Tech Is Cutting Workers to Buy GPUs: How the AI Restructuring Creates Winners and Losers

Something unusual is happening in Silicon Valley right now. Companies are laying off workers at an accelerating pace, but they're not in trouble. They're redirecting those salary dollars into artificial intelligence spending. Prediction markets are telling us exactly where the money is flowing, and the picture is more nuanced than a simple "tech is doomed" narrative.

Let's start with the numbers that paint the picture.

Betting markets put an 86% probability that tech layoffs in 2026 will exceed 2025 levels. At the same time, those same markets say there's a 59% chance that Anthropic (the company behind Claude) will have the best AI model by December 2026. Google comes in second at 22%, OpenAI at just 10%, and Elon Musk's xAI at 9%. Meanwhile, the probability of a SpaceX IPO by June has risen to 17% (up 2.5% in the last 24 hours), and the chance of the Nasdaq-100 falling below 19,000 by year-end sits at 20%, though that number dropped 4.7% in a single day, suggesting traders are becoming less pessimistic about a tech crash.

Put these signals together and a clear story emerges: tech isn't collapsing. It's restructuring. Companies are swapping headcount for compute power, and the value is concentrating around whoever wins the AI race and, more importantly, whoever builds the infrastructure that all AI companies need.

Think of it like the California Gold Rush. Some prospectors struck gold, most didn't. But the people who sold shovels, picks, and blue jeans made money no matter what. The same logic applies here.

The Productivity Cycle in Action

Ray Dalio talks about productivity cycles where labor gets displaced by technology. What we're seeing now fits that framework almost perfectly. The self-reinforcing loop works like this:

1. Companies recognize that AI can automate tasks previously done by employees.
2. They lay off workers and redirect those salary savings into AI spending (cloud compute, GPU clusters, enterprise AI platforms).
3. That AI spending flows to infrastructure providers: chip makers, data center builders, networking companies, and power generators.
4. The companies that automate fastest gain a productivity advantage, forcing competitors to do the same.
5. This accelerates more layoffs and more AI spending, feeding step one again.

This cycle explains why layoffs and AI investment are rising simultaneously. It's not contradictory. It's the same phenomenon viewed from two angles.

The AI Race: Anthropic's Surprising Lead

The prediction market probabilities for best AI model by December 2026 are striking. Anthropic at 59% is a commanding lead over Google at 22%, and the gap between prediction market sentiment and stock market pricing creates actionable information.

AMZN is the clearest beneficiary of an Anthropic-led future. Amazon has invested over $4 billion in Anthropic and AWS is the cloud backbone running Anthropic's models. If Anthropic wins the AI race, Amazon captures value through both its equity stake and cloud compute revenue. Amazon is also automating its own warehouses aggressively, making it both a gold miner (through the Anthropic bet) and a shovel seller (through AWS compute for everyone else's AI workloads). Confidence: 72%.

GOOGL presents a trickier picture. At 22% for best AI, Google is still the second most likely winner, and its Cloud division benefits from AI spending regardless of whether its own Gemini model leads the pack. Google has genuine advantages in TPU custom silicon, DeepMind talent, and massive proprietary data. But the equity market still prices Google as if it's competing at parity in AI, while prediction markets are more skeptical. That mismatch means there's potential downside risk if Google's AI efforts disappoint. The Department of Justice antitrust case around search adds another layer of uncertainty. Confidence: 58%.

PLTR occupies an interesting position as the enterprise AI deployment layer. Palantir's AIP platform is model-agnostic, meaning it wins regardless of whether Anthropic, Google, or OpenAI leads. The 86% probability of accelerating layoffs is directly bullish for platforms that help companies replace human workflows with AI. The problem is valuation: Palantir trades at over 100 times forward earnings, which prices in an enormous amount of the success scenario already. Strong thematic fit, stretched price tag. Confidence: 52%.

The Shovels: Infrastructure Plays That Win Regardless

This is where the analysis gets most interesting for anyone building a portfolio around the AI theme. Instead of betting on which AI company wins, you can invest in the companies that supply all of them.

NVDA is the quintessential shovel seller. Every AI lab on the planet, whether it's Anthropic, Google, OpenAI, or xAI, trains models on NVIDIA's H100, H200, and Blackwell GPU clusters. The CUDA software ecosystem creates switching costs that AMD and Intel haven't been able to overcome at scale. Data center revenue now exceeds 80% of NVIDIA's total. The layoff-to-capex rotation directly translates into GPU purchase orders. The risk, of course, is that at 35 times forward earnings, the market already knows this trade. Any slowdown in hyperscaler spending would hit the stock hard. Confidence: 78%. Infrastructure relevance score: 92 out of 100.

ASML sits two levels upstream. This Dutch company is the sole manufacturer of EUV lithography machines, the equipment used to manufacture every advanced AI chip in existence. There is literally no competitor on Earth that makes these machines. Every NVIDIA GPU, every AMD chip, every Apple silicon processor requires ASML equipment to produce. This is the purest shovel seller in the entire AI supply chain. Geopolitical risk around Dutch export controls to China is real, and semiconductor capex is cyclical, but the monopoly position is extraordinary. Confidence: 72%. Infrastructure relevance: 88.

