EU and US dismantle AI guardrails as Nvidia earnings fuel investment frenzy

The regulatory architecture that Europe spent years constructing around artificial intelligence is being quietly taken apart. In Brussels, the European Commission has confirmed it will delay central provisions of the AI Act, only months after the legislation entered force in August 2024, and rewrite the General Data Protection Regulation so that companies can feed personal data into training models without seeking fresh consent. Across the Atlantic, the United States is going further: congressional language tucked into the National Defense Authorization Act would bar states from passing their own AI rules and authorise the Department of Justice to sue any that try.

Brussels chooses growth over guardrails

The Commission's "digital omnibus" packages the AI Act, GDPR, the ePrivacy directive and the Data Act into a single streamlining exercise. The stated aim is to reduce compliance costs and end what officials call "cookie banner fatigue". In practice, the GDPR changes would create a broad lawful basis for processing personal data for AI development, removing the need for explicit consent in many cases. The AI Act delays push back obligations for high-risk systems, general-purpose models and conformity assessments that were due to start applying in 2025 and 2026.

The shift follows a year of pressure from capitals and industry. Mario Draghi, the former Italian prime minister and European Central Bank president, warned in a 2024 report commissioned by the Commission that the EU had fallen behind the United States and China in the technologies that would determine future growth. His diagnosis, fragmented markets, insufficient venture capital, regulatory drag, has become the consensus in the Berlaymont. The economy commissioner, Valdis Dombrovskis, has echoed the argument that Europe cannot regulate its way to competitiveness.

Civil society groups describe the omnibus as a massive rollback of digital protections. The European Data Protection Board has not yet issued a formal opinion on the GDPR amendments, but its previous guidance makes clear that purpose limitation and data minimisation principles are not optional. The Commission's impact assessment, published alongside the proposal, acknowledges that the changes increase the risk of function creep but argues they are proportionate to the economic gain.

Washington goes further: federal pre-emption of state law

If Brussels is loosening the reins, Washington is cutting them. The NDAA provision, inserted by Republican leadership with White House backing, would establish a ten-year moratorium on state-level AI regulation. The Justice Department would be directed to sue non-compliant states, almost certainly California and Colorado, which have already enacted transparency and bias-audit requirements for automated decision systems. A similar executive order drafted by Donald Trump last week frames the patchwork of fifty state regimes as a strategic liability. "You can't go through 50 states," Trump told the US-Saudi Investment Forum. "Fifty is a disaster. You'll have one woke state and you'll have to do all woke."

The move has provoked a rare coalition of opposition. More than 200 state representatives and senators from across the political spectrum signed a letter arguing that the provision violates the Tenth Amendment and strips legislatures of their police powers to protect consumers from discrimination, deepfakes and algorithmic harm. A previous attempt at a federal moratorium failed in the Senate by a 99-1 vote earlier this year. The NDAA must still pass both chambers and survive a likely veto threat, but the momentum is with the pre-emption camp.

Nvidia's numbers keep the bubble inflated

The deregulatory push on both sides of the Atlantic is propelled by a single commercial reality: the capital intensity of the current AI wave is unprecedented, and the company supplying the shovels is printing money. Nvidia reported $57.01 billion in revenue for the quarter ended 26 October, up 62% year on year. Diluted earnings per share came in at $1.30, ahead of the $1.26 consensus. Datacentre revenue, the segment that sells the H100 and Blackwell GPUs powering large-language-model training, hit $51.2 billion against expectations of $49 billion. The company guided for $65 billion in the current quarter, well above the $61 billion analysts had modelled.

Jensen Huang, the co-founder who has run Nvidia since 1993, dismissed the bubble narrative directly on the earnings call. "There's been a lot of talk about an AI bubble," he said. "From our vantage point, we see something very different. As a reminder, Nvidia is unlike any other accelerator. We excel at every phase of AI from pre-training to post-training to inference." The market cheered initially, pushing the S&P 500 and Nasdaq higher, but the rally evaporated within 24 hours. The S&P 500 closed down 1.6%, the Nasdaq 2.2%. Robert Pavlik, senior portfolio manager at Dakota Wealth, captured the scepticism: "The people who are selling the semiconductors to help power AI doesn't alleviate the concerns that some of these hyper-scalers are spending way too much money on building the AI infrastructure."

Antitrust cases collapse under the weight of AI competition

The same dynamic, massive AI investment as proof of competitive vitality, has now rescued two US tech giants from structural breakup. In the District of Columbia, Judge James Boasberg dismissed the Federal Trade Commission's case seeking to force Meta to divest Instagram and WhatsApp. Boasberg wrote that the market had "changed markedly" since the suit was filed in 2020, citing TikTok's explosive growth and YouTube's entrenched position as evidence that Meta no longer holds monopoly power in social networking. "Even if YouTube is out, including TikTok alone defeats the FTC's case," he concluded.

Weeks earlier, Judge Amit Mehta in the Google search monopoly case reached a parallel conclusion. While he found that Google maintains an illegal monopoly in general search, controlling roughly 90% of the global market, he declined to order the divestiture of Chrome, the world's dominant browser. Generative AI, he ruled, had "altered that market permanently" and introduced competition Google had not faced in decades. OpenAI, Anthropic and others were now credible challengers. In both cases, the judges accepted the defendants' argument that the AI arms race itself constitutes a competitive constraint that did not exist when the complaints were drafted.

Europe's strategic dilemma sharpens

For European policymakers, the US rulings create an uncomfortable feedback loop. The Commission is weakening its own rules to help European firms compete with US hyperscalers, while US courts cite the very existence of those hyperscalers' AI investments as a reason not to break them up. The result is a transatlantic convergence on deregulation, but one that leaves European companies structurally dependent on American infrastructure. No European firm trains frontier models at the scale of OpenAI, Anthropic, xAI or Google DeepMind. Mistral AI, the French champion, has raised just over €1 billion, a rounding error against the $51 billion Nvidia booked in datacentre revenue last quarter alone.

The GDPR changes illustrate the bind. By creating a broad lawful basis for AI training, the Commission hopes to unlock European data for European models. But the same provision applies to any company offering services in the EU, including the US giants that already hold the compute, the talent and the distribution. European data may simply flow more freely into American models. The AI Act delay compounds the problem: high-risk obligations for general-purpose models are postponed, giving non-EU providers more time to entrench before compliance costs bite.

What happens next

The digital omnibus now moves to the European Parliament and Council, where co-legislators can amend or reject the GDPR and AI Act changes. The Parliament's lead committees, LIBE for data protection, IMCO for the AI Act, have signalled resistance to weakening consent requirements. A trilogue before the 2025 summer recess is possible but not guaranteed. In Washington, the NDAA faces a conference committee between the House and Senate versions; the pre-emption language survives in the House bill but has not yet cleared the Senate. The 200-plus state legislators have vowed to lobby every Senate office. Meanwhile, Nvidia's next earnings call in February 2026 will test whether the $65 billion guidance holds. If it does, the deregulatory momentum on both sides of the Atlantic will only accelerate.