Canadian AI Stocks: Ranked on Measured AI Exposure

Affiliate Disclosure: Bestcanadianstocks.ca may earn a commission when you open an account or make a purchase through links on this page. This comes at no additional cost to you and helps us continue providing free financial content to Canadian investors.

Celestica closed at $483.09 on September 21, 2026, up 1,506% in three years. Thomson Reuters closed at $134.01, down 38% in twelve months. Both appear on essentially every list of Canadian AI stocks published this year, usually within a few lines of each other, usually described in the same admiring language.

That is the problem with the category. “Canadian AI stock” has become a label rather than a measurement, and it is applied to a semiconductor-adjacent hardware manufacturer whose revenue grew 62% last quarter and to a legal-information publisher whose revenue grew 9%, as though the AI boom were doing the same thing to both.

So we measured it. We took every TSX and TSX Venture name commonly presented as a Canadian AI play, regressed each one’s daily return on Nvidia’s over the last 504 trading sessions, and published the R-squared: the share of each stock’s day-to-day movement that the AI trade actually explains. The answer for Celestica is 30.9%. The answer for Thomson Reuters is 0.0%. Between them sit two dozen companies, most of them far closer to zero than they are to Celestica.

This page ranks ten Canadian AI stocks on a criterion the label does not capture: how much of the artificial-intelligence build-out actually reaches the company’s own income statement, corroborated against what the market has independently concluded. Every company figure comes from that company’s own filing, named and linked beside it. Every technical claim is computed from price history and published with the number of observations behind it, so a nine-event median reads as a nine-event median.

Prices and moving averages are the September 21, 2026 close. The AI exposure regression is a fixed 504-session window ended September 11, 2026 and is dated separately below.

What the Measurement Actually Shows

• The AI trade explains 30.9% of Celestica’s daily variance and nothing at all of Thomson Reuters’. Across the twenty-five names we tested, the median R-squared is 5.0%. Most Canadian AI stocks are not, by the market’s own revealed behaviour, AI stocks.
• The three highest readings are not software companies. Celestica builds data-centre hardware, Hut 8 builds and powers data centres, and Brookfield finances and owns the infrastructure underneath them. Compute, power and land are where the AI money is landing in Canada.
• The Canadian software names have been sold, hard, in the year AI was supposed to make them. Constellation Software is down 36% over twelve months, Thomson Reuters 41%, OpenText 32%, Coveo 55%. The market is not pricing these as AI winners. It is pricing several of them as AI victims.
• Constellation publishes the number that debate turns on and almost nobody reads it. Its own quarterly disclosure puts organic growth in maintenance and other recurring revenue, the line AI is supposed to erode, at 4% in the second quarter of 2026 and positive in all nine quarters it discloses.
• The pure-play is gone. TELUS Digital, the closest thing Canada had to a listed AI data-services company, was taken private by TELUS on October 31, 2025 at US$4.50 per share and delisted from the TSX and NYSE. Canadians can no longer buy it.
• Celestica’s own guidance is the loudest AI signal on the TSX. The company raised its 2026 revenue outlook to $20.5 billion from $19.0 billion and its adjusted EPS outlook to $11.30 from $10.15, and told shareholders to expect 2027 growth to accelerate beyond the 65% it expects this year.

The Ten, at a Glance

#	Company	Ticker	AI exposure (R²)	12-month return	What it actually sells into AI
1	Celestica	TSX: CLS	30.9%	+41.2%	Custom servers, storage and 800G/1.6T networking for hyperscalers
2	Hut 8	TSX: HUT	21.0%	+209.3%	Powered land and AI data-centre capacity under long-term lease
3	Brookfield	TSX: BN	20.3%	-16.0%	Capital, power and sites for the build-out, plus the fees on them
4	Shopify	TSX: SHOP	14.9%	-11.1%	Commerce infrastructure that AI agents transact through
5	Capital Power	TSX: CPX	14.5%	+8.8%	Electricity, contracted to a hyperscaler for more than ten years
6	OpenText	TSX: OTEX	5.6%	-31.6%	Enterprise content and its Aviator AI agents, attached to cloud deals
7	Descartes Systems	TSX: DSG	5.0%	-28.8%	Logistics network data, the scarce input AI models need
8	Constellation Software	TSX: CSU	2.3%	-36.1%	Nothing yet, which is the whole argument
9	Kinaxis	TSX: KXS	1.7%	-6.3%	Agentic supply-chain planning, sold as Maestro
10	Thomson Reuters	TSX: TRI	0.0%	-41.4%	CoCounsel and its own legal LLM, into the profession AI threatens most

R-squared is our own regression of each stock’s daily log return on Nvidia’s over the 504 trading sessions ended September 11, 2026. Returns are price returns over twelve months. Neither is a forecast, and a high reading is not a recommendation: three of the four highest readings on the wider list are down over a year.

How To Buy AI Stocks in Canada

Before any of the ranking matters you need an account that can hold these shares, and a clear view of what owning them costs. Most pages on this subject skip this part. It is the part that decides whether you end up owning anything.

Which account to use

Nine of the ten names below are listed on the Toronto Stock Exchange and trade in Canadian dollars. Hut 8 is dual-listed on the TSX and Nasdaq, and a Canadian can simply buy the Toronto line. All ten can be held in any registered account.

• A TFSA shelters capital gains permanently and nothing is taxed on the way out. For a group this volatile that cuts both ways: a loss inside a TFSA also destroys contribution room you never get back. Our TFSA stock rankings approach the same question from the account side.
• An RRSP matters most if you intend to hold US-listed AI names alongside these. Article XXI(2) of the Canada-US tax convention exempts US dividends paid into an RRSP from the 15% withholding tax that a TFSA cannot recover, which is the argument our page on RRSP stocks ranked on tax saved works through in full. For non-dividend-paying AI names the point is close to academic, because there is no dividend to withhold.
• An FHSA can legally hold any of these, and for most people should not. A stock that can fall 40% in a year is a poor match for money with a dated purpose, which is the argument in our FHSA investment strategy page.
• A taxable account is the default once registered room runs out, and for a group that mostly pays no dividend it is less punishing than it sounds: nothing is taxed until you sell.

