Chips, Data, Mergers and Acquisitions: The Survival Line of Enterprises in the AI Era Rewritten by Regulations

“Can You Do It” Is More Valuable Than “Will You Do It”

What is the biggest asset of an AI company? In the past, everyone said it was computing power, models, or GPUs. But looking at the current situation in 2025, the real scarcity has become a series of licenses—a license to “exist.”

The Chinese Ministry of Commerce’s review of Meta’s acquisition of AI agent developer Manus appears to be a routine investment review on the surface, but behind the scenes, it’s a layered check on cross-border data transfer, technology export, and foreign control rights. Manus generates over $100 million annually, headquartered in Singapore but with Chinese roots. Such cases are perfect “test beds” for regulators.

The question is no longer “What can you build,” but “Who can you sell to,” “What can you buy,” and “Where are you willing to move data.”

Chip Trade Turns Into a Geopolitical Game

Nvidia’s H200 chips are extremely popular in China—Chinese companies have ordered over 1.2 million units at $27,000 each, but Nvidia only has 700,000 units in stock. Under normal circumstances, this would be a business deal, but what’s abnormal is that the Chinese government has directly ordered local tech companies to suspend new orders, citing the need to balance domestic and imported chips.

This is the new game rule: national policies directly determine market demand.

To mitigate risks, Nvidia now requires full prepayment from Chinese customers, with no refunds, changes, or withdrawals. Some orders even require guarantees or insurance instead of cash. The reason is simple—recently, the company wrote off $5.5 billion due to US export bans. A misjudgment could leave millions of chips stranded.

Listen to Nvidia CEO Jensen Huang’s words: “If the orders are still being fulfilled, it’s because those orders can be fulfilled.” Implicitly, orders that cannot be fulfilled have already been canceled. In December last year, the US claimed it would allow the export of H200 chips to China, imposing a 25% tariff and oversight by the Commerce Department. But then there were reports of a review of advanced chip sales. Policies swing like a pendulum, and companies must always be prepared for a slap in the face.

By the end of 2024, Beijing banned state-owned data centers from purchasing foreign AI chips. The result? Nvidia’s market share in China plummeted from 95% in 2022 to 0% in 2025. It’s not a competitor choking it out; it’s a regulatory ban that pushed it out.

Mergers and Acquisitions Are No Longer Business Deals, But Geopolitical Events

Meta’s acquisition of Manus, the EU’s review of Alphabet’s $32 billion Wiz deal—these are not ordinary M&A. Regulators are not just looking at financial figures but at risks related to technology transfer, cross-border data flow, and who controls AI decision-making.

Meta says it will not operate Manus in China post-acquisition, and Manus will cease its China operations—trying to satisfy both sides with this promise. But the question remains: how do you ensure that the expertise of an AI team “does not cross borders”? The talents, architecture, and code that can influence the real world—how are these defined as “exports”?

Recently, the EU strengthened its standards for foreign investment review. Before February 10, regulators must decide whether to approve the Wiz deal or launch an in-depth investigation. The reason is the so-called “AI stack,” which involves the security levels of cloud infrastructure and the building of trust in cloud computing. In other words, as long as it touches AI and cloud infrastructure, the deal could be halted.

Cross-Border Data, From a Technical Issue to a Political One

The EU AI Act begins phased implementation from February: prohibitions in February, general model requirements in August, full compliance by August 2026. Bruxelles has made no extensions. This is not a suggestion; it’s a hard requirement.

Meanwhile, the EU is investigating whether AWS and Microsoft Azure constitute “gatekeepers” of cloud computing. Once defined, regulation will extend from model behavior to infrastructure, data, and distribution networks—all the backend systems supporting AI training and deployment become objects of scrutiny.

On the US side, there’s another chaos: the Trump administration requires federal agencies to review state-level AI laws, claiming some state regulations on information disclosure and system result changes violate the Constitution. This creates new uncertainties—companies must navigate state laws while guarding against federal interference, with standards constantly shifting.

Even more complex is fiscal policy. The US Treasury’s investment review process now restricts US investments in sensitive technologies (including AI) in risk countries, requiring reporting and licensing. In other words, “Can you finance” is now tied to “Can you export” and “Can you hold shares,” all subject to regulatory approval.

Rules Are Changing, Companies Are Adapting

The reality is this: successful AI companies are no longer those with the strongest technology, but those that survive the ever-changing regulatory storms the longest. Compliance is not a cost center but a strategic core. Contracts are not just business tools but risk hedging instruments. Corporate structures cannot focus solely on efficiency—they must also consider geopolitical resilience.

The key question has shifted from “How” to “Can you”—can you do it? And the definition of “can” is being rewritten constantly by policies from regulators worldwide. Companies that can survive will be those that can innovate and adapt at any moment, showing the greatest resilience in the face of cross-border data transfer, export controls, and investment reviews.