Contrary to the prevailing interpretation, Goldman Sachs' declaration that "the most capital-hungry investment cycle in history has arrived" is not a bullish signal for the sectors it names. It is a confession. A bank that intermediates the flow of global capital does not publish a note titled "unprecedented capital intensity" because it wants investors to hold cash. It publishes because the deal pipeline demands one thing above all: allocation.
Read the sentence again. A capital-intensive cycle that will "reshape global economic structures" is a polite description of a multi-trillion-dollar reallocation of savings into fixed assets โ data centers, transmission lines, power plants โ that will not produce meaningful cash flow for years. In crypto, we have a precise term for a structure where the promoter's incentive to sell the asset is structurally higher than the asset's ability to return present value. That term is a rug pull. Goldman's capital cycle is a rug pull scaled to the entire global economy, executed not through a malicious smart contract, but through the far more effective mechanism of institutional consensus. The exit liquidity is you, your pension fund, and every bond buyer who believes that the adjective "unprecedented" converts into the noun "profit."
The macro map needs to be drawn before the crypto implications become legible. The capital-hungry cycle is overwhelmingly concentrated in one cluster of demand: the physical layer of artificial intelligence. Hyperscaler capital expenditure has moved from a maintenance run-rate of roughly $150โ200 billion per year in the early 2020s toward consensus projections of over $1 trillion annually by 2028. This is not a speculative forecast in the ordinary sense; the cumulative backlog of signed power purchase agreements, data center pre-leases, and grid interconnection requests constitutes a committed pipeline that infrastructure finance has never seen.
The capital structure question is where crypto analysis must begin. This cycle is debt-heavy at the margin. Investment-grade bond issuance has been increasingly routed toward AI infrastructure and energy assets, while the equity component remains concentrated among a handful of vertically integrated hyperscalers. The global financial system is therefore being asked to warehouse trillions of dollars of long-duration, fixed-cash-flow obligations at a moment when the dominant trend in the developed economies is fiscal deficit expansion, not savings growth.
Here is the part the bank's note does not emphasize: a capital cycle is a liquidity transfer mechanism. It takes liquid, flexible capital from the margin and converts it into illiquid, geographically fixed steel and silicon. Every dollar that becomes a data center is a dollar that no longer functions as a risk cushion. I observed this mechanism first in 2021, during the NFT boom, when I traced the correlation between NFT trading volume and Ethereum gas price spikes. The apparent demand for digital collectibles was, in significant part, institutional wash-trading inflating perceived interest while draining actual liquidity. The analytical mirror is identical today. The demand for AI infrastructure is real. The demanders are borrowing against their own future cash flows to finance it, and the stablecoin supply โ the marginal liquidity of the crypto market โ is conspicuously failing to mint its way into this cycle. The capital is flowing into concrete, cables, and chips rather than into on-chain settlement layers.
This is the first technical signal that crypto's best trades in this cycle are not the ones that touch the infrastructure buildout directly. To understand where the opportunity actually sits, the cycle must be decomposed into three channels: the energy-compute arbitrage, the decentralized physical infrastructure (DePIN) narrative, and the tokenization of infrastructure debt. Each channel has a distinct incentive structure, and each has a distinct failure mode. The decomposition follows.
The energy-compute arbitrage is real. The tokens are not.
Let me begin with the one channel where the convergence is verifiable. In 2024, I published a framework predicting that AI computing markets would converge with crypto mining economics. The mechanism was obvious in retrospect: bitcoin miners spent a decade perfecting the procurement of firm, low-cost power with interconnected transmission capacity. These are precisely the inputs the hyperscaler AI buildout cannot source fast enough. The transactions that subsequently emerged โ miners leasing data center capacity, converting entire facilities to AI compute contracting, or entering multi-year hosting agreements with AI labs and cloud providers โ are not narrative plays. They are committed off-chain revenue contracts, frequently structured at valuations multiples of the miner's public equity trading price.
The technical detail that most infrastructure-token analysis misses is where the value is captured. It is captured at the parent-company level, not the protocol level. When a publicly listed miner announces an AI hosting agreement, the equity re-rates based on contracted future revenue. The token โ if the company has one โ holds no contractual claim to that AI revenue. It carries only a residual claim on hashrate, which the company can redirect at its discretion. This is a structural separation of the asset from the cash flow, and it has a direct analogue in the financial architecture of the last bull cycle.
