UK Abandons AI Copyright Opt-Out, NVIDIA Restarts China H200 Sales, and AI Bubble Fears Grow

Three stories are converging to reshape the AI industry's economic foundations this week. Each is significant independently. Together, they paint a picture of an industry entering a new phase where regulatory costs, geopolitical constraints, and valuation discipline collide with the relentless pace of technical progress.

Story 1: UK Government Kills AI Copyright Opt-Out

What Happened

On March 18, the UK government published its long-awaited AI copyright report under the Data (Use and Access) Act 2025. After receiving more than 11,500 consultation responses — with 90 percent opposing the opt-out model — the government abandoned its preferred approach entirely.

The opt-out model would have allowed AI companies to train on copyrighted works unless creators explicitly blocked them. Instead, the UK is pursuing transparency obligations requiring AI companies to disclose their training data, combined with market-driven licensing mechanisms.

Why It Matters

This is the first G7 nation to formally reject permissive AI training data policies after a comprehensive consultation process. The UK's pivot to transparency creates policy convergence with the EU AI Act's Article 53 requirements, establishing a transatlantic regulatory baseline that AI companies must meet.

For a deeper analysis of what this means for the global licensing landscape, read my full breakdown of the UK copyright decision and its implications.

The decision provides early validation for my prediction that at least five G7 nations will enact mandatory AI training data licensing by Q4 2027. The UK is domino number one.

Key Details

• The government is not legislating a specific licensing framework yet
• Transparency obligations will require AI companies to disclose training data sources
• Dispute resolution mechanisms for licensing negotiations are being developed
• No retroactive requirements — existing models are not immediately affected
• Timeline for final regulations remains undefined

Sources

• GOV.UK Official Report on Copyright and AI
• Hogan Lovells Analysis — Copyright and AI Statement of Progress
• BM Magazine — Licensing Emerges as the Battleground

---

Story 2: NVIDIA Restarts H200 Production for China

What Happened

Jensen Huang announced on March 17-18 that NVIDIA is restarting H200 chip manufacturing for China after securing multiple export licenses from the US government. The potential scale is enormous — up to 75,000 chips per customer, with a total of up to one million processors under consideration. A 25 percent duty and inspection requirements apply.

This reverses weeks of uncertainty during which NVIDIA had only one approved license for China sales. The shift suggests the Trump administration has calculated that controlled access is preferable to driving Chinese companies toward domestic chip alternatives.

Why It Matters

This is a significant geopolitical development on multiple levels:

1. Revenue impact: One million H200 chips at estimated pricing represents $20 billion or more in revenue for NVIDIA — enough to materially affect quarterly earnings.

2. Competitive dynamics: If Chinese companies receive H200 access, it reduces their urgency to develop domestic alternatives. This may be the strategic intent — keeping China dependent on American silicon rather than accelerating their chip independence.

3. Copyright intersection: Chinese AI companies operating under more permissive copyright frameworks will train models on this hardware without the licensing costs their Western competitors face. The training cost asymmetry grows.

For context on the chip geopolitics, see my analysis of GPU geolocation verification and the AI chip battlefield.

Sources

• Axios — NVIDIA Fires Up H200 for China
• The Register — NVIDIA H200 China Sales Resume
• Semafor — NVIDIA Restarts H200 Manufacturing for China

---

Story 3: AI Bubble Fears Reach the Mainstream

What Happened

Bloomberg published a major feature on March 18 titled "Is an AI Bubble Set to Burst?" examining whether current AI valuations and spending levels are sustainable. The same day, Benchmark partner Bill Gurley warned that AI capex-to-revenue ratios now exceed dot-com era levels. Moody's has modeled scenarios with a 40 percent AI valuation drop.

Meanwhile, SaaS companies are already feeling the pressure. Salesforce and ServiceNow have lost more than 20 percent of their market capitalization in 2026 as AI agents undercut their subscription-based pricing models.

Why It Matters

The bubble question is not new, but three factors make this moment different:

1. Mainstream financial media engagement: Bloomberg and Moody's are not tech blogs speculating. When the credit rating agencies start modeling AI downside scenarios, institutional investors pay attention.

2. SaaS carnage provides a template: The 20 percent market cap destruction in traditional software companies shows that AI disruption is real — but it is destroying existing value faster than it creates new revenue for AI companies.

3. Regulatory cost layer: The UK copyright decision adds a new cost category (licensing) on top of compute, talent, and infrastructure. This compresses the already-narrow path to profitability for most AI companies.

I have been tracking this dynamic across multiple analyses. The $2.5 Trillion Paradox explored why 90 percent of companies see zero returns from AI spending. My prediction on enterprise AI spending correction in Q2 2026 remains active. The convergence of regulatory costs, valuation pressure, and the capex-revenue gap suggests the correction may be broader than pure enterprise spending.

Sources

• Bloomberg — Is an AI Bubble Set to Burst?
• Fortune — Bill Gurley AI Spending Warning

---

The Convergence

These three stories are not isolated events. They describe a single dynamic: the AI industry is entering its regulatory and economic maturity phase. The period of unconstrained spending, unquestioned valuations, and unregulated data practices is ending.

What replaces it will be messier, slower, and more expensive. It will also be more sustainable. Industries that survive their bubble phases emerge stronger, more efficient, and more focused on actual value creation. The internet went through this after 2001. Cloud computing went through it after 2010. AI is going through it now.

The companies that survive will be those that treat regulatory compliance as a competitive moat rather than a cost burden, that build licensing relationships proactively rather than reactively, and that demonstrate real revenue — not just user growth — to justify their capital consumption.