What it costs

Zero-commission trading on Canadian-listed stocks and ETFs is the baseline at the discount brokers now, so the headline cost of buying anything on this page is nothing. What remains is the bid-ask spread on the thinner names, and currency conversion if you buy a US listing of a company that also trades in Toronto. There is no reason to pay a conversion spread to buy Hut 8 in US dollars when HUT trades on the TSX. If you do end up holding US-listed AI names, Norbert’s gambit is the cheap way to move the money.

We use Questrade for this. Zero commissions on Canadian and US-listed stocks and ETFs, registered accounts with no annual fee, and the dual-currency setup that matters the moment you add Nvidia or Microsoft beside these. Our Questrade review sets out the full fee schedule, and Questrade against Wealthsimple is the comparison most readers are actually running. Wealthsimple is the simpler entry point if you want fewer decisions, and our Wealthsimple review is honest about what that simplicity costs. If you are starting from scratch, the best investing apps in Canada ranks the whole field and the best broker for beginners narrows it to a first account.

What you need to open one

A Social Insurance Number, government photo identification, your employment and income details, and about fifteen minutes. There is no minimum deposit at the major discount brokers. Our walkthrough on how to open a brokerage account in Canada covers what the application asks and why, and how to buy your first stock covers the order screen that comes after.

The four steps

1. Open and fund the account. Electronic funds transfer from your bank clears in one to three business days.
2. Decide which layer of AI you want before you decide a ticker. Compute, power, data or applications. That decision explains far more of your outcome than the choice between two software companies, and it is the argument of the next section.
3. Use a limit order. These names move several per cent on a single headline out of Santa Clara. A market order placed into a gap fills at whatever the gap is paying. How to read a stock quote covers the bid, the ask, and why the difference matters more here than on a bank stock.
4. Size it as the volatile position it is. Celestica has risen 1,368% in three years. It has also fallen more than 20% from a high four separate times inside that run. Both facts describe the same holding.

How We Ranked Them

Three tests, applied in order. AI exposure is the axis here, not growth: our Canadian growth stock rankings rank five of these same companies on a different test entirely, five years of growth per share.

First, does the AI build-out show up in the filings? Not in the press release adjectives, in the reported numbers. Celestica’s second quarter revenue rose 62% because hyperscalers ordered hardware. Hut 8 has signed leases. Capital Power has a signed electricity supply agreement with Meta. Those are AI dollars with a document behind them. A software company that added an AI feature and reported 3% organic growth has an AI product, not AI revenue, and the ranking treats those differently.

Second, does the business survive being wrong about AI? The single most likely way to lose money in this category is to buy a company whose entire valuation assumes a build-out that slows. So the ranking rewards businesses with something underneath the AI story: Brookfield’s fee streams, Capital Power’s contracted generation, Shopify’s merchant base, Descartes’ network.

Third, what has the market already concluded? This is where the regression comes in. It is a corroborating test, not the ranking itself, and in two places it argues against a name we still rank.

Read that chart carefully, because it contains a warning as well as a finding. High exposure has not been the same thing as a good year:

One methodological note, because it matters for how much weight this deserves. An R-squared computed on two years of daily data is a description of a specific regime: the two years in which Nvidia was the market’s AI proxy. If the AI trade broadens or breaks, these relationships change. We publish the sample size and the window for exactly that reason, and we would not rank on this measure alone.

The Latest Reported Quarter, From the Filings

Before the individual cases, the scorecard. Every figure below is the company’s own reported number for its most recently completed quarter, taken from the filing named in each section rather than from a data aggregator.

Company	Quarter	Revenue	Change	Bottom line	The AI-specific disclosure
Celestica	Q2 2026 (Jun 30)	$4.70B USD	+62%	Adj. EPS $2.54 vs $1.39	2026 outlook raised to $20.5B revenue, $11.30 adj. EPS
Hut 8	Q2 2026 (Jun 30)	$74.9M USD	+81%	Net loss $177.1M	949 MW contracted, ~$26.6B base-term contract value
Brookfield	FY2025 (Dec 31)	$75.1B USD	-12.7%	DE before realizations $5,386M	Inaugural AI infrastructure fund launched
Shopify	Q2 2026 (Jun 30)	$3,583M USD	+34%	Operating income $488M vs $291M	GMV $115.6B, free cash flow margin 18%
Capital Power	Q2 2026 (Jun 30)	$740M CAD	+67.8%	AFFO $328M vs $235M	250 MW, 10-year-plus supply agreement with Meta
OpenText	Q4 FY2026 (Jun 30)	$1.349B USD	+2.9%	Adj. EBITDA margin 37.6%	Enterprise cloud bookings $295M, +24.1%
Descartes	Q2 FY2027 (Jul 31)	$201.1M USD	+12%	Gross margin $156.7M vs $138.2M	Services 94% of revenue
Constellation	Q2 2026 (Jun 30)	$3,335M USD	+17%	Organic growth 3%, 1% ex-currency	Maintenance organic growth +4%
Kinaxis	Q2 2026 (Jun 30)	$158.8M USD	+16%	Adj. EBITDA margin 26%	ARR $465.6M, +19%
Thomson Reuters	Q2 2026 (Jun 30)	$1,954M USD	+9%	Diluted EPS $1.02 vs $0.69	“Big 3” organic growth +10%

Brookfield’s line is a fiscal year rather than a quarter because its full-year release is the most recent document we could verify byte for byte from its own site.

The Technical Position of All Ten, in One Place

Each section below carries its own crossover record. This table is the same data side by side, so the sample sizes are visible together rather than one at a time. Every figure is computed from ten years of closing prices to September 21, 2026.

