If Kevin O'Leary had to pick only one investment theme for the next three years, he said he would choose energy.
Speaking on The Deep End, a podcast under Stock Sharks, the Canadian investor and Shark Tank regular said the center of gravity in AI investing is moving away from trying to identify the strongest model developer and toward the infrastructure needed to keep the industry running. In his view, the question is no longer just who builds the best model. It is also who can supply the power, land, data-center capacity and long-term contracts that large-scale computing requires.
The interview ranged well beyond AI. O'Leary also discussed crypto, public blockchains, quantum computing and the portfolio rules he uses to manage uncertainty when the eventual winners are still unclear.
From model leadership to resource constraints
O'Leary did not dismiss the gains already made in the last AI cycle. He said large technology companies, chip makers and model firms have already delivered strong returns, and investors holding those stocks or related indexes have benefited.
His focus now is different. If no one can say with confidence which model company will dominate in the end, he asked, how can investors still participate in the expansion of AI?
His answer was energy.
He described the idea as a classic picks-and-shovels strategy. In a gold rush, the better trade may be backing the companies that sell tools rather than trying to guess who finds the gold. He sees a similar structure in AI. Model companies need chips, chips need data centers, and data centers need large amounts of reliable electricity. Whether OpenAI, Anthropic or another company gains more share, power demand could keep rising as computing infrastructure expands.
For that reason, O'Leary said his main areas of focus over the next 36 months are energy companies, natural gas generation, turbines, power infrastructure and uranium. What matters to him is not whether those assets carry an AI label, but whether they become necessary inputs for the industry as it scales.
Why Canada and the Nordics stand out
O'Leary repeatedly pointed to Finland, Norway and Alberta in Canada as places with an energy advantage. He said those regions have relatively low-cost electricity, and that some long-term power contracts can come in below 6 cents per kilowatt-hour.
For large data centers that need to run continuously, electricity costs can have a direct effect on project margins. Even if the competitive order among model companies changes, businesses with land, grid access and long-term power contracts may still be able to serve a range of customers.
He cited Bitzero as a concrete example of that thesis. According to O'Leary, the company has generated cash flow from Bitcoin mining, but what he values more is its long-term power contracts and infrastructure footprint in Norway and Finland.
In that framework, mining is only one current business line. The more durable asset may be the company’s ability to secure energy. If cloud-computing or AI firms decide to build data centers in those markets, Bitzero could use its existing resources to take part. The appeal of that model, as O'Leary described it, is that an infrastructure supplier does not need to know in advance which AI model will win.
He did not present access to power as a guarantee of profits. Projects still have to solve for grid connection, equipment procurement, customer contracts and capital spending, and low contract prices still need to cover total construction and operating costs. Still, he argued that energy infrastructure offers another way to gain exposure to AI growth.
From electricity to uranium and SMRs
O'Leary said his energy work does not stop with today’s power supply. He is also looking at the energy mix over the next seven to eight years, with particular interest in small modular reactors, or SMRs.
SMRs are a nuclear technology path built around smaller reactor modules, with some designs aiming to improve flexibility through modular manufacturing and deployment. O'Leary said that if AI data centers keep driving long-term power demand, nuclear energy could become an important source of stable supply, with uranium sitting upstream in that chain.
That is why he has been paying more attention to uranium investments. He also made clear that the thesis still depends on commercialization. Whether SMRs can deliver competitive generation costs on the expected timeline will depend on engineering, regulatory approvals, financing and fuel supply.
The broader logic stays the same: do not rush to predict the final technology winner; study the base resources that large-scale adoption will require.
Canada as a resource allocation trade
Canada is a major part of that view. O'Leary said the country has natural gas, uranium, potash, aluminum and other strategic resources, while also sitting next to the United States, a large energy-consuming market. If North America continues to add computing capacity and power infrastructure, he believes Canada could benefit.
He said his team built roughly a 10% allocation to Canada through the large-cap ETF XIU. The trade did not initially have unanimous support inside his team because trade friction between Canada and the US had increased concern about the Canadian economy.