In my structural audit of Uniswap V2 in 2017, the uncomfortable discovery was that the market was pricing the existence of a mechanism rather than its failure modes. Investors assumed that the constant product formula was the business, when the business was the liquidity that the formula incentivized. A formula without liquidity is a mathematical curiosity. The equivalent here: investors treat the mining token as a proxy for the AI business, when the AI business runs through the corporate entity and the banking syndicate, not through the token. There is no hook in the token contract that captures the new revenue.
Uniswap V4's hook architecture solves a different problem โ the ability to customize liquidity provision around concentrated positions and external triggers โ but the principle is instructive. A financial architecture that cannot attach a claim to new cash flows cannot capture the value of those cash flows. The AI-converted miner's token is precisely such an architecture: it is infrastructure with a governance token wrapper and no dividend claim. The same structural critique I have applied to DAO governance tokens โ that they are non-dividend stock whose only holder return is the next buyer's entry price โ applies here in full force. The AI miner token is, ultimately, a non-dividend share of an operation whose free cash flow has been contracted to someone else.
The data problem runs parallel. I have long argued that dedicated data availability layers are overbuilt for the actual market they serve; 99% of rollups do not generate enough transaction data to justify a dedicated DA market. They settle finality on Ethereum or a general-purpose chain, and the DA token's value accrues only if a handful of high-throughput applications materialize. The physical infrastructure cycle is now being mirrored in the digital world: capital is being raised for capacity that demand has not yet justified, and the token is the financing vehicle. Capacity without demand is warehousing. Warehousing is a cost, not a revenue.
DePIN is the attention economy of the capital cycle.
The sector that has attached itself most aggressively to the capital cycle narrative is decentralized physical infrastructure networks. The sales logic is seductive: if Goldman is correct that trillions will flow into physical infrastructure, then a tokenized version of physical infrastructure must capture some fraction of those flows. This is the rug pull in its purest form โ not because the founders are dishonest, but because the incentive alignment is inverted. The token price is the product. The physical infrastructure is the marketing budget.
Consider the structural condition of the cycle. In a capital-intensive environment, the marginal cost of capital rises. DePIN projects are capital-constrained by definition; they ask retail token holders to fund physical hardware โ routers, sensors, storage drives, bandwidth nodes โ in exchange for inflationary token emissions. When a competing project offers the same hardware acquisition with a higher emission schedule, the hardware migrates. The network acquires no economic moat because the underlying infrastructure is commoditized. This inverts the edge of the capital cycle's primary beneficiaries. Hyperscalers win because they control scale and capital costs. DePIN projects lose because their token is a lottery ticket on future adoption, and the capital cycle narrative is simply the lottery's advertising budget.
The information asymmetry is the core mechanic, and it cuts entirely in the direction of the developers. Protocol teams know hardware utilization data before anyone else. They know that honest utilization rates for most deployed physical networks rest in the single digits, expressed as a percentage of stated capacity. The market sees the token's market cap, the dashboard estimating "nodes deployed," and the partnership announcements. This is precisely the setup I quantified during DeFi Summer in 2020, when I tracked over 50,000 on-chain transactions across Compound and Aave pools to demonstrate that leveraged farming produced net-negative returns after gas fees and token depreciation. The framework holds: emissions are inflation, hardware purchases are capital, and utilization is the only revenue. When utilization cannot be verified on-chain โ and in most DePIN networks, utilization data flows through the project's own oracle โ the token is a confidence instrument. It is the same as lending against unaudited collateral in a yield farm, and it ends the same way.
RWA securitization is the finance sector's growth โ and the newest counterparty market.
Goldman identifies finance as a growth sector in this cycle, and this is the line where crypto has its most consequential intersection. The tokenization of infrastructure debt โ the securitization of the very AI data center loans being underwritten today โ is migrating on-chain through real-world asset protocols, mostly below the daily commentary threshold. This is the genuine meeting point: not "crypto as an infrastructure asset," but "infrastructure debt needing a new distribution channel."