Company	Close	50-day	200-day	Current regime	Completed golden crosses	Median peak gain	Made a new high
Celestica	$483.09	$451.49	$455.67	Death since 2026-09-14	6	+17.8%	1 of 6
Hut 8	$145.10	$129.88	$108.20	Golden since 2025-08-11	7	+104.0%	2 of 7
Brookfield	$53.37	$58.02	$60.57	Death since 2026-07-28	6	+35.1%	5 of 6
Shopify	$193.38	$190.40	$180.09	Golden since 2026-08-28	6	+83.5%	4 of 6
Capital Power	$62.29	$66.37	$64.40	Golden since 2025-06-24	6	+40.3%	5 of 6
OpenText	$32.42	$33.15	$33.88	Death since 2026-02-04	9	+21.2%	3 of 9
Descartes	$111.33	$106.35	$104.10	Golden since 2026-09-03	5	+43.8%	4 of 5
Constellation	$2,786.97	$2,959.14	$2,795.78	Golden since 2026-08-07	4	+56.5%	4 of 4
Kinaxis	$173.90	$168.38	$153.87	Golden since 2026-07-28	8	+19.1%	3 of 8
Thomson Reuters	$134.01	$139.45	$135.34	Golden since 2026-09-10	5	+56.8%	3 of 5

Read the last two columns together. Constellation’s four-for-four record looks like the strongest signal on the page and rests on four observations. Kinaxis has twice as many observations and a median gain a quarter the size. Celestica’s median is still the lowest in the table even after its 4,347% regime ended on September 14, 2026 and joined the completed sample, which lifted that median from 13.9% to 17.8% on a single observation. That is what happens when a business changes completely inside a sample. None of these are base rates. They are records, and short ones.

1. Celestica (TSX: CLS)

The one Canadian company whose income statement is visibly made of AI.

Celestica designs and builds the hardware inside hyperscale data centres: custom compute racks, storage, and the 800G and 1.6T optical networking that moves data between GPUs. It is not an AI company in the sense of owning a model. It is the company that gets paid when someone else decides to build a cluster.

The macro reason. The capital-expenditure cycle at the largest cloud operators is the single largest industrial spending programme currently running, and Celestica’s Connectivity and Cloud Solutions segment sells directly into it. In the second quarter of 2026 that segment grew 84% year over year and accounted for 81% of company revenue, per Celestica’s own second-quarter results release, filed with the SEC as an exhibit to its Form 8-K. Total revenue was $4.70 billion against $2.89 billion a year earlier, an increase of 62%. GAAP earnings per share were $3.17 against $1.82, and adjusted EPS was $2.54 against $1.39. Adjusted operating margin reached 8.2%, which the company described as a new high.

Then the part that matters more than the quarter. Celestica raised its full-year 2026 outlook to $20.5 billion of revenue from $19.0 billion, and its adjusted EPS outlook to $11.30 from $10.15, implying growth of 65% and 87% respectively. It also told shareholders to expect 2027 revenue growth to accelerate beyond that 65% rate, on 800G and 1.6T networking ramps and new custom programmes. Third-quarter guidance is $5.25 billion to $5.55 billion of revenue and $2.88 to $3.08 of adjusted EPS.

Set the two charts beside each other. Revenue roughly doubled between fiscal 2021 and fiscal 2025. Earnings per share went up nearly five times over the reported window, because this is a business with operating leverage that only shows up at scale.

The technical picture, and it has changed. Celestica closed at $483.09 on September 21, 2026 with the 50-day at $451.49 and the 200-day at $455.67, the fast average 0.9% below the slow one. When this page last reported the pair they were within 0.04% of each other and we called it a genuine inflection. It resolved, and not in the direction the price action since would suggest: the 50-day crossed below the 200-day on September 14, 2026, ending the golden-cross regime that had run since December 6, 2022.

The precedent, and the honest version of it. Celestica has had six golden crosses in the last ten years, and the enormous one is now over. The regime that fired on December 6, 2022 peaked 4,347% above its signal price and ended on September 14, 2026 with the stock still 2,895% higher than where the signal fired. That single observation is the entire AI trade in one line, and it is also why the median is nearly useless here: across all six completed golden-cross regimes the median peak gain is 17.8%, the second best is 39.7%, and the stock made a new all-time high in one of six. Six observations, one of which is two orders of magnitude larger than every other. Anyone quoting Celestica’s crossover statistics as a base rate is quoting a sample of six from a business that changed completely in 2023.

What the new signal has done so far. The death cross fired on September 14, 2026 with the stock at $439.70, and the stock has gone up since: $483.09 at the September 21 close, 9.9% above the signal. That is not out of character. Across Celestica’s five completed death-cross regimes the median peak gain from the signal was 20.8%, the best 139.0% and the worst 2.9%, and the stock made a new all-time high in none of the five. In this sample a Celestica death cross has marked a pause rather than a top. Five observations is not a base rate either, and the reason to report the signal is that it ends a four-year trend, not that it predicts the next one.

The risk, stated plainly. Celestica’s revenue is concentrated in a small number of hyperscale customers whose spending decisions are made annually. The 2027 acceleration the company is guiding to depends on programmes that have been won but not yet shipped. A single large customer deferring a build is a double-digit revenue event.

If you want the recent trading context, our coverage of why Celestica jumped 6.64% on a growth reshuffle and the 5.5% drop that preceded it on no news at all both deal with a stock that now moves on sentiment about someone else’s capital budget.

2. Hut 8 (TSX: HUT)

The largest contracted AI data-centre pipeline of any Canadian-listed company, attached to the smallest revenue on this list.

Hut 8 began as a bitcoin miner. It is now, by its own description, an energy infrastructure platform, and the reason it ranks second is that it has done the hardest thing in the AI build-out: secured power and land, then signed investment-grade tenants to long leases on it.

The macro reason. The binding constraint on AI capacity is no longer chips, it is interconnection queues and megawatts. Whoever holds permitted, powered sites holds the scarce asset. Per Hut 8’s second-quarter results, filed with the SEC as an exhibit to its Form 8-K, the company has scaled to approximately 949 MW of contracted AI data-centre capacity with roughly $26.6 billion of expected aggregate base-term contract value, more than $1.75 billion of expected average annual net operating income, leased or backstopped exclusively by investment-grade counterparties. It closed $7.5 billion of fully amortizing investment-grade project financing across two offerings in a single quarter, non-dilutive and without recourse to the parent. At Beacon Point, an existing tenant returned within months of the first lease to commit to a second 352 MW lease, commercializing the campus’s full gigawatt of utility capacity. Facilities representing 1,330 MW are in active construction at River Bend and Beacon Point, with initial data hall delivery targeted for the second and third quarters of 2027.