O'Leary said he saw that pessimism as a contrarian opportunity. In his view, changes in Canada’s policy environment and progress on resource-development projects could lead the market to reassess local asset values.
He added that the position at one point outperformed the S&P 500 by 158 basis points, or 1.58 percentage points. The interview did not provide the full comparison period or pricing basis, so the figure stands only as his description of the trade’s performance.
Whether Canada can outperform over a longer period, he said, will still depend on project execution, commodity prices, trade relations and policy follow-through. For him, the trade is not simply a call on economic recovery. It is a call on the strategic value of energy and natural resources in a world of expanding technology infrastructure.
Crypto valuations need more than market flows
O'Leary applied a similar re-think to crypto. In his telling, blockchain assets that were once valued largely through trading activity and capital inflows will need to show sustainable economic value through real business use.
He said a key assumption in earlier crypto investing was that institutional money entering the market would broaden allocation demand across the sector. What he has come to question is whether that money will spread evenly across digital assets.
Recalling an industry conference with heavy institutional participation, O'Leary said some analysts argued that Bitcoin and Ethereum alone could provide about 97% of the price-volatility exposure to the crypto market they were studying. He presented that as a research view he was relaying, not as a statement about BTC and ETH market share.
The implication is straightforward. If a small number of large assets already satisfy most institutional crypto exposure needs, the case for owning a wide range of smaller tokens weakens, and capital concentration becomes more pronounced.
O'Leary said his portfolio once held 27 crypto-related positions. He later sold some of them, and some of the remaining investments suffered significant losses. The main assets he kept were Bitcoin and Ethereum.
That did not lead him to conclude that every other chain had lost investment value. Instead, he said he started looking for a different test: if broad market appreciation is no longer enough, how does another blockchain prove its economics?
Why he is revisiting Avalanche
His answer may lie in enterprise adoption.
Ethereum has often been treated as one of the strongest candidates to become general-purpose blockchain infrastructure. O'Leary said he has started to question whether every industry will really use the same chain. Finance, logistics, real estate, contract management and sports assets do not all have the same requirements.
Financial institutions may care more about stablecoin payments and asset transfers. Logistics companies may focus on supply-chain management and contract execution. Sports clubs may want blockchain systems for collectibles and digital assets. Different business needs could lead to different technical architectures.
That is one reason he has been studying Avalanche again. He specifically pointed to Avalanche’s ability to support customized blockchain deployments for enterprises, saying that model could fit sports assets, collectibles and other use cases that need separate business networks.
In his view, companies may not need to place every operation on one public network. They may choose different blockchain infrastructure depending on the industry. That leaves two possible paths for the market: a small number of general-purpose networks dominate, or multiple networks serve different industries and connect through technical interfaces.
He did not say which path will win. He did stress that business adoption and token returns are not automatically linked. A company using Avalanche technology does not necessarily mean demand for the AVAX token rises in parallel. Investors still need to examine how network fees are paid, how business revenue is generated and whether that value reaches token holders.
For that reason, O'Leary did not say he was preparing to buy altcoins in size again. He said he is considering smaller positions in public chains after studying specific applications. The focus has shifted from liquidity-driven crypto investing to whether enterprises truly need these networks and whether that demand can produce recurring revenue.
Quantum computing as a long-term crypto risk variable
As he revisits the commercial case for blockchains, O'Leary is also watching a separate risk that could affect the entire digital-asset sector: quantum computing.
In the past, Bitcoin risk was usually framed around price volatility, liquidity, regulation and macro conditions. As quantum technology advances, he said, the future resilience of current cryptographic systems has become another issue worth studying.
O'Leary referred to the point at which quantum computing might break some existing cryptographic algorithms as Q-Day.
He was not saying Bitcoin is under quantum attack today. His point was that expectations about a future threat could affect prices before any technical breakthrough arrives. If institutional investors begin to worry that quantum computers may one day attack current digital-signature systems, that concern alone could reduce willingness to allocate capital.
