The AI Gold Rush Is Splitting Tech in Two. Here's Who Sells the Shovels.

Something unusual is happening in the tech economy right now, and prediction markets are making it very visible. The companies building AI are spending like never before, hiring the best researchers, and racing to dominate what might be the most transformative technology since the internet. At the same time, the broader tech workforce is getting laid off at accelerating rates. These two facts aren't contradictory. They're two sides of the same coin, and understanding that coin is the key to figuring out where to put your money.

Prediction markets, where people bet real money on future outcomes, are painting a strikingly clear picture of how concentrated the AI race has become. Anthropic, the company behind Claude, currently leads the "best AI" race with a 42% probability of being recognized as the top AI by the end of 2026, and that number climbed 2.3% in just the last 24 hours. Google sits at 27%, Elon Musk's xAI at 14%, and OpenAI, the company most people would probably guess is winning, trails at just 11%.

Four companies control roughly 94% of the probability in this market. That kind of concentration matters enormously for investors.

Meanwhile, the same prediction markets show an 85.9% chance that tech layoffs in 2026 will exceed 2025 levels. Companies are shedding workers and plowing the savings into AI infrastructure. Netflix has a 74% probability of raising subscription prices. Elon Musk has a 75% chance of becoming a trillionaire by 2027, driven largely by SpaceX, which carries an 88% probability of going public before 2027 and a 16% chance of IPO-ing before June 2026.

Put it all together and you get what legendary investor Ray Dalio would call a paradigm shift. Wealth and power aren't spreading across the tech industry. They're concentrating into fewer and fewer hands, while the companies doing the concentrating automate away the jobs they used to need humans for.

This creates a bifurcated tech economy: a small number of AI winners and platform monopolists extracting enormous value on one side, and a hollowing-out middle tier on the other. The question for investors is how to position for this split.

The Gold Miners: Direct AI Plays

The most obvious approach is to buy the AI leaders directly. The problem is that Anthropic and xAI are private companies. You can't buy their stock. That narrows the field considerably.

GOOGL is the most investable direct AI leader, sitting at 27% in the prediction market with DeepMind, Gemini, and massive cloud infrastructure behind it. Google can integrate AI across Search, Cloud, YouTube, and its autonomous driving unit Waymo, giving it multiple ways to monetize even if it doesn't "win" the frontier model race outright. The flat 24-hour movement in prediction markets, though, suggests the market has mostly priced in Google's position already. The DOJ antitrust case looms large, and there's a real risk that AI-powered search (think ChatGPT or Claude answering questions directly) cannibalizes Google's core advertising model. Massive capital spending on AI infrastructure could also squeeze profit margins for years.

GOOG, the other share class, gets a weaker buy signal for similar reasons. Google's moat is wider than most people appreciate because it doesn't need to win the "best AI" crown to make enormous amounts of money from AI. It just needs to weave AI into the products that already dominate their categories. But the risks are real: regulatory action could force breakups, and losing the perception battle could accelerate the erosion of Search traffic.

AMZN occupies an interesting dual role. Amazon Web Services is the cloud infrastructure that Anthropic, the current prediction market leader, runs on. Amazon has invested over $8 billion in Anthropic. That means Anthropic's surge to 42% probability directly benefits Amazon's economics. AWS also serves countless other AI companies, making Amazon both a direct player through its Anthropic investment and an infrastructure provider that profits regardless of which model wins. The risk is that Azure and Google Cloud are competitive alternatives, and Amazon's retail business faces pressure from ultra-cheap competitors like Temu and Shein.

TSLA gets only a weak buy signal despite Musk's 75% trillionaire probability. Tesla is the only publicly tradeable piece of the Musk ecosystem, which includes SpaceX and xAI. But this is a high-risk play. You're paying an enormous premium for optionality on things that don't exist yet at scale: robotaxis, the Optimus robot, energy storage. Tesla's core auto business is deteriorating with declining market share and margin compression. Perhaps most importantly, if SpaceX actually does IPO, investors who currently use Tesla as a Musk-ecosystem proxy might sell Tesla to buy SpaceX directly. That rotation risk is real and underappreciated.

The Shovel Sellers: Where the Smart Money Looks

During the California Gold Rush of 1849, most individual miners went broke. The people who got rich were the ones selling picks, shovels, and denim jeans to the miners. The same logic applies to AI.

The AI race has four major contenders spending billions on compute infrastructure. Every single one of them needs the same underlying hardware. That creates a set of companies that win no matter which AI lab comes out on top.

NVDA is the canonical shovel seller. Whether Anthropic, Google, xAI, or OpenAI wins, all of them must buy NVIDIA's GPUs, the specialized chips that power AI training and operation. NVIDIA's software ecosystem, called CUDA, creates a lock-in effect that competitors like AMD haven't been able to replicate. Data center revenue now accounts for roughly 88% of NVIDIA's business. The concentration in the AI race actually helps NVIDIA because fewer, richer winners spend more per company on compute as they engage in an arms race against each other.

The risk is valuation. At over 35 times forward earnings, NVIDIA is priced for perfection. Any stumble in demand, any pause in the AI spending cycle, and the stock could drop sharply. Hyperscalers are also developing their own custom chips (Google's TPUs, Amazon's Trainium, Microsoft's Maia) specifically to reduce their dependency on NVIDIA. Export controls to China have already shrunk the addressable market.

ASML operates two levels upstream. ASML makes the extreme ultraviolet lithography machines that factories like TSMC and Samsung use to manufacture the most advanced chips in the world. There is literally no alternative supplier. Whether NVIDIA, AMD, or custom-designed chips win the chip war, they all need to be manufactured on ASML's machines. This is a monopoly in the truest sense of the word. The risk is geopolitical: export restrictions to China reduce the market, and the semiconductor industry is cyclical.