The uncomfortable angle is that the debt being tokenized is the debt the traditional finance sector cannot fully absorb on its own balance sheet. Tokenization creates a new bid from crypto-native liquidity. That appears as an inflow to the ecosystem. Structurally, it is the opposite. When a consortium led by a traditional bank tokenizes a construction loan for an AI data center and offers the tranche to DeFi, they are not bringing infrastructure to crypto. They are selling the riskiest, longest-duration, least-liquid paper of the most leveraged sector of the global economy to the buyer class with the least capacity for counterparty stress.
When I stress-tested lending protocols after the Terra collapse in 2022 โ and later moved 60% of my fund's assets into stablecoins and shorted over-leveraged lenders like Celsius ahead of the FTX failure โ the analytical discipline was the same as now: trace the asset to its ultimate borrower, then ask whether the borrower survives a 200-basis-point repricing. Applied to tokenized infrastructure debt, the end borrower is a hyperscaler or an independent power producer servicing a fixed-rate construction loan. The collateral is a power plant or a data center under construction. The token holder sits at the bottom of the capital structure with the worst information. The senior secured bank debt controls the physical asset. The token holder controls a smart contract whose proceeds flow only after the entire waterfall is exhausted. Asymmetric information is the defining feature of the position, and asymmetric information is the classic precondition for underpriced risk.
Historical precedent is not a metaphor; it is a balance sheet. In the telecom fiber buildout of 1998โ2001, capital markets financed precisely this kind of cycle with precisely this consensus: unprecedented demand for bandwidth justified unprecedented capital intensity. At the peak, the fiber network was priced as a perpetuity. Hundreds of billions of dollars of fiber-optic capacity eventually returned near zero on invested capital. The physical assets survived and became commercially viable at a fraction of the build cost, but only after the capital owners were wiped out. The surviving internet giants acquired the infrastructure through bankruptcy auctions, not through dividend payments.
The current buildout has the same structure. The hyperscaler does not need the return on a data center to be positive within ten years. It needs AI to consolidate control over its market. Capital is being deployed for strategic dominance, not marginal return. The marginal buyer of the debt โ including the token holder in the RWA tranche โ is accepting the assumption that the hyperscaler's strategic control is equivalent to their own yield. That is a fragile liability all the way down the chain. When the refinancing cycle begins, and the lenders of last resort are already underwater on construction loans, the queue of token holders will be long and the legal clarity will be short. An exogenous shock to the cost of capital โ a spike in the ten-year yield, a credit event in the power sector โ would reset the entire markdown of the RWA book overnight.
The position I will defend with capital, not commentary, is that the crypto asset class's most rational role in this capital-intensive cycle is to remain the hedge against it, not to join it. Every historical cycle of "unprecedented capital intensity" has concluded with the destruction of the capital and the emergence of a smaller, more efficient set of survivors from the wreckage. The winners were not the infrastructure owners. They were the holders of the scarce resource the infrastructure builders needed to refinance: liquidity.
The most capital-hungry cycle in history is also the most leverage-heavy cycle since 2007, with the most fragile counterparty chain ever assembled โ token holders, RWA protocols, underwriters, construction lenders, hyperscalers. When the cycle rolls over, the entities that survive will be the ones that decoupled from the infrastructure debt narrative early. Bitcoin's role thereby becomes a structural position rather than an investment narrative: an energy-backed monetary asset with no issuer risk is the precise inverse of a long-duration capital asset with no cash flow. As the buildout consumes the global debt capacity, the scarcity of honest collateral becomes the most valuable property in the world. The end of every leverage cycle is the discovery that the asset that looked like infrastructure was actually just leverage โ and the rug pull is executed by the system's own refinancing needs.
Goldman's direction is accurate. The capital is real, the turbines will spin, and the data centers will hum. The asymmetry between the narrative and the return profile is the largest I have observed in 19 years of market observation. In this sideways market, my positioning is precise: hold liquidity, refuse the RWA infrastructure-debt tranche at the margin, and wait for the refinancing event that follows every expansion of this kind. In crypto, the last investor to understand that the cycle was a rug pull is the one who bought the lease on someone else's exit.
