The semiconductor industry is the unspoken foundation of every blockchain. From the ASICs that secure Bitcoin to the GPUs that power Ethereum’s transitions and the HBM stacks that feed AI training, the physical hardware beneath our digital consensus is often treated as a black box. But when Goldman Sachs raised its semiconductor equipment cycle forecast in August 2025, projecting wafer fabrication equipment (WFE) spending to surge from $1,100 billion in 2025 to $2,810 billion by 2028, they did more than revise a financial model—they exposed the fragile, centralized supply chain that will determine whether decentralized AI can scale. Trace the code back to the conscience behind it, and you find atoms, not bits.
For years, I have argued that blockchain’s promise of trustless coordination depends on open, auditable infrastructure. But the hardware that runs our nodes, validates our transactions, and trains our models is anything but open. The top five semiconductor equipment suppliers—ASML, Applied Materials, Lam Research, Tokyo Electron, and KLA—control over 80% of the market. Their extreme ultraviolet (EUV) lithography machines, priced at over $300 million each, are built by a single Dutch company. This is not the decentralized world we envisioned. It is a monoculture, and Goldman’s forecast confirms that the monoculture is about to get worse before it gets better.
Context: The Hardware That Powers the Blockchain
To understand the stakes, we must first map the semiconductor supply chain to the blockchain stack. The most obvious connection is cryptocurrency mining. Bitcoin’s ASICs rely on advanced logic processes (7nm and below), which require the same EUV tools that are now in short supply. Ethereum’s shift to proof-of-stake reduced that dependency, but the rise of AI-driven blockchains—like those integrating decentralized inference for on-chain agents—has created a new demand for high-bandwidth memory (HBM) and cutting-edge GPUs. HBM, the memory technology that stacks DRAM dies vertically, is the backbone of AI accelerators. It is also the core driver of Goldman’s revised forecast. According to the analysis, HBM production consumes three to four times more wafer capacity per unit than standard DDR5, and the three memory giants (SK Hynix, Samsung, Micron) are racing to expand capacity. The forecast implies that HBM supply will remain tight through 2028, which means the GPUs that power decentralized AI networks will be expensive and scarce.
But the bottleneck goes deeper. The WFE growth is not just about HBM. It is about advanced logic foundry capacity for 3nm and 2nm nodes, which are critical for next-generation ASICs and AI processors. TSMC alone is investing hundreds of billions of dollars to build out its 2nm fabs, and the equipment needed for these nodes—high-NA EUV machines, advanced deposition and etch tools—is produced by a handful of suppliers. Every new fab requires lead times of 12 to 24 months, and the delivery cycles for EUV tools are already stretching to 18 months or more. For a blockchain ecosystem that prides itself on speed and agility, this hardware lag is a silent killer.
Core: The Technical Analysis of the Equipment Cycle and Its Blockchain Implications
Goldman’s forecast breaks down WFE spending by segment: memory (DRAM and HBM) and logic (foundry). The memory segment is expected to grow fastest, driven by AI demand for HBM. The logic segment is driven by foundry expansion for 3nm and 2nm. Let us examine each through a blockchain lens.
First, memory. The HBM market is dominated by SK Hynix, which holds over 50% market share. The transition from HBM3E to HBM4 (expected in 2025-2026) will increase stack layers from 8-12 to 16, further straining wafer capacity. Every blockchain application that relies on AI inference—whether it is a decentralized autonomous organization (DAO) using on-chain oracles or a DeFi protocol that incorporates machine learning for risk assessment—will depend on HBM availability. The forecast implicitly assumes that HBM technology iterations will proceed on schedule, but any delay in HBM4 ramp-up could create a ripple effect, driving up costs for AI-capable GPUs and limiting the growth of decentralized AI platforms. Based on my experience auditing ERC-20 standards in 2017, I saw how a single technical flaw could cascade into systemic risk. The same is true here: a delay in HBM yield improvements could make the entire decentralized AI narrative unaffordable.
