Written by: Charlie Little Sun
Last Wednesday, during the highly anticipated Nvidia earnings call, investors were thrilled to see impressive results, and the weight that had been hanging over the hearts of countless investors finally lifted: revenue grew by over 60% year-on-year, the data center business was so hot it sold out, and the performance guidance was raised again.
However, the capital market reacted differently. After a brief surge, Nvidia's stock price fell back, and broader AI concept stocks collectively declined, with the credit spreads of companies aggressively expanding AI infrastructure widening. The public market even experienced a 2.5% drop in just over an hour.
In fact, discussions about an "AI bubble" have been rising recently: MIT stated that 95% of corporate AI pilot projects failed to generate measurable returns on investment, central bank governors warned that valuations are as distorted as they were at the end of the 1990s, and the media began to dig into the circular revenues among major AI companies.
In other words, despite the high revenue figures, the market has begun to question whether the underlying fundamentals of the entire industry can support this valuation.
The Real Bottleneck of AI: Power and Capital
Recently, Goldman Sachs, in an industry report on energy and power, made an interesting analogy, linking the current moment to two historical super cycles of infrastructure.
The railway construction of the 19th century gave rise to modern investment banking and bonds as a widely popular asset class.
The internet construction at the end of the 20th century nurtured venture capital and ignited high-risk growth stocks and IPOs.
In the current AI era, traditional stocks and bonds cannot meet the demand brought about by the explosion of power and computing power; we need a new model of capital formation, even a new capital market.
Moreover, the fundamental constraint lies in whether we can provide sufficient AI-level power and finance it without overwhelming the financial system.
The Power Dilemma
Over the past twenty years, the annual growth rate of the U.S. power grid has been less than 1%—which was manageable during the web server and smartphone era, but is disastrous for AI factories.
Analysis shows that to meet the combined demand of new data centers, electric vehicles, and industrial reshoring, the U.S. now needs to add about 80 gigawatts of generating capacity each year. However, the actual annual increase is only 50-60 gigawatts, resulting in a gap of about 20 gigawatts each year—enough to support two or three cities the size of New York.
The first instinct to fill the gap is always the intuitive option: more natural gas power plants, accelerated deployment of wind and solar storage, and hopes for a nuclear revival. But none of these can meet the demand in a reasonable timeframe:
New natural gas power plants sound appealing on paper, but have become projects that average four years in duration, with turbine supply bottlenecks leading to equipment delivery cycles of three to five years, not including approval and grid connection waiting times.
Onshore wind power, including preliminary planning and grid connection studies, typically takes three to four years, and may even drag on for nearly a decade, despite the physical construction phase only requiring six to twenty-four months.
Utility-scale solar is more modular and faster to build, with a typical construction cycle of 12-18 months, and an average battery storage development cycle of less than two years, which is why "solar + storage" accounts for over 80% of the expected new installed capacity in the U.S. by 2025.
Nuclear power, especially small modular reactors, may be the most compelling long-term answer for 24/7 AI-level power, but the first round of SMR projects in North America is also targeting commercial operation around 2030-2035.
All these solutions are essential, but in a world where grid connection queues can take four to seven years, they are merely medium to long-term solutions.
The only way to significantly speed up the process is to reuse sites that already have land, high-capacity grid connections, and power infrastructure—especially large Bitcoin mining sites. In practice, upgrading existing mining sites to AI facilities requires only a few months of renovation work (liquid cooling, power distribution, GPUs), rather than the lengthy four to seven-year journey required to apply for new grid connections from scratch.
This is precisely why AI companies are acquiring or partnering with mining companies: CoreWeave's acquisition of CoreScientific aims to repurpose its approximately 1.3 gigawatts of mining infrastructure for AI.
Although the stunning performance of Gemini 3 has led to speculation about whether TPU will replace GPU in the future, thus reducing power demand, the gradually forming consensus in the market remains that "GPU is primary, TPU is secondary." Just as the emergence of DeepSeek raised questions about GPU demand, Nvidia's GPUs have once again withstood the pressure, and expectations for power demand remain strong.