VRT provides the power management and cooling systems that every data center needs. AI data centers run incredibly hot and consume staggering amounts of electricity. Vertiv's products are the unsexy but essential infrastructure that keeps the lights on and the servers from melting. The company is one of the top three providers in its space alongside Schneider Electric and Eaton.

ETN supplies the electrical infrastructure for data centers: transformers, switchgear, uninterruptible power supply systems, and power distribution equipment. Every new AI data center requires massive electrical buildout regardless of who the tenant is. The tech layoff trend actually benefits Eaton because companies redirecting labor spending toward infrastructure spending is exactly the substitution pattern this company captures. The knock is that Eaton is a diversified conglomerate, so AI exposure is diluted across aerospace, vehicle, and eMobility segments.

ANET makes the high-speed networking switches that connect GPU clusters inside AI data centers. As AI training runs scale from thousands to hundreds of thousands of GPUs, the networking fabric connecting them becomes a critical bottleneck. Meta and Microsoft are major customers. The bifurcated economy pattern, with fewer and larger AI data center operators, actually benefits Arista because bigger customers buy at larger scale. Customer concentration is the primary risk: Meta and Microsoft represent over 40% of revenue.

CEG might be the most overlooked infrastructure play in this entire pattern. Constellation Energy is the largest nuclear power operator in the United States, and nuclear is the only scalable, round-the-clock, carbon-free power source that hyperscalers can contract for AI workloads. Microsoft already signed a 20-year power purchase agreement with Constellation for the Three Mile Island restart. As AI data centers demand more and more electricity, someone has to generate it.

ORCL has emerged as an unexpected beneficiary. Oracle Cloud Infrastructure has reportedly become the preferred training cloud for AI labs that want an alternative to the AWS/Azure/Google oligopoly, including large xAI workloads. With xAI sitting at 14% in the AI race, its infrastructure spending flows disproportionately to Oracle. Oracle's enterprise database dominance also means that every company deploying AI against existing business data likely runs through Oracle's middleware.

EQIX and VST round out the infrastructure tier with weaker buy signals. Equinix is the world's largest data center REIT (a real estate investment trust, meaning it owns and operates data center buildings) with over 270 facilities globally. It provides the neutral ground where cloud providers and AI companies physically interconnect. Vistra is a large competitive power generator with nuclear and natural gas assets, particularly exposed to Texas, where Musk's AI and SpaceX operations are concentrated. Both face more competition in their respective niches than the higher-conviction picks.

Think of the infrastructure thesis as a cascade. Here's how it flows:

1. AI companies raise billions and spend it on compute.
2. That compute requires NVIDIA GPUs, manufactured using ASML machines.
3. Those GPUs sit in data centers built with Eaton electrical infrastructure and Vertiv cooling systems.
4. The data centers are connected internally by Arista networking equipment.
5. The data centers are powered by Constellation and Vistra nuclear and gas plants.
6. The data centers are physically housed in Equinix and similar colocation facilities.
7. The cloud platforms (AWS, Oracle Cloud) provide the software layer on top.

At every single layer, someone is selling shovels. And unlike the gold miners, the shovel sellers don't need to guess which AI model wins.

The Risks Nobody Wants to Talk About

This pattern is not a sure thing. Several risks deserve serious attention.

The biggest is that the entire AI capex cycle could slow down or pause. If AI capability improvement plateaus, or if companies realize the return on their AI investments is taking longer than expected, the spending spree could stop abruptly. Every infrastructure company in this thesis would be hit.

Valuation is stretched across the board. NVIDIA, ASML, Vertiv, and Arista have all seen their stocks run up significantly on the AI narrative. If growth disappoints even slightly, the stocks don't just stall. They fall hard. When you're priced for perfection, "pretty good" counts as a miss.

Regulatory risk is omnipresent. Google faces a DOJ antitrust case that could force structural changes. Amazon's Anthropic investment could draw antitrust scrutiny. Export controls on advanced chips to China keep expanding. Nuclear power faces relicensing uncertainty.

The hyperscaler self-build trend is real. Google, Amazon, and Microsoft are all developing their own custom AI chips to reduce reliance on NVIDIA. They're building their own data centers rather than leasing from Equinix. If this trend accelerates, it could undermine several infrastructure plays simultaneously.

And there's a second-order risk that gets less attention: if tech layoffs truly accelerate (which the 85.9% probability suggests they will), consumer spending in tech-heavy cities like San Francisco, Seattle, and Austin could weaken meaningfully. That has knock-on effects for housing, retail, and local economies that ripple far beyond the tech sector.

Why This Matters for Your Money

You don't need to be a tech investor for this pattern to affect you. If you have a 401(k) or any index fund that tracks the S&P 500, you already own significant exposure to these companies. NVIDIA, Google, Amazon, and Tesla are among the largest weights in virtually every major index.

The concentration dynamic matters for everyone. When wealth and power consolidate into fewer companies and fewer individuals, it changes the economy in ways you can feel. Netflix raising prices (with a 74% probability) is a small example. The broader phenomenon of companies replacing workers with AI while charging consumers more is a pattern that affects grocery bills, subscription costs, and job security.

The shovel-seller thesis isn't just about making money. It's about understanding where value is actually flowing in the economy. Right now, it's flowing toward the physical infrastructure that makes AI possible: chips, power, cooling, networking, and data centers. That flow is likely to continue regardless of which AI lab is leading at any given moment, because the arms race itself is the demand driver, not any single participant's victory.

The bifurcation in tech is real, it's accelerating, and prediction markets are pricing it in with remarkable clarity. The question isn't whether this shift is happening. It's whether your portfolio is positioned for the right side of it.

Analysis based on prediction market data as of March 19, 2026. This is not investment advice.