Second, logic. The advanced foundry capacity expansion is concentrated in Taiwan and South Korea, with TSMC and Samsung leading the charge. The geopolitical risk is obvious: a disruption in the Taiwan Strait could cut off the supply of the world’s most advanced chips. But even without a conflict, the concentration of manufacturing creates a single point of failure. Blockchain projects that aim to build decentralized physical infrastructure networks (DePIN)—such as distributed computing marketplaces or edge AI nodes—are already feeling the pinch. The VCs that fund these projects often ignore the hardware reality, assuming that Moore’s Law will continue indefinitely. Goldman’s forecast suggests otherwise: the equipment spending required to maintain Moore’s Law is exponential, and the barriers to entry are rising. The next generation of high-NA EUV machines cost over $400 million each, and only ASML makes them. This is not a free market; it is a monopoly.
The forecast also reveals a hidden assumption: that the semiconductor supply chain is resilient enough to handle the demand. But the equipment supply chain itself is fragile. ASML’s EUV production capacity is roughly 50-60 units per year. If demand exceeds that, lead times stretch, and projects are delayed. For blockchain miners, this means that new ASIC designs may take longer to reach production, reducing the rate of hashrate growth and potentially increasing centralization among those who can secure early supply. For decentralized AI, it means that the cost of training a model on a decentralized network could remain prohibitively high compared to centralized cloud providers like AWS, who have long-term contracts with GPU suppliers. The promise of democratized AI access remains unfulfilled if the hardware is locked behind a oligopoly.
Contrarian: The Blind Spots in Goldman’s Optimism
Goldman’s forecast is bullish, but it has blind spots that a blockchain-centric analysis reveals. First, the forecast assumes that AI demand will remain robust through 2028. However, the AI industry is currently in a hype cycle that may peak before then. The forecast’s own trajectory shows WFE growth peaking at 45% in 2027 before decelerating to 29% in 2028, suggesting that Goldman expects the AI investment wave to mature around 2027-2028. If the AI bubble bursts earlier—perhaps due to disappointing returns on large language models or regulatory crackdowns—the demand for HBM and advanced logic could collapse, leaving memory and foundry overcapacity. In that scenario, blockchain projects that have built their roadmaps around cheap AI hardware would be left with expensive, stranded assets.
Second, the forecast underestimates the impact of Chinese equipment localization. China is investing heavily in domestic semiconductor equipment, and export controls are accelerating that process. While Chinese equipment is still years behind in advanced nodes, the maturation of 28nm and above fabs could reduce the global WFE spending growth forecast. For blockchain, this is a double-edged sword: Chinese manufacturers could produce cheaper ASICs and GPUs for mining and AI, but those chips may be subject to export controls, fragmenting the global market. The result could be a bifurcated blockchain ecosystem—one with access to cutting-edge hardware and one without. Education is the only true decentralized currency, but hardware access is the gatekeeper.
Third, the forecast does not account for the potential of open-source hardware initiatives. Projects like RISC-V, which offer an open instruction set architecture, could reduce dependence on proprietary chip designs. However, RISC-V still requires a foundry to manufacture the chips, and the foundry models are still dominated by TSMC and Samsung. The open-source hardware movement is promising, but it will take years to build a fabrication ecosystem that is truly decentralized. In the meantime, every blockchain project that relies on advanced chips is betting on a centralized supply chain.
Takeaway: Building Bridges, Not Just Blocks
Goldman Sachs’ forecast is a wake-up call for the blockchain community. We have spent years building decentralized protocols on top of centralized hardware. The forecast shows that the hardware bottleneck is only going to tighten. To truly achieve sovereignty, we must invest in decentralized manufacturing, open-source chip designs, and community-owned computational resources. The narrative that blockchain is just about software is a lie. Every line of code is a hand extended in trust, but that hand is holding a microchip made by a single company in a single country.
We need to fund projects that explore alternative materials, foundry cooperatives, and on-chain governance of hardware production. The DePIN movement is a start, but it must go beyond coordinating compute resources to coordinating the production of those resources. Perhaps the most radical step is to tokenize semiconductor fabrication—creating a DAO that owns a stake in a fab, or a token that represents a claim on future wafer output. This is not science fiction; it is the logical extension of the blockchain ethos. We build bridges, not just blocks, between people. The next bridge must span the gap between code and silicon.
The silicon ceiling is real. Goldman Sachs has shown us the numbers. Now it is up to us to build the infrastructure that breaks through it.