India’s political nerve
Alphabet spent $44.9 billion on capital expenditure in the June quarter, overwhelmingly on AI infrastructure, and reported negative free cash flow of $5.9 billion. Meta expects capital spending of $130-145 billion in 2026. Microsoft added 31 data centres across five continents in its latest quarter and says it is on course to double its overall capacity in two years. Amazon expects capital expenditure of around $220 billion this year.
The scale is important because artificial intelligence is reversing the economics that created the first generation of tech monopolies. Software businesses could once grow with relatively little physical capital. Frontier AI requires chips, data centres, land, electricity, networks and huge quantities of data. An incumbent with advertising, cloud or e-commerce profits can finance those assets from existing cash flows. A start-up cannot, except by renting computing capacity from one of the incumbents or raising capital from investors associated with them.
Big Tech is buying the commanding heights of AI
That changes the character of competition. Microsoft, Amazon and Google are not merely developing their own artificial intelligence. They own much of the cloud infrastructure on which other developers build and run models. The giants can compete at several levels at once: computing, models, applications and distribution.
The danger, therefore, is not necessarily that one company will own artificial intelligence. An oligopoly can exercise formidable power without producing a single monopolist. Control over computing capacity can determine which models get built; control over operating systems, search engines and app stores can determine which ones reach consumers. Existing businesses provide both the money to finance expansion and the distribution channels needed to recover the investment.
READ | Data protection law risks fortifying Big Tech
There is already ample evidence of what entrenched market power can mean in the technology business. A US federal court found in August 2024 that Google had unlawfully maintained a monopoly in online search. The Justice Department argued that Google accounted for about 90 per cent of US search queries. The remedies imposed in September 2025 specifically prevent Google from extending some of the same exclusionary arrangements to Gemini and other generative AI products.
The lesson should have been obvious. Once a digital market tips towards an incumbent, competition law can spend years trying to reopen it. The Google search case was filed in 2020. Judgment on liability came four years later and remedies another year after that. AI infrastructure is being built on a much faster clock.
Big tech regulation runs behind investment
US policy is nevertheless moving in the opposite direction on AI regulation. President Donald Trump revoked the Biden administration’s 2023 AI executive order soon after returning to office. The White House’s July 2025 AI Action Plan called for the removal of federal rules considered obstacles to AI development. It also proposed taking a state’s regulatory approach into account when distributing some federal funds and reviewing earlier Federal Trade Commission actions that might unduly burden AI innovation.
The administration went further in December 2025, seeking a national framework that could pre-empt state AI laws regarded as obstacles to innovation. Its March 2026 legislative recommendations asked Congress for a “minimally burdensome” national standard.
There is a legitimate case against 50 incompatible state regimes. Regulation can also favour incumbents because large companies find compliance easier to finance than small ones. Neither proposition requires the government to leave the structure of the AI market to the companies financing its infrastructure.
This distinction has received less attention than it deserves. Rules governing what an AI model may say are different from rules governing who controls the market in which the model operates. Safety regulation can be excessive while competition policy is inadequate. Accelerating permits for data centres does not require accepting exclusionary contracts, tying arrangements or acquisitions that close markets to rivals.
Europe has made a different choice. Its Digital Markets Act imposes obligations on designated digital gatekeepers before every disputed practice has travelled through years of antitrust litigation. On July 23, the European Commission fined Google €890 million for self-preferencing in search and restrictions that prevented businesses from steering Google Play users towards alternative purchase channels. European regulation can be cumbersome, but its premise is worth noting: companies that control gateways to digital markets require constraints different from those imposed on ordinary firms.
READ | Why breaking up of Big Tech won’t set consumers free
Artificial intelligence carries American market power abroad
The consequences will not stop at the US border. The White House AI Action Plan explicitly seeks to make American semiconductors, models and applications the global standard. Success would strengthen American technological power. It could also leave governments, businesses and universities elsewhere dependent on a handful of private American companies for computing capacity and increasingly for software through which information is searched, analysed and created.
For India, this is not an argument for copying the EU rule book. It is an argument for treating dependence on AI infrastructure as an economic-policy question rather than merely a technology purchase. A country that moves much of its commerce, government computing and knowledge production onto infrastructure controlled elsewhere acquires a vulnerability that cannot be repaired by writing a tougher data-protection clause into a contract.
The United States still has strong antitrust laws and officials willing to use them. The Justice Department’s Google victories prove as much. But prosecuting yesterday’s monopoly while public policy helps finance the conditions for tomorrow’s oligopoly is an expensive way to protect competition.
AI leadership and Big Tech dominance are not synonyms. If Washington discovers the distinction only after the market has tipped, breaking the grip will take much longer than creating it.