The Capital Dilemma
Since the end of 2022, when ChatGPT ignited the AI boom, the demand for AI data centers has skyrocketed, and the financing model has undergone several stages of evolution.
The first stage was almost entirely supported by the operating cash flow of hyperscale companies. When you generate hundreds of billions of dollars in free cash flow each year, you can quietly build numerous data centers and lock in a large number of GPUs. However, the scale of the current vision—the global multi-trillion-dollar AI stack—has begun to put pressure on these balance sheets.
Thus, we entered the second stage: debt and private credit. The phenomenon of investment-grade borrowing funding AI construction has surged; high-yield issuers (Bitcoin miners transforming into AI, new data center developers) have entered the junk bond market; the rapidly growing private credit system has layered customized loans, sale-leaseback arrangements, and revenue-sharing facilities on top of this.
Notably, much of the funding has never appeared as simple "debt" on balance sheets, but rather as off-balance-sheet private credit: they exist in project joint ventures, structured leases, and other off-balance-sheet instruments, turning capital expenditures into long-term obligations, making the entire stack resemble shadow financing. If the trillion-dollar AI capital expenditure forecasts are roughly accurate, banks and bondholders will not be able to support it; by 2028, private credit and these quasi-invisible structures are expected to provide a significant share of the capital behind AI data centers and power trading—possibly even the majority.
Even so, this is still insufficient, leading us to see early signs of the third stage: securitization. Asset-backed securities for data center rents and leases have quietly grown to about $80 billion in outstanding value, expected to reach about $115 billion by 2026. In terms of equity, REIT-like tools and joint ventures have split the economic interests of "land + shell + power vs. GPU vs. AI application revenue."
The public credit market has already taken note of these potential risks. Bloomberg's criticism of Meta's $27 billion off-balance-sheet data center joint venture's "creative financing," as well as comments on Oracle's aggressive leasing and lending strategies, all point to the same conclusion: tech giants cannot fully self-fund AI construction, and every new financing trick they adopt makes bond investors more anxious.
So, is this an AI bubble? To some extent, yes—but not in the way the headlines suggest.
In terms of equity, valuations are indeed eye-popping. AI-related companies occupy too large a share of market earnings, the S&P 500 trades at valuation multiples reminiscent of the internet era, and Nvidia's market cap briefly exceeded the GDP of almost all countries except China and the U.S. But equity investors at least believe they understand how to price growth and hype.
The more interesting—and dangerous—actions lie in the capital stack behind these. The issue is not that AI lacks practical uses, but that we are trying to finance a generation of infrastructure using tools and intermediaries not designed for this specific risk profile (long-term physical risks: power plants, grid upgrades; short-cycle technology risks: old GPUs may become obsolete within five years).
Returning to the historical analogy mentioned earlier: railroads were not financed solely through generic loans for crude oil, but rather the funding needs for thousands of miles of tracks and locomotives gave rise to modern investment banks and standardized railroad bonds; the internet was not simply grafted onto corporate balance sheets; it nurtured the venture capital partnership model and norms around funding other loss-making companies with equity, due to the extremely asymmetric return distribution and appreciation potential.
Therefore, the real question is: in the AI era, what should a more effective capital formation mechanism look like? What are its native financial instruments?
RWA: Financial Instruments of the New Era
On the surface, it seems Wall Street has found the answer.
"RWA" has become the annual buzzword in earnings calls and regulatory speeches, referring to tokenized government bonds, stocks, bank deposits, and on-chain repurchase experiments, and is seen as the foundational infrastructure of the financial market in the new era.
According to the SEC's narrative, it seems to be inherently the financial infrastructure of the AI era, just as railroad bonds were to steel, and startup equity was to the internet.