He speculated that some institutions may avoid Bitcoin over the long term because of quantum risk, though he did not cite public data to support a specific share and presented it as his own view.
He also drew a technical distinction. A quantum threat to some public-key cryptography does not mean every part of Bitcoin’s security fails at the same moment. Networks could also adopt post-quantum cryptography through protocol upgrades.
For O'Leary, that risk also points to a possible investment theme. If financial institutions need to upgrade security systems, companies that can provide quantum-computing technology and quantum-resistant security solutions may see new demand. He said that is why he has started watching the quantum businesses of IBM and Google.
At the same time, he cautioned that research capability, commercial revenue and post-quantum security products are not the same thing. Investors should not treat technical progress as a direct proxy for earnings growth.
Position limits over concentrated bets
Even with positive views on energy, blockchain infrastructure and quantum computing, O'Leary said investors should not concentrate too much capital in a single technology trend. He described diversification as one of his most important disciplines.
His rules are simple: no single stock should exceed 5% of the portfolio, and no single industry should exceed 20%.
He tied that framework to an earlier experience during the internet bubble, when he shorted Yahoo. As the stock kept rising, he faced repeated margin calls. By his recollection, the trade led to a drawdown of roughly 40% to 50% in his net worth. Yahoo later fell, but the experience of carrying losses and posting more margin convinced him to stop shorting stocks, and it changed the way he manages risk.
Today he prefers to build core exposure through index funds and then add to companies he likes, while keeping a cap on any single name.
He used Tesla as an example. O'Leary said he initially rejected Tesla because of its valuation, but his son argued that Tesla was not just a car maker and should also be viewed as a technology company collecting large amounts of vehicle data. He eventually took a position and trimmed it several times as the stock rose so it would not become too large in the portfolio.
That approach may leave some upside on the table, but it also reduces the effect of one stock on total performance. He said that for some individual stock trades, he typically uses about 17% as a minimum return target and considers selling more than half the position once that threshold is reached.
He presented those as his own trading rules, not as a guarantee that fixed profit-taking produces better long-term returns. The larger point is that he does not want the fate of the portfolio to rest on a single prediction.
Private-market investing still comes back to cash flow
O'Leary said he uses operating data to test investment ideas in private companies as well. The two metrics he emphasized most were customer acquisition cost, or CAC, and churn rate. The first shows how much a company must spend to win a customer. The second helps show whether existing customers stay.
If a business has to keep raising marketing spend just to maintain revenue growth while customers continue to leave, headline growth may not be economically attractive.
He also looks at cash flow, debt levels and operating data across multiple quarters. That is one reason he remains constructive on enterprise AI applications. In private companies he has invested in, AI tools are already being used for financial analysis, customer acquisition and day-to-day operations. He said one of AI’s real commercial benefits is the ability to cut operating costs and improve margins.
He added that his own company team chose Anthropic’s product as an internal tool. That experience showed him that model rankings alone do not determine enterprise purchasing decisions, because actual performance in business use can differ.
Whichever supplier a company chooses, he said, the final test is still operational improvement. A technology can be advanced without producing profits for the buyer. That is why, even while staying positive on AI over the long run, he is spending so much attention on energy, infrastructure and cash flow.
The common thread across AI, crypto and quantum
Across AI power infrastructure, enterprise blockchains and quantum security, O'Leary’s argument was less about chasing the hottest narrative and more about asking where value can be captured consistently.
His framework comes down to a single idea: rather than trying to predict the final winner in every technology race, study which companies can keep supplying essential resources, generate commercial revenue and turn growth into cash flow as the industry expands.
For energy companies, that means asking whether low-cost power can support profitable long-term contracts. For blockchains, it means testing whether enterprise adoption becomes real network revenue. For quantum computing, it means separating research progress from actual products. For AI applications, it means checking whether businesses can keep lowering costs and improving margins.
If those conditions do not materialize, long-term industry growth alone may not be enough to justify the prices investors are willing to pay.