Now the counterweight, from the same filing. Total revenue in the quarter was $74.9 million against $41.3 million a year earlier. Compute revenue was $72.5 million, Power $1.2 million, Digital Infrastructure $1.3 million. The company reported a net loss of $177.1 million, which included $138.6 million of primarily unrealized losses on digital assets, against net income of $137.5 million a year earlier when those marks ran the other way. Adjusted EBITDA, under a definition the company revised this quarter to exclude digital-asset marks, was $10.4 million. Including those marks it was negative $94.6 million.

So the investment case is entirely forward. Roughly $26.6 billion of contracted value sits against $75 million of quarterly revenue, and the gap closes only when data halls deliver in 2027 and beyond. That is not a criticism, it is the shape of the trade, and it should be sized as such.

The technical picture. Hut 8 closed at $145.10 on September 21, 2026 with the 50-day at $129.88 and the 200-day at $108.20, the fast average 20.0% above the slow one. That is one of the steepest uptrend configurations on the TSX.

The precedent. Hut 8 has had eight golden crosses since 2018. The current one, from August 11, 2025, has produced a peak gain of 549.3% and is up 411.8% as at the September 21, 2026 close. Across the seven completed regimes the median peak gain was 104.0%, the best 586.5% and the worst 12.3%, with a new all-time high reached in two of seven. Seven observations, all inside a business that has changed identity twice, and a stock whose drawdowns inside those winning regimes reached 41.8%. The direction of the record is genuinely favourable. The precision it appears to offer is not there.

The risk. Bitcoin still moves this stock and still moves its reported earnings, as the $138.6 million of unrealized losses shows. Construction risk on 1,330 MW is real. And a counterparty that is investment grade today is signing leases whose economics depend on AI demand still being there in 2029.

3. Brookfield Corporation (TSX: BN)

Owning the build-out rather than betting on it, at a price the market has marked down 17% this year.

The macro reason. Someone has to finance data centres, buy the land, contract the power and take the construction risk, and Brookfield does all four while charging fees on other people’s capital for the privilege. Per Brookfield’s own full-year results release, distributable earnings before realizations were $5,386 million for fiscal 2025 ($2.27 per share) against $4,871 million the prior year, and total distributable earnings were $6,008 million ($2.54 per share). Fee-bearing capital rose 12% to $603 billion, supporting a 22% increase in fee-related earnings to $3.0 billion for the year. In the same release the company flagged the launch of its inaugural AI infrastructure fund.

That is the difference between Brookfield and everything else on this list. The AI exposure is real, the measured relationship to the AI trade is the third highest in the group at 20.3%, and yet the business does not depend on AI: it collects fees on $603 billion whether the next cluster gets built or not.

A note on what is missing. Brookfield’s fiscal 2025 annual filing restates fiscal 2024 diluted earnings per share at $0.20 where the fiscal 2024 filing reported $0.31. Our extraction cross-checks every fiscal year that appears in more than one filing and refuses to stitch a disagreement, so there is no EPS chart here rather than a chart built on a basis we picked by hand.

The technical picture. Brookfield closed at $53.37 on September 21, 2026 with the 50-day at $58.02 and the 200-day at $60.57, the fast average 4.2% below the slow one. It has been in a death cross since July 28, 2026 and is 10.6% lower than the signal price.

The precedent, which does not flatter the chart. Across five completed death-cross regimes Brookfield’s median peak gain from the signal was 13.7%, and it made a new all-time high in zero of five. The golden-cross record is the opposite: six regimes, median peak gain 35.1%, a new high reached in five of six. Both samples are small, and Brookfield’s two recent death crosses in 2025 and early 2026 both produced double-digit rallies before rolling over, so “death cross” here has meant chop rather than collapse. The honest summary is that Brookfield’s crossovers have been better at marking the start of an advance than the start of a decline in this sample, and the current signal is the wrong one.

The risk. Brookfield’s earnings are sensitive to the discount rate applied to very long-duration assets, and the Bank of Canada’s policy rate has been held at 2.25% since October 30, 2025 while US rate expectations have moved the other way. A higher-for-longer US rate is a direct headwind to the value of everything Brookfield owns, and that, rather than anything to do with AI, is the best explanation for the 2026 share price.

If infrastructure and real assets are the part of this that interests you, our Canadian REIT rankings cover the listed property side of the same theme, and Canadian energy stocks cover the generation and pipeline assets that AI demand is now competing for.

4. Shopify (TSX: SHOP)

The Canadian company most likely to be paid by AI agents, and the one whose AI story is hardest to see in the numbers.

The macro reason. If software agents start doing the buying, the transaction still has to land somewhere: an inventory system, a payment rail, a merchant of record. Shopify is the largest such rail outside Amazon, and it has spent 2026 wiring its platform for agent-initiated commerce. That is a genuine structural position rather than a feature announcement.

What the filings show. Second-quarter 2026 revenue was $3,583 million against $2,680 million a year earlier, growth of 34% and 33% in constant currency. Gross merchandise volume was $115,567 million against $87,837 million. Monthly recurring revenue reached $221 million against $185 million. Operating income was $488 million against $291 million, and free cash flow $654 million against $422 million, an 18% margin against 16%. Those figures are Shopify’s own, from the second-quarter release filed with the SEC.

The honest problem with the AI thesis here. None of the figures above are attributable to AI. Shopify does not break out agentic-commerce revenue and there is no reason to think it is material yet. What you are buying is a 34%-growth business at a price that has fallen 17.5% this year, plus a free option on being the rail that agents use. Rank it fourth on the strength of the business, not on the AI story, and the position makes sense either way.

The technical picture. Shopify closed at $193.38 on September 21, 2026 with the 50-day at $190.40 and the 200-day at $180.09. A golden cross fired on August 28, 2026 and the stock is 9.0% lower than the signal price, having been 16% lower in mid-September. The signal fired essentially at the local top, which is an uncomfortable configuration, but the gap has been closing.

The precedent. Six completed golden-cross regimes since 2016, median peak gain 83.5%, best 848.2%, worst 46.1%, and a new all-time high reached in four of six. That is the strongest golden-cross record on this page, and it is six observations from a stock that went up ninefold and then fell 85% inside the sample. Note also what the current regime has already done: every one of those six regimes eventually produced at least a 46% peak gain, and this one is still below its signal price. Either the pattern breaks here or the entry is early.

