Amazon confronts data centre backlash as AI’s data infrastructure faces cost and compliance tests.
The AI infrastructure race and the ROI question.
Massive investments in data infrastructure are driving the next phase of enterprise AI - but not without growing pains. Companies worldwide are pouring capital into data centres and cloud infrastructure purpose-built for AI. Samsung, for example, just committed US$1 billion to a KKR-backed AI data centre venture ([1]), reflecting industry conviction that more storage and compute capacity are critical to AI leadership. Analysts forecast that the global build-out for AI data centres could reach an astounding $30 trillion or more by 2050 ([2]), dwarfing past tech infrastructure cycles. This “AI infrastructure boom” is being likened to railroads or electrification, only bigger, as firms race to provide the digital plumbing for intelligent systems.
Yet business leaders are increasingly asking whether these investments will translate into real value. A new Reuters analysis highlighted concerns that AI’s commercial returns may not arrive quickly enough to justify the enormous spend. JPMorgan analysts noted that broad productivity gains from AI remain elusive so far ([3]), and Bain & Co. has argued that efficiency improvements alone won’t create the trillions in new revenue needed to support such spending ([4]). The sobering reality is that many early AI projects haven’t delivered ROI: a global survey by McKinsey found that while 88% of companies had embedded AI in at least one function, only 39% reported any bottom-line impact ([5]). And last year, MIT researchers concluded a stunning 95% of enterprise generative AI pilots failed to produce measurable profit ([6]). These figures highlight a looming gap between AI ambitions and business outcomes.
The takeaway for executives is clear: infrastructure alone isn’t a silver bullet. The companies pulling ahead are those coupling big bets on AI compute with equally strong data strategies. This means ensuring that data is high-quality, integrated and accessible so that advanced models can actually drive insights and automation that matter to the business. Indeed, Forrester Research found firms with a mature, unified data strategy are four times more likely to see their AI initiatives exceed expectations ([7]). In the AI gold rush, the real competitive moat isn’t just how many GPUs and data centres you can buy - it’s how effectively you harness the data flowing through them. Leaders are now focused on aligning these infrastructure investments with concrete use cases and robust data governance, to turn expenditure into sustainable ROI.
Data governance becomes non-negotiable.
Even as companies scale up AI infrastructure, they face another make-or-break challenge: governing and protecting the data that fuels AI. This week saw a vivid example from the consumer tech world that carries lessons for every enterprise. Apple announced new safeguards in macOS to rein in how much of a user’s data AI-driven apps can access ([1]). The move comes after a controversy where a desktop AI assistant reportedly read a user’s private messages without explicit permission, exploiting an overly broad “Full Disk Access” setting ([2]). Apple’s response - requiring more explicit user consent for apps wanting to scan files, emails or messages - shows how vital data privacy and control have become as AI systems grow more capable. Although focused on consumer devices, the principle applies equally in business: without strong data governance and clear guardrails, AI can inadvertently expose sensitive information or make decisions on faulty, biased data.
For corporate leaders rolling out AI, trust and quality are as important as raw computing power. An AI model is only as good as the data it’s trained on and the data it’s allowed to consume. Flaws in data - whether inaccurate, biased, or insecurely handled - can lead to everything from bad business decisions to regulatory penalties. Ensuring data quality and consistency across the enterprise is now seen as a prerequisite for AI readiness. Indeed, industry analysts have long warned that most organisations attempting to scale AI will fail without modernised data architecture and governance in place ([3]). Companies at the forefront are investing in data catalogs, master data management, and real-time integration platforms to break down silos and prevent “Garbage In, Garbage Out” syndrome in AI outputs. They’re also implementing strict access controls, audit trails and privacy safeguards, so that employees and AI systems alike can only access the data they need. The goal is to build an environment of trusted, high-quality data that AI can leverage effectively - because without trust in the data, even the best AI models will falter.
Regulators and the shifting data advantage.
Meanwhile, the regulatory environment around data is heating up, potentially reshaping how companies think about data as a competitive asset. In the EU, new Digital Markets Act measures are compelling tech giants to open up their data vaults - for instance, by requiring Google to share anonymised search data with rival search engines and AI assistants by 2027 ([1]). Regulators argue this will “rebalance the playing field” in online search and AI development, since incumbents’ massive data troves are seen as key sources of their market power. Google is fiercely resisting such data-sharing mandates on privacy grounds and has appealed the EU order in court, warning of “irreversible harm” if it’s forced to hand over its valuable search query data ([2]) ([3]). The outcome of this battle could set a global precedent for how data is regulated in the AI era.
On the other side of the Atlantic, the U.S. government is also signalling a tougher stance on AI oversight. President Trump recently appointed a national “AI czar” to lead a federal task force on AI’s risks and opportunities, with a report due in 120 days ([4]). Meanwhile, one of the world’s most-watched AI startups, Anthropic, made headlines with an unusually stark warning in its IPO prospectus: it cautioned that increasing government scrutiny of advanced AI could damage its relationships with customers and partners ([5]) ([6]). Notably, Anthropic derives less than 1% of its revenue from government contracts ([7]) - the concern is more about a chilling effect on the wider market if regulators or public opinion turn against AI. The fact that a company eyeing a multi-billion (or even multi-trillion) valuation would highlight “catastrophic” AI risks in a filing shows how seriously the industry now takes regulatory and ethical issues.
For business leaders, these developments are a clear signal. Whether it’s meeting new data-sharing rules, ensuring compliance with privacy laws, or addressing public ethical concerns, a proactive data strategy is now essential. Companies that treat data as a competitive moat must be prepared for greater oversight and demands for transparency. Those that build flexibility - such as robust data governance, consent mechanisms, and auditability - into their AI data architectures will not only stay on the right side of regulators, but also stand a better chance of earning and keeping customer trust in an increasingly data-conscious market.