In the past decade, global financial innovation has predominantly revolved around “digitalization”: internet platforms, ETFs, stablecoins, and now RWA (real-world asset) tokenization. But with the rapid development of AI, a deeper question has begun to emerge: What truly supports the era of AI?

The answer may not lie in the code itself, but in the reality behind the code.

AI is redefining asset layer structure

For a long time, AI has often been imagined as a “dematerialization” technological revolution, but in reality, it is quite the opposite. AI has not weakened the dependence on the real world; rather, it is strengthening this dependence. Every instance of model training, inference, and deployment requires substantial real-world infrastructure, including data centers, power grids, cooling facilities, advanced hardware, and industrial resources. In other words, what seems like a digital system is essentially an industrial system.

And within this system, what is truly irreplaceable are metals and real resources. Materials like copper, silver, and gold collectively determine conductivity, durability, and infrastructure performance, and these capabilities are not replaceable by software.

S&P Global predicts that the demand for copper from data centers alone will grow from 1.1 million tons in 2025 to 2.5 million tons by 2040. At the same time, the market anticipates a global refined copper shortfall of 304,000 tons by 2025, which could expand to 6 million tons by 2035. As AI infrastructure continues to expand, demand for these real resources is rapidly increasing, while supply remains structurally constrained.

More and more industry observers believe that this is not a short-term cyclical problem, but a kind of long-term structural change. What truly limits AI expansion may no longer just be computing power itself, but the “physical layer” constituted by energy, metals, and real infrastructure. This layer is also beginning to form its own logic of scarcity, pricing logic, and asset system.

A new “asset layer” structure is forming

In this context, the market is beginning to reinterpret the relationship between the physical layer, financial layer, and digital layer:

• Physical Layer: metals, energy, real resources

• Financial Layer: government bonds, ETFs, structured products

• Digital Layer: tokenized infrastructure, programmable assets

The digital layer is built upon the financial layer, which ultimately relies on the physical layer of the real world. For decades, the market has long rewarded “upper layer assets,” including stocks, ETFs, internet platforms, and digital financial infrastructure; but now, AI is redirecting market attention back to the underlying real resources themselves.

Tokenization will not create value out of thin air

This also explains why most RWA projects have not truly materialized. The issue is not entirely with the technology itself, but with asset selection.

Tokenization does not create value from nothing; it simply reconnects the assets that the market has already trusted. For an asset to achieve true tokenization, it usually needs to possess mature demand, deep liquidity, and institutional consensus; otherwise, tokenization often only brings complexity, not value.

From this perspective, the current path of tokenization development is actually very reasonable. The first to be tokenized are sovereign debts because they have the most mature liquidity and credit systems globally; then came gold, which has a centuries-old global consensus; and next is silver, which possesses both reserve properties and industrial demand. The direction that can truly expand in the future will be the industrial materials that the real economy genuinely relies upon.

It is worth noting that the order of tokenization does not solely depend on the importance of these assets to AI infrastructure. The significance of copper and industrial metals is not lower than that of gold. What truly determines the order is where market consensus is initially established, and each step inherits the credibility accumulated from the previous one.

This is also the core logic of the RWA tokenization platform Matrixdock: starting with assets that the market has already built long-term trust in, including sovereign debts, gold, and silver. Currently, Matrixdock manages over $200 million in on-chain assets and serves institutional clients that require both the stability of real-world assets and the programmability of on-chain infrastructure.

Gold ETFs and gold tokens are heading in different directions

In the field of gold, a new change is also emerging.

Gold ETFs have been one of the most successful financial innovations of the past twenty years. They solved the issues of physical gold being difficult to trade, lacking liquidity, and high holding costs, allowing gold to be easily bought and sold by ordinary investors for the first time, much like stocks.

However, the core logic of ETFs essentially allows investors to “have exposure to gold” rather than actually bringing gold into the financial system. The gold in ETFs primarily remains within the traditional financial holding system, making it difficult to realize programmable settlement, native collateralization, or cross-system interaction like on-chain assets.

With the development of programmable finance and on-chain finance, the market is beginning to pose new questions: Beyond “holding,” can gold actually participate in financial activities? For instance, can it achieve instant settlement, cross-border collateralization, and flow without custodial intermediaries?

In a sense, this is also a fundamental difference between gold tokens and gold ETFs. Gold ETFs address the “investability” of gold, while gold tokens explore a broader functionality of gold within the digital financial system.

Matrixdock's gold token XAUm is developed based on this logic. Currently, XAUm has around $74 million in gold assets under management (AUM) and accumulated transaction volume exceeds $100 million. Its goal is not merely to replicate ETFs but to enable gold to begin entering the on-chain financial system.

From “store of value” to “functional asset”

And gold may just be the starting point.

As AI infrastructure continues to expand, more and more industrial materials are transitioning from “bulk commodities” to “strategic resources.” Silver is used for conductivity, while copper supports energy and connectivity infrastructure, and industrial metals are becoming the real physical underlayer behind AI infrastructure.

Especially for silver, its supply and demand structure has already begun to change. Silver has faced a structural supply shortage for the fifth consecutive year, and it is projected that the gap will widen to 46.3 million ounces by 2026. Industrial demand from solar energy, electric vehicles, and AI infrastructure is consistently driving up consumption while mineral supply growth struggles to keep pace.

If gold represents “store of value,” then industrial metals are more like “functional assets.” However, the tokenization of industrial metals will not completely replicate the pathway of gold. Since industrial metals are consumed, the focus is not just on reserve properties but also on how to establish connections for operation and circulation between the real commodity system and digital infrastructure.

Matrixdock's silver token XAGm is the first step in this direction. Its positioning is to connect the reserve logic of precious metals with the functional demands of industrial metals. As the roadmap further delves into the “physical layer,” the direction is becoming clearer: those industrial metals that AI infrastructure heavily relies upon may be becoming important components of the next stage of the on-chain asset system.

In a sense, the asset layer is evolving towards a more strategically significant direction that is more grounded in the real physical world and is also more programmable. The assets that are truly worthy of tokenization in the future may not just be the “easiest to digitize” assets but rather those important assets that the real economy has long depended on.

Original link