The risk. Shopify’s valuation has never been forgiving, its growth is levered to consumer spending, and the agentic-commerce thesis is one that Amazon, Stripe and the model developers themselves are all also pursuing. Our coverage of why Shopify fell 4.7% on an oil shock is a reminder that on most days it trades as a high-multiple growth stock rather than as an AI asset, which is exactly what the 14.9% R-squared says.

5. Capital Power (TSX: CPX)

The bottleneck, contracted.

The macro reason. Every megawatt of AI compute needs a megawatt of firm power, and Alberta has the deregulated market, the gas and the land. Capital Power converted that into a document. On July 8, 2026 it announced a long-term energy supply agreement of more than ten years for 250 megawatts of capacity and energy supporting a data centre in Sturgeon County, Alberta being developed by Meta Platforms, with load anticipated in service in the back half of 2028. That agreement began as a binding memorandum with an unnamed investment-grade developer in December 2025 and is now signed with a named counterparty.

What the filings show. Per Capital Power’s 2026 second quarter report, the company generated adjusted funds from operations of $328 million against $235 million a year earlier, adjusted EBITDA of $351 million against $322 million, and net cash flows from operating activities of $214 million against $143 million. It reported a net loss of $43 million against a net loss of $131 million. Revenues and other income were $740 million against $441 million. Generation rose to 10,137 GWh from 9,022 GWh while facility availability fell to 87% from 93% on planned maintenance. For the full 2025 year the company reported AFFO of $1,066 million, adjusted EBITDA of $1,580 million and net income of $159 million, alongside the roughly $3.0 billion (US$2.2 billion) acquisition of the Hummel and Rolling Hills facilities in PJM, which added about 2.2 GW and took the US portfolio to roughly 60% of capacity and adjusted EBITDA.

That chart is the reason Capital Power ranks where it does rather than lower. Nine of the ten names here pay you nothing while you wait for the AI thesis to resolve. This one has raised its dividend every year for thirteen years, from a generation fleet that would exist whether or not a single additional data centre is built.

The technical picture, which is not good. Capital Power closed at $62.29 on September 21, 2026. The 50-day is $66.37 and the 200-day is $64.40, so the fast average is still 3.1% above the slow one and the golden cross from June 24, 2025 is technically intact, but the price has fallen through both and the gap between the averages has narrowed from 5.6% in mid-September. That combination usually resolves into a death cross within weeks unless the price recovers.

The precedent. Across six completed golden-cross regimes the median peak gain was 40.3%, the best 80.8%, and a new all-time high was reached in five of six. The current regime peaked at 40.3% and has given back most of it, sitting at 12.3% above the signal. The death-cross record is milder: six regimes, median peak gain 11.5%, no new highs. On six observations either way, the useful reading is directional only: Capital Power’s uptrends have run long in every case on record, and this one has not yet ended.

The risk. The Meta load does not arrive until the back half of 2028, and Alberta power prices between now and then are a merchant exposure. Add the interest-rate sensitivity of any capital-intensive utility and the possibility that Alberta’s data-centre queue converts more slowly than announced. If it is the power theme rather than this particular name that interests you, our Canadian energy stock rankings cover the generation and pipeline businesses on the same demand curve.

6. OpenText (TSX: OTEX)

Cheap, cash-generative, and the only name here whose AI product is visibly changing what customers buy.

The macro reason. Enterprise AI needs enterprise data, and most of it is sitting in unstructured content repositories that nobody has touched in a decade. OpenText owns a great many of those repositories. Its Aviator agents are the attempt to turn that archive from a storage cost into an AI substrate.

What the filings show. In the fourth quarter of fiscal 2026, ended June 30, 2026, OpenText reported total revenue of $1.349 billion, up 2.9%, cloud revenue of $503 million, up 6.0% and a twenty-second consecutive quarter of cloud organic growth, and enterprise cloud bookings of $295 million, up 24.1%. Adjusted EBITDA was $507 million at a 37.6% margin. GAAP net income was $156 million and GAAP diluted EPS $0.64, with non-GAAP figures of $299 million and $1.23. Those are the company’s own numbers from its fourth-quarter and fiscal-year results release.

The bookings line is the one that matters. Revenue growing 2.9% while forward cloud bookings grow 24.1% is the signature of a business whose mix is turning over, and OpenText attributes the turn to AI-attached cloud deals.

The technical picture. OpenText closed at $32.42 on September 21, 2026 with the 50-day at $33.15 and the 200-day at $33.88, the fast average 2.2% below the slow one. It has been in a death cross since February 4, 2026.

The precedent, which is a genuine warning. OpenText has had nine death crosses in ten years. Across the eight completed ones it made a new all-time high in zero, and the median peak gain from the signal was only 12.0%. The golden-cross record is barely better: nine regimes, median peak gain 21.2%, a new high in three of nine. This is a stock whose crossovers have produced tradeable bounces and almost never a durable trend. On nine observations that is a pattern worth noting and not a rule, but combined with a share price down 33% over three years it argues that “cheap” has been the correct description of OpenText for a long time without being a catalyst.

The risk. The debt taken on for Micro Focus is still being worked down, organic growth outside cloud is negative, and the Aviator attach rate is a company-supplied narrative rather than a disclosed revenue line. Our coverage of the software selloff that took the whole Canadian tech complex down in September puts OpenText’s year in the sector context it belongs in.

7. Descartes Systems (TSX: DSG)

The best-run business on this list, with the weakest claim to being an AI stock.

The macro reason. Descartes operates the Global Logistics Network, which carries the customs filings, carrier connections and shipment data of a large share of North American trade. Proprietary transaction data of that kind is exactly what general-purpose models do not have, and it is the input that makes an AI application in logistics defensible rather than a wrapper. The tariff and trade volatility of the last two years has also made the network itself more valuable, because complexity is what customers pay Descartes to absorb.

What the filings show. In the second quarter of fiscal 2027, the quarter ended July 31, 2026, Descartes reported record revenue of $201.1 million, up 12% from $179.8 million a year earlier and up 4% from $193.6 million in the prior quarter. Services revenue was $188.6 million, 94% of the total, up 13%. Gross margin was $156.7 million against $138.2 million. The company’s baseline calibration, disclosed in the same release, put estimated baseline revenue for the third quarter of fiscal 2027 at approximately $181.0 million against baseline operating expenses of approximately $111.5 million.

