Goldman Sachs just dropped a $7.5 trillion bombshell on the AI infrastructure space. Over five years, that's the predicted investment into chips, data centers, and power grids to fuel the next wave of artificial intelligence. But before you start cheering for a universal compute boom, ask yourself: where does crypto mining fit into this picture? The ledger doesn't lie, and right now, the math suggests a collision course.
Let's break down the numbers. $7.5 trillion over five years averages $1.5 trillion annually. For context, the entire global semiconductor market is currently worth about $600 billion. This forecast implicitly demands a tripling of chip production capacity, most of it for AI accelerators like NVIDIA's H100 and B200. The energy requirement alone is staggering: if half that money goes to hardware, we're looking at 1,500-2,000 GW of installed AI compute by 2028, consuming maybe 10-15% of global electricity.
But here's where the crypto native should lean in. That same silicon is what powers proof-of-work mining and, increasingly, decentralized GPU networks like Render or Akash. If Goldman's prediction holds, corporate AI spending will crowd out every other demand source for high-end chips. I've seen this play out before: during the 2020 DeFi summer, I audited a yield aggregator whose liquidity was drained by a single exploit. The fundamental flaw wasn't the code—it was the assumption that liquidity would always flow. Similarly, the assumption here is that AI compute demand will remain elastic. It won't.
Between the hype cycle and the blockchain reality, we need to examine the hidden assumptions. Goldman's model assumes that AI models will keep scaling—more parameters, more training data, more inference queries—without a fundamental breakthrough in efficiency. It assumes no regulatory intervention on energy consumption, no geopolitical fragmentation of supply chains, and no Capability Ceiling. More critically for crypto, it assumes that every GPU will be snapped up by hyperscalers and AI startups, leaving little room for mining or decentralized compute.
The core insight is that this forecast is a feature, not a bug, for centralized AI. The very scale of investment needed to hit $7.5 trillion requires dominant players—Microsoft, Google, Amazon, NVIDIA—to consolidate their control over the hardware stack. Whales like MicroStrategy might survive by pivoting to AI infrastructure (as they've hinted), but the median GPU miner will face either extinction or a forced migration to ASICs. Smart contracts don't lie, but they also don't protect you from market forces when your hardware is outbid by a $200 billion cloud contract.
Now the contrarian angle: the real story isn't AI at all—it's the centralization of compute resources. Crypto was supposed to democratize access to capital and computation. Yet here we have a single investment thesis that, if realized, would funnel the world's most advanced chips into a handful of data centers, controlled by entities that have no interest in supporting decentralized networks. Is it innovation, or just a liquidity trap in pixels? The same critique I leveled at NFT art in 2021 applies here: these are social signaling mechanisms dressed as technological progress.
History offers a cautionary tale. The internet infrastructure bubble of the late '90s saw $1.5 trillion in fiber optic investment, much of it left dark for years. AI infrastructure has a shorter depreciation cycle—3-5 years for chips vs. 15-20 years for fiber. If application revenue doesn't materialize to the tune of $2-3 trillion annually by 2028, we'll see a wave of asset write-downs that makes the 2022 crypto winter look like a mild frost. Code is law, but audits are the truth we chase, and the balance sheets here don't add up.
Where does this leave crypto? Watch for three signals. First, NVIDIA's quarterly data center guidance—if it shows deceleration from 200% YoY growth, the AI narrative loses steam. Second, GPU mining profitability—if hashprice drops as miners get squeezed by AI demand, it's time to rotate into ASIC-heavy assets like Bitcoin. Third, DePIN projects that actually deliver compute to retail users—they might become the only escape hatch from centralized AI infrastructure.
The takeaway is not to dismiss the $7.5 trillion figure outright. Rather, treat it as a stress test for crypto's original promise of decentralized computation. The speed of news is fast, but the chain is slower. And right now, the chain is telling us that the biggest AI infrastructure buildout in history may also be the biggest threat to mining and GPU-based crypto networks. Valuing the intangible in a tangible world is hard enough; try doing it when your hardware supplier just sold your next shipment to OpenAI.