ANET builds the networking fabric inside AI data centers. Think of it this way: NVIDIA makes the brains, but Arista makes the nervous system that connects them. AI training clusters require ultra-low-latency, high-bandwidth connections between thousands of GPUs, and Arista's 400G/800G switches and EOS software are the standard in hyperscale environments. Microsoft and Meta are top customers, both deep in AI infrastructure buildout. Confidence: 74%. Infrastructure relevance: 82.

VRT handles the cooling and power management that every GPU cluster needs. As anyone who has felt a laptop get hot under load knows, computing generates enormous heat. At data center scale, thermal management is a genuine engineering challenge, and Vertiv is one of few pure-play providers. The stock has already tripled in 2024, which limits the remaining upside, but the demand tailwind is strong. Confidence: 70%. Infrastructure relevance: 82.

For the power generation layer, two names stand out. VST (Vistra) is a merchant power generator with nuclear and natural gas assets concentrated in Texas, where AI data center buildout is booming. Every compute cycle burns watts, and energy is the ultimate upstream input regardless of which AI company wins. Confidence: 71%. CEG (Constellation Energy) is the nation's largest nuclear fleet operator, and Microsoft already signed a 20-year power purchase agreement with Constellation to restart Three Mile Island specifically to power AI data centers. That's a direct, contracted, publicly disclosed link between nuclear power and AI spending. Confidence: 70%.

EQIX serves as the real estate layer underneath the AI stack. Equinix operates the world's largest network of data center colocation facilities, and landlords always collect rent. The thousands of mid-tier AI companies that can't afford to build their own data centers lease space from Equinix. The REIT structure provides income while the thesis plays out, though it also makes the stock sensitive to interest rate movements. Confidence: 68%. Infrastructure relevance: 74.

ETN (Eaton) makes the electrical plumbing, including transformers, UPS systems, switchgear, and power distribution units for data centers. Every new AI facility needs massive electrical infrastructure, and Eaton is one of few companies that can deliver it at scale. The AI theme is somewhat diluted because Eaton also serves aerospace, automotive, and general industrial markets. Confidence: 62%.

SMCI (Super Micro Computer) builds AI-optimized server systems that house NVIDIA GPUs with liquid cooling. The thematic alignment is strong, but this comes with a critical caveat: SMCI has faced serious accounting issues, including delayed SEC filings, an auditor resignation, and concerns about a DOJ investigation. This is a high-risk, speculative inclusion at best. Confidence: 52%.

The SpaceX Signal

The rising probability of a SpaceX IPO by June (17% and climbing) adds another data point. When major private companies start exploring public listings, it often signals they see a window that might be closing. SpaceX going public could suggest insiders believe current market conditions, while still favorable, may not last. It's not a definitive bearish signal, but it's worth noting as context.

The Risks You Need to Know

Every thesis has vulnerabilities, and being honest about them is more valuable than ignoring them.

The biggest risk across all these names is that AI spending turns out to be a bubble. If hyperscalers collectively decide they've overbuilt, every infrastructure play from NVIDIA to Vertiv to Constellation gets hit. GPU purchase orders get delayed. Data center construction slows. Power demand flatlines.

For AMZN specifically, Amazon's Anthropic stake is still a minority investment with limited governance control. Anthropic could pivot away from AWS exclusivity. Antitrust scrutiny on Amazon's cloud and AI vertical integration is intensifying.

For GOOGL, the 22% AI leadership probability versus a $2 trillion equity valuation suggests the stock market is significantly more optimistic than prediction markets. That gap is a risk. Add the DOJ antitrust ruling that could force structural changes to Google's search monopoly, and the Gemini brand damage from high-profile AI errors, and you get a stock with more downside risk than the headline narrative suggests.

For NVDA, AMD's MI300X is gaining traction, export controls to China reduce the addressable market, and hyperscalers like Google (TPUs) and Amazon (Trainium) are building custom chips that could eventually reduce NVIDIA dependence.

For the power plays (VST and CEG), merchant power pricing is inherently volatile. A mild winter, a shift in energy policy, or faster-than-expected renewable scaling could all compress margins.

And prediction market probabilities themselves are volatile. The numbers quoted here could shift meaningfully in a matter of days. They reflect crowd expectations, not certainties.

Why This Matters for Your Money

If you have a 401(k) or any exposure to technology stocks, this pattern matters. The old playbook of buying a broad tech index fund and assuming everything rises together may not work as well in a restructuring environment. Value is concentrating, not spreading evenly. The companies that build AI infrastructure are capturing an increasing share of tech spending, while companies that can't compete in AI face margin pressure and workforce cuts.

This doesn't mean you should panic and sell your index funds. The Nasdaq-100 has only a 20% chance of falling below 19,000, and that probability is declining. But it does mean that understanding where the dollars are flowing, from salary budgets into GPU clusters, from traditional software into AI platforms, from coal plants into nuclear power purchase agreements, gives you a meaningful edge in thinking about which parts of the tech sector are likely to outperform.

The Gold Rush analogy holds: most prospectors went bust, but the people who sold shovels built lasting fortunes. In the AI race, the shovels are chips, networking switches, cooling systems, electricity, and data center space. And those companies get paid regardless of whether Anthropic, Google, or OpenAI finds the gold.

Analysis based on prediction market data as of April 1, 2026. This is not investment advice.