The technical picture. Descartes closed at $111.33 on September 21, 2026 with the 50-day at $106.35 and the 200-day at $104.10. A golden cross fired on September 3, 2026 and the stock has recovered to within 0.1% of the signal price, after trading 5.5% below it in mid-September.

The precedent. Five completed golden-cross regimes, median peak gain 43.8%, best 83.7%, worst 21.2%, with a new all-time high in four of five. That is a strong record and the strongest worst-case on this page: the least profitable golden cross Descartes has produced in a decade still reached a 21.2% peak. Five observations, so treat it as directional. The death-cross record is unusually benign too, with a median peak gain of 15.7% from the signal, which tells you this is a low-volatility compounder whose chart signals mean less than they would elsewhere.

The risk, and the reason it ranks seventh. Descartes is down 28.8% over twelve months despite that record, because a 12%-growth business trading at a premium multiple is vulnerable when the market decides to pay less for software. Its measured AI exposure is 5.0%. Nothing about the AI build-out is visible in its revenue. It earns its place because the network data is a genuine AI asset and the business is excellent, not because the AI trade is paying for it. Our analysis of why the “EPS miss” label misread Descartes’ quarter goes through the reported numbers in detail.

8. Constellation Software (TSX: CSU)

The stock the market has decided AI will destroy, and the disclosure that says it has not started.

The macro reason, inverted. Every other name on this page is ranked on AI as an opportunity. Constellation is ranked on AI as a threat, because that is what the last twelve months have been about. The company owns hundreds of small vertical-market software businesses, each selling a niche system to a niche industry, each defended by switching costs rather than by technology. If a general-purpose model can rebuild a marina-management system in a weekend, that moat is gone. The market has priced a meaningful probability of exactly that: the shares fell from roughly C$5,060 in June 2025 to a low near C$2,196 in January 2026, and are down 36.1% over twelve months.

What the company’s own disclosure actually shows. This is where the argument gets interesting, because Constellation publishes the number the thesis depends on and it is not widely read. In its management’s discussion and analysis for the quarter ended June 30, 2026, total revenue was $3,335 million, up 17% from $2,844 million, with six-month revenue of $6,516 million against $5,498 million. Organic growth was 3% for the quarter, and 1% after adjusting for currency. Broken out by revenue type: licences $92 million, professional services $573 million, hardware and other $116 million, and maintenance and other recurring $2,554 million, up 19% in total and 4% organically.

Read the black line. Maintenance and other recurring revenue is the subscription base, roughly 77% of company revenue, and the first thing that would fall if customers were being lost. It has grown organically in every quarter Constellation discloses, and 4% in the most recent one. That is not the signature of a business being hollowed out. It is also, at 4% and decelerating from 9%, not the signature of a business that is winning.

The technical picture. Constellation closed at $2,786.97 on September 21, 2026 with the 50-day at $2,959.14 and the 200-day at $2,795.78. A golden cross fired on August 7, 2026 and the stock is 13.4% below the signal price, which means the recovery attempt that produced the cross has already failed once.

The precedent, which is the most striking on this page. Constellation has had five golden crosses in ten years. Across the four completed ones it made a new all-time high in four of four, with a median peak gain of 56.5% and a worst case of 38.4%. There is no losing observation in the sample. There are also only four observations, all of them inside a fifteen-year secular run that the current drawdown may have ended, and the most recent death cross, from September 8, 2025, ran to a 29.5% loss. A base rate of four is a story, not a probability.

The risk. The AI thesis against Constellation cannot be disproved by four quarters of maintenance revenue, because software attrition is slow and shows up at renewal. Organic growth of 1% after currency is thin cover if acquisition multiples rise. The honest position is that the disclosure contradicts the most aggressive version of the bear case and does not refute the patient version.

9. Kinaxis (TSX: KXS)

Real agentic AI in production, sold into supply chains, and priced as though none of it exists.

The macro reason. Supply-chain planning is the rare enterprise problem where an agent has a defensible job: continuous re-planning against changing constraints, at a speed no human team matches. Kinaxis sells Maestro, which its chief executive describes as “a composable agentic AI platform that helps customers connect data, decisions, and actions.” Tariff volatility has made that pitch considerably easier to make.

What the filings show. In its second-quarter 2026 results, Kinaxis reported total revenue of $158.8 million, up 16% from $136.4 million, with SaaS revenue of $106.5 million, up 20% from $88.4 million. Annual recurring revenue reached $465.6 million, up 19% and 21% in constant currency, and remaining performance obligations reached $983.5 million, up 18%. Gross profit was $104.4 million at a 66% margin against 64%. Profit was $21.2 million, or $0.76 per diluted share, against $18.4 million and $0.64. Adjusted EBITDA was $41.4 million, a 26% margin and 130 basis points better than a year earlier. Operating cash flow was $30.7 million against $22.6 million. The company raised full-year 2026 guidance to revenue of $625 million to $640 million and SaaS growth of 18% to 20%, and reaffirmed an adjusted EBITDA margin of 25% to 26%.

That is a business accelerating, expanding margins and raising guidance. The market’s response over twelve months has been minus 6.3%, and the measured AI exposure is 1.7%: the AI trade is giving Kinaxis essentially no credit at all.

A gap we are not papering over. There is no multi-year filing chart in this section. Kinaxis files on SEDAR+ and its investor-relations host and newswire both refuse our requests, so its historical annual filings were not retrievable within this update. Every Kinaxis figure above comes from the company’s own release, and we would rather say that plainly than build a chart from a data vendor’s version of its financials.

The technical picture. Kinaxis closed at $173.90 on September 21, 2026 with the 50-day at $168.38 and the 200-day at $153.87, the fast average 9.4% above the slow one. A golden cross fired on July 28, 2026 and the stock is 4.3% above the signal, having reached 8.1%.

The precedent, which argues for caution. Nine golden crosses in ten years, eight of them completed: median peak gain 19.1%, best 169.4%, worst 0.4%, and a new all-time high in three of eight. The median drawdown inside those regimes was around 3%. Kinaxis crosses frequently and follows through rarely, which is what a stock that has traded in a range since 2021 looks like on a moving-average chart. Eight observations, and the distribution is wide enough that the median is doing very little work.

