Goldman Sachs just dropped a number that should make every crypto miner, ASIC designer, and DePIN enthusiast sit up: global wafer fab equipment (WFE) spending is projected to hit $281 billion by 2028. That's a 37% CAGR from 2025, driven by AI demand for advanced logic, DRAM, and HBM. But here's the signal in the noise: this super-cycle isn't just about chips for chatbots. It's about the hardware that will underpin the next phase of crypto mining, proof-of-stake infrastructure, and decentralized compute networks.
Context: The Hardware War
Crypto's relationship with silicon has always been adversarial. Bitcoin mining consumes ASICs that are designed for a single algorithm. Ethereum's shift to proof-of-stake killed the GPU mining boom overnight. But the underlying tension remains: the semiconductor supply chain is the bottleneck for every blockchain that relies on physical compute. Goldman's forecast confirms that the gear—the EUV lithography machines, the HBM stacks, the advanced packaging—is being prioritized for AI, not for crypto. The question is whether crypto can piggyback on this wave or gets squeezed out.
Look at the numbers. Goldman's report, sourced from a deep dive into the semiconductor equipment cycle, predicts WFE climbing from ~$150 billion in 2026 to $218 billion in 2027 and $281 billion in 2028. The primary drivers: HBM (high-bandwidth memory) for AI accelerators, advanced logic (3nm/2nm GAA) for GPUs and custom ASICs, and DRAM for servers. Crypto mining, traditionally a buyer of trailing-edge chips, now faces a market where the cutting edge is consumed by hyperscalers. But that's not the whole story.
Core: The Narrative Mechanism
Let's decode the mechanics. Goldman's optimism hinges on three assumptions: AI capex stays elevated, export controls don't tighten further, and equipment supply chains can deliver. For crypto, each assumption has a dual edge.
First, AI demand. The report notes that AI training chips (NVIDIA H100/B200) consume massive wafer area—single GPU dies over 800mm², eating 2-3 wafers each after yield loss. This pushes foundries like TSMC to prioritize 5nm and below for AI, leaving less capacity for legacy nodes that produce mining ASICs. But the counterpoint: AI inference chips, which are more numerous and cheaper, are expected to overtake training by 2026. These inference chips often use older nodes (7nm, 12nm) that overlap with ASIC production. If the AI inference boom materializes, it could actually increase availability of mature process capacity for crypto hardware.
Second, HBM. The report highlights HBM as a standalone growth driver, separate from logic. HBM3E and HBM4 require advanced packaging (TSV, CoWoS) and specialized equipment for stacking. This is a direct boost for memory manufacturers like SK Hynix and Samsung. But for crypto, HBM is also crucial for high-performance mining rigs that need fast memory bandwidth. The equipment needed for HBM—TSV etchers, bonders, temporary bonders—is distinct from logic equipment, meaning it adds a second engine to WFE spending. This could alleviate some supply pressure on logic equipment, indirectly benefiting crypto.
Third, the geopolitical angle. The report's hidden assumption (confidence 6/10) is that export controls won't spiral. If they do, China's domestic chip production, including for mining ASICs, will be severely constrained. China currently manufactures a significant portion of Bitcoin mining hardware (Bitmain, MicroBT, Canaan). The report notes that Chinese equipment self-sufficiency is only ~20-25%, and advanced node (≤7nm) equipment is effectively blocked. This means the next generation of mining ASICs, which require 7nm or 5nm for efficiency, may be harder to produce in China. The implication: mining hardware innovation could shift to foundries in Taiwan, South Korea, or the US, but at higher cost and longer lead times.
Contrarian: The Blind Spot
Here's the contrarian angle that most crypto analysts miss: the equipment super-cycle is not a threat to crypto mining; it's a catalyst for consolidation and vertical integration. The narrative that "chip shortage kills mining" is a meme. The reality is that the WFE boom is creating a two-tier market. Top-tier miners (Marathon, Riot, Bitfarms) with long-term contracts and access to foundry capacity will get the best ASICs. Second-tier miners relying on spot markets will face delayed deliveries and higher prices. The result is a shakeout that favors scale—exactly what we saw after the 2022 bear market.
But the real blind spot is that the equipment cycle is also enabling new crypto-hardware paradigms. The report mentions that CoWoS (chip-on-wafer-on-substrate) capacity is doubling every year, from 40k wafers/month in 2024 to 120k+ by 2026. This technology is essential for chiplets—a design approach where multiple smaller dies are integrated into one package. For crypto, chiplets could allow specialized mining ASICs to be combined with general-purpose compute, blurring the line between mining and AI inference. Imagine a chip that can both mine Bitcoin and serve lightweight AI models—that's the future CoWoS enables.
Follow the protocol, not the influencer. The protocol here is the semiconductor roadmap. The report states that by 2028, the number of new fabs equivalent to the WFE spending would be 14-19 large-scale facilities, each costing $15-20 billion. That's a massive addition to global manufacturing capacity. If historical trends hold, a portion of that capacity will eventually be repurposed for crypto hardware as the AI boom plateaus. The equipment has a multi-year lifecycle; after the initial AI demand is satisfied, foundries will seek new customers. Crypto miners, with their predictable demand for hashing power, could become the steady-state buyers.
Takeaway: The Next Narrative
The next narrative for crypto hardware is not about scarcity but about sovereignty. The miners that survive the next cycle will be those that control their supply chain—either through direct investment in foundry capacity or through partnerships with equipment makers. The Goldman report implicitly validates this: the equipment market is becoming a seller's market, and those who can secure long-term supply will have a structural advantage.
History repeats, but the code evolves. The 2017 mining boom was about buying GPUs off the shelf. The 2025-2028 cycle will be about negotiating with ASML and Applied Materials. The signal is clear: the silicon ceiling is rising, but it's also hardening. The question is not whether crypto can afford the chips, but whether it can afford to wait.