However, in essence, tokenized RWA is not a new form of capital; it is merely a new packaging of financial products we are familiar with: behind it still lies senior and mezzanine debt; common and preferred stock; revenue-sharing agreements, etc.
In the context of energy or data centers, this could mean tokenized shares of 20-year power purchase agreements; tokenized project equity with on-chain waterfall logic; tokenized REIT units; or short-term over-collateralized notes supported by contract GPU revenues.
So, if RWA is not novel, what real advantages can it bring compared to traditional financial instruments beyond the noise and hype? Through analysis of some early projects, we can see four practical benefits:
Fine Granularity: A $50 million project share can be split into thousands of on-chain positions, allowing position sizes to match a broader range of investment requirements.
Global Reach: As long as securities rules are followed, the same instrument can be held by funds, family offices, DAOs, or corporations from different jurisdictions without needing to rewire the underlying pipeline each time.
Programmable Cash Flow Distribution: Smart contracts can host stablecoins, enforce waterfalls and contracts, and automatically pay interest or revenue shares based on verifiable performance data, without relying on spreadsheets and intermediaries.
Fast Settlement Based on Dollar Stablecoins: You can transfer principal and interest across time zones and weekends within minutes, although the depth of the secondary market is still far thinner than traditional bond markets.
All of this sounds like a financial upgrade, but it still feels like it doesn't address the deeper capital formation issues.
In the railroad era, bonds were effective because there was a whole set of mechanisms around them that could convert steel and land into standardized securities; in the internet era, high-growth equity was effective because the venture capital partnership model could transform chaotic startups into financeable pipelines. But tokenized RWA cannot miraculously create that flywheel out of thin air.
The real financial challenge to be solved in the AI + energy cycle is not how the leading AI companies can continue to "smartly" use financial engineering to leverage debt to build AI data centers and power plants, but how to initiate, aggregate, and de-risk thousands of small distributed assets (solar rooftops, batteries, micro data centers, flexible loads), and express their cash flows in a way that global capital can truly trust.
This is precisely the gap that DePIN RWA seeks to fill, and why, in this context, energy and computing networks are more important than another vague "RWA narrative."
Energy DePIN: Long-Tail Capital Formation
This is where DePIN—the idea of using tokens to coordinate the deployment of physical infrastructure—becomes interesting.
Currently, the scale of DePIN is still small. According to Messari's 2024 report, the entire sector has about 350 tokens with a total market value of about $50 billion, trading at about 100 times combined revenue. Specifically, the energy DePIN subcategory has about 65 projects, with a total market value of less than $500 million.
If you are a traditional infrastructure investor, these numbers seem laughably small in the face of trillion-dollar AI capital expenditure plans. However, the best-designed energy DePIN forms almost perfectly align with the power bottlenecks that the AI stack is currently facing.
Taking Daylight as an example.
Its core logic is that distributed energy—rooftop solar, home batteries, electric vehicle charging stations—can be coordinated into a software-defined power plant if it can detect and pay for "flexibility," rather than just raw generation. Its "flexibility proof" mechanism pays in $GRID tokens when smart devices commit to adjusting consumption or charge/discharge behavior during high-pressure moments; energy companies burn $GRID to purchase access to that flexible capacity.
On this basis, $GRID serves as a currency backed by energy, touching every part of the stack: installation discounts for homeowners; payments for data and analytics; staking and derivatives for regional capacity mispricing; insurance for off-chain capacity commitments. In its U.S.-only model, the total across physical and financial energy markets is about $1 trillion annually.
Daylight's model is tightly coupled with the existing grid. If you believe that AI data centers will primarily be located within or near the current transmission grid, and that utility companies are willing to pay high prices for flexibility, then this is an important selling point. If grid connection delays and regulatory issues slow everything down, that is also a risk.
In contrast, we have Arkreen.
If Daylight is "grid-native, U.S.-centric," then Arkreen is "grid-agnostic, globally oriented."