The risk. Kinaxis competes with SAP and Blue Yonder, both of which are shipping their own agents, and its customer base is concentrated in manufacturers whose capital budgets tighten in a slowdown. The 2026 acceleration is real; whether it survives a competitive response from a vendor already inside the customer’s ERP is the open question.

10. Thomson Reuters (TSX: TRI)

The most advanced AI product portfolio in Canadian large-cap, and the market thinks it is a liability.

The macro reason, and the reason this ranks tenth. Thomson Reuters sells legal, tax and accounting research to professionals who bill by the hour. It has built CoCounsel, released its own domain LLM, and is embedding both across Westlaw and its tax products. It is, on paper, the clearest case in Canada of an incumbent turning AI into product. And it has been the worst-performing large-cap on this list, down 41.4% over twelve months, because the market has decided the same technology that makes CoCounsel possible also makes a legal-research subscription optional.

What the filings show, which is not a company in trouble. In its second-quarter 2026 results release, filed with the SEC, Thomson Reuters reported revenues of $1,954 million against $1,785 million, up 9%, with organic revenue up 8% and organic growth of 10% across its “Big 3” segments of Legal Professionals, Corporates, and Tax, Audit and Accounting Professionals. Recurring revenue, 82% of the total, grew 9%, and transactions revenue grew 16%, offset by a 3% decline in Global Print. Operating profit was $558 million against $436 million, up 28%. Diluted EPS was $1.02 against $0.69, up 48%. Net cash provided by operating activities was $920 million against $746 million and free cash flow was $727 million against $566 million, up 29%. Adjusted EBITDA was $745 million at a 38.1% margin. The company raised its full-year 2026 outlook to approximately 8.0% total revenue growth and 9.5% to 10.0% organic growth for the Big 3, and announced a definitive agreement with KKR to form a joint venture for its Global Print business, selling a 51% stake for approximately $500 million of gross proceeds on closing.

Read those numbers again against a share price down 41.4%. Earnings per share grew 48% in the quarter. Free cash flow grew 29%. Chief executive Steve Hasker’s stated priority for the second half is “further deepening our leadership in trusted Fiduciary-Grade AI solutions,” and the company points to the release of CoCounsel Legal and “the first production ready version of the Thomson LLM.” The market is not disputing any of that. It is disputing how long people will pay for it.

That pair of charts is the argument. Revenue up every year, operating cash flow up every year, reported earnings per share down, and a share price down 41%. Some of the EPS decline is disposal gains rolling out of the base. The rest of the gap is the market applying a lower multiple to a business it thinks has a shorter life than it used to.

The technical picture, and it is recent. Thomson Reuters closed at $134.01 on September 21, 2026 with the 50-day at $139.45 and the 200-day at $135.34. A golden cross fired on September 10, 2026, ending a death-cross regime that began on September 25, 2025 and cost 37.3%. The stock is 1.1% above the signal price and has been as much as 10.9% above it.

The precedent. Five completed golden-cross regimes in ten years: median peak gain 56.8%, best 104.9%, worst 0.0%, with a new all-time high in three of five. The two largest ran 386 and 730 days. That is a favourable record and it is five observations, one of which produced nothing at all. What makes it worth stating is the contrast with the death-cross record, where the median peak gain was 10.1% and no regime made a new high. Thomson Reuters’ crossovers have been more informative than most on this page, and the one that just fired is the bullish kind.

The risk, which is the whole thesis. If large language models make general legal research a commodity, Thomson Reuters’ subscription base erodes regardless of how good CoCounsel is, because the company’s own product would be cannibalising a higher-priced one. The bull case requires that professionals keep paying for verified, citable, indemnified answers. The bear case requires only that they stop. Nothing in the last four quarters of disclosure resolves that, which is why this ranks tenth despite being the best business on the list by most conventional measures.

Where To Buy These Stocks

Nine of the ten trade on the TSX in Canadian dollars, Hut 8 is dual-listed and can be bought in Toronto, and every one can be held in a TFSA, an RRSP, an FHSA or a taxable account. There is nothing exotic to arrange, so the only decisions left are where you hold them and what you pay to get in.

Open a Questrade account if you want zero commissions on Canadian and US-listed stocks and ETFs, no annual registered account fee, and the dual-currency registered accounts that matter the moment you put Nvidia or Microsoft beside these. The full fee schedule is in our Questrade review. Wealthsimple is the simpler option if you want fewer decisions and a cleaner app, reviewed here. If you are still choosing, the best investing apps in Canada ranks the whole field and the best broker for beginners narrows it for a first account.

Two things worth doing before you size a position. The mutual fund fee calculator will show you what an AI-themed fund’s management expense ratio costs over a holding period, which is the relevant comparison if you are choosing between the ETFs further down this page and buying the shares directly. And if you are buying in a taxable account, the capital gains tax calculator is worth running before you sell rather than after.

Names That Came Off This List, and Why

This page has carried a dozen or more names over its history. Saying what left, and why, is more useful than quietly dropping them.

CGI (TSX: GIB.A) was removed on the merits. CGI markets itself as an AI consulting partner and has the client base to make that plausible. The measurement does not support it: its R-squared against Nvidia is 0.4%, among the three lowest of the twenty-five names we tested, its beta is 0.036, and the shares are down 24.8% over twelve months and 30.0% over three years. Its 50-day moving average sits 4.6% below its 200-day. A consulting business that bills hours to implement other people’s AI is a real business, but it is not a way to own the AI build-out, and on this page’s criterion it does not earn a place ahead of the ten above it.

TELUS Digital (formerly TIXT) cannot be bought at all. It was the closest thing Canada had to a listed AI data-services pure play, doing the annotation and human-in-the-loop work that model training requires. TELUS acquired the shares it did not already own for US$4.50 per share in cash and stock, roughly US$539 million in aggregate, and completed the privatization on October 31, 2025. The subordinate voting shares were delisted from the TSX and the NYSE and the company ceased to be a reporting issuer. TELUS’s own media release announcing the completion is not linkable from here: telus.com refuses automated requests, and we would rather leave the credit unlinked than point you at somebody else’s write-up of it. Any list that still carries it is out of date.