It connects distributed renewable energy resources to a Web3-enabled data and asset network. Participants install "mining machines" or connect via API; the network records verifiable green energy generation data and tokenizes it into renewable energy certificates and other green assets.
Arkreen has connected over 200,000 renewable energy data nodes, issued over 100 million kilowatt-hour tokenized RECs, and facilitated thousands of on-chain climate actions.
Its vision is clear globalization and long-tail: a peer-to-peer energy asset trading network where households and small producers can connect their DERs to the DePIN system, earn tokens through "impact profit" activities, and indirectly form virtual power plants or green AI offsets.
Each of these projects, when considered individually, cannot fund the next 1-gigawatt data park for hyperscale enterprises. But they point to possible forms of "capital formation" in the AI era—if we stop thinking only in terms of nine-figure project financing blocks and start thinking in atomic-level "kilowatt-hours."
This is also where the concerns about centralization highlighted by the crypto + AI narrative and a16z's latest crypto status report come into play: unregulated AI tends to centralize—large models, large clusters, large clouds. In contrast, blockchain excels at aggregating numerous small distributed contributions and granting them access to a liquid global market.
Connecting Kilowatt-Hours with AI Tokens: The Crypto Bridge
Currently, the value chain from marginal "kilowatt-hours" to "AI tokens" is fragmented.
Power plants sign PPAs with utility companies; utility companies or developers contract with data centers; data centers sign contracts with cloud providers and AI companies; AI companies sell API usage rights or seats; somewhere at the top of the stack, users pay a few dollars to run an inference.
Each link is independently financed, with different investors, risk models, and jurisdictional constraints. The opportunity, and the crypto-native version of "capital formation," lies in making this chain transparent and programmable.
On the supply side, you can tokenize kilowatt-hour-related outputs, representing claims to specific renewable energy generation flows; tokenize RECs and carbon credits; and tokens representing flexible capacity commitments from batteries, smart devices, and VPPs. Projects like Arkreen demonstrate that this is technically and commercially feasible at a reasonable scale.
In the midstream, you can express infrastructure as tokenized RWAs: equity and debt of data centers, grid connection upgrades, behind-the-meter generation and storage, GPU clusters. Here, traditional securitization is still happening, but on-chain tracks can make it more transparent: when investors purchase layered products, they know exactly which assets back them, and cash flows settle in stablecoins that move in minutes rather than days.
On the demand side, you can link energy and computing with AI-native tools: GPU hour tokens, inference second credits, or even application-layer "AI service" tokens. As agent AI systems mature, some of these tokens will be directly held and spent by software agents—programs capable of assessing where to purchase computing and power at the margin and dynamically arbitraging between providers.
Thus, every marginal kilowatt-hour used by AI models, from its origin (rooftops, solar power plants, nuclear SMRs) to its consumption in GPU racks, and its monetization in AI applications, is traceable, priceable, and hedgeable.
This does not require every link to be on the same chain or priced in the same token. Rather, it means that the state of each link is machine-readable and can be stitched together by smart contracts and agents.
If achieved, you effectively create a new form of capital: any investor, anywhere, can choose where to take on risk in this chain—energy, grid, data center, GPU, AI application—and purchase tokenized exposure of corresponding scale and duration.
The balance sheets of hyperscale enterprises will not disappear, but they will no longer be the only way to warehouse that risk.
Conclusion
This story is not a foregone conclusion.
The combination of "big companies + big capital" is sufficient to accomplish all of this independently. Hyperscale enterprises may decide to directly own the energy stack through vertical integration and keep cash flows internal.
Long-tail energy DePIN may never surpass centralized projects.
But even if only a small portion of AI-related energy and computing is ultimately financed and coordinated through DePIN and tokenized RWAs, we have answered the open questions left by Goldman Sachs and a16z's call for decentralization.
At this moment, computing power and electricity are intertwining in unprecedented ways, and the form of capital is quietly reshaping.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