Cohere is not investable. Canada’s most significant AI company by any technical measure is private, Toronto-based, and has no listed security. There is no way for a retail investor in Canada to own it, and no listed proxy that meaningfully tracks it. This is the single largest gap between “Canadian AI” as an industry and “Canadian AI stocks” as an investable set.

Four names we tested and did not rank. HIVE Digital has genuine AI exposure at 17.5% R-squared through its high-performance computing business, and it is down 10.3% over twelve months with a share price of $4.20; it sits just outside on scale and balance-sheet volatility. Coveo is the purest listed Canadian generative-AI software business and is down 55.2% over twelve months at $4.09, which is either the opportunity or the verdict, and we are not confident enough to rank it. BlackBerry is up 105.7% over twelve months on QNX rather than on AI. Kraken Robotics has an R-squared of 8.9% and a three-year return of 866%, and belongs in a conversation about robotics rather than this one.

If small, volatile Canadian names are what you are actually looking for, our Canadian penny stock rankings apply a balance-sheet screen to that end of the market, and our Canadian fintech stock page covers the software names that sit adjacent to this list.

Are AI Stocks a Buy Right Now?

The honest answer is that “AI stocks” is not one asset and the question cannot be answered at that level. What the data on this page supports is narrower and more useful.

The compute and power layer is being paid today. Celestica’s revenue grew 62% last quarter and its own guidance implies 65% for the year. Hut 8 has signed roughly $26.6 billion of base-term contract value. Capital Power has a signed ten-year-plus agreement with Meta. These are not forecasts, they are documents. The risk in this layer is that it is cyclical capital spending, priced as though it is not.

The application layer is being priced as a casualty. Constellation is down 36% over twelve months, Thomson Reuters 41%, OpenText 32%, Coveo 55%. In every case the operating numbers are fine and the multiple has collapsed. The market is expressing a view that AI compresses the value of packaged enterprise software, and it is expressing it hard. If that view is wrong, this is where the return is. If it is right, no amount of current cash flow protects these businesses.

The two layers are not diversification. They are opposite sides of the same bet. Owning Celestica and Constellation together is not a hedge, it is a straddle on how far the technology goes.

One number worth keeping in view. Of the twenty-five names we measured, the median R-squared against Nvidia is 5.0%. If you are buying a Canadian stock for AI exposure, the base case is that you are getting very little of it, and you should be able to say what else you are buying. Our page on Canadian blue-chip stocks is the better starting point if the answer turns out to be “a good business,” and our Canadian ETF rankings are the better starting point if the answer is “broad exposure I do not have to time.”

Nothing here is a recommendation to buy or sell any security, and the fact that a company appears on this page is not a prediction that it will rise.

How To Pick AI Stocks

Six tests, in the order we apply them.

1. Find the revenue line. Ask which line on the income statement contains the AI money, then check whether the company discloses it. Celestica’s CCS segment is disclosed and grew 84%. If nobody can point at a line, the AI exposure is a story about the future, which is a legitimate thing to buy but a different thing to buy.

2. Separate the layers. Chips and hardware, power and land, data, models, applications. Each layer has a different cost structure, a different cycle and a different set of competitors. A portfolio of five “AI stocks” from the same layer is one position.

3. Ask what happens if AI is smaller than expected. Every name on this page except Hut 8 has a business that exists without AI. That is the difference between a growth option and a binary.

4. Ask what happens if AI is bigger than expected. This is the test people skip. For Constellation, Thomson Reuters and much of enterprise software, a bigger AI is worse, not better. Work out the sign of the exposure before assuming it is positive.

5. Measure rather than assume. The regression on this page took an afternoon and it contradicted the marketing on roughly half the names we tested. You do not need to run one yourself, but you should be suspicious of any list that ranks companies as AI plays without checking whether the market treats them as such.

6. Check the technical context, then check the sample size. A moving-average crossover tells you what regime a stock is in. The historical record of what happened after past crossovers tells you something only if there are enough of them. Every crossover statistic on this page is published with the number of events behind it precisely because five or eight observations is directional, not predictive. What moves a stock price is worth reading if that distinction is new, and how the stock market works covers the mechanics underneath it.

Can You Invest in AI ETFs?

Yes, and for most people this is the more sensible route than picking from the ten above. A thematic fund removes the single-name risk that dominates this category, at the cost of a management fee and of owning some companies you would not have chosen.

CI Global Artificial Intelligence Fund (TSX: CIAI) is the largest Canadian-listed option. Per CI Financial’s own fund page, it held $1.1 billion of net assets as at September 11, 2026 at a net asset value of $39.6462 per unit, with 29,900,000 units outstanding. The management fee is 0.55% and the management expense ratio was 0.76% as at June 30, 2026. It launched in May 2024, is actively managed, and invests primarily in global equities, which in practice means the large US names rather than the TSX.

Evolve Artificial Intelligence Fund (TSX: ARTI) takes a different approach. Per Evolve’s own product page, it is actively managed with portfolio selection powered by Boosted.ai, rebalances quarterly, carries a 0.60% management fee plus applicable sales taxes, and listed on the TSX on March 22, 2024. It is a small fund.

Global X also lists an AI and technology fund on the TSX under AIGO. We are not quoting its fee or assets here because Global X’s Canadian site refuses our requests and we will not publish a figure we could not read from the manager’s own page. Check the fund page directly before buying.

Two things to understand before choosing any of them. First, essentially every AI ETF available in Canada is dominated by the same handful of US mega-caps, so buying two of them mostly doubles a position rather than diversifying it. Second, a 0.76% MER on a $50,000 position is $380 a year, which the mutual fund fee calculator will compound out over a realistic holding period. Compare that number against the cost of simply buying three or four of the names above at zero commission, and against a broad index fund, before deciding the theme is worth the fee. Our Canadian ETF rankings put these thematic funds in the context of the core holdings most portfolios should own first.

US AI Leaders Canadians Can Buy

The largest AI companies in the world are American, and Canadians can own all of them from an ordinary Canadian brokerage account. Nvidia designs the accelerators that Celestica builds into racks. Microsoft, Alphabet and Amazon are simultaneously the largest buyers of that hardware and the largest sellers of AI capacity. AMD is the credible second source in accelerators, and its Helios platform is one of the programmes Celestica has publicly tied its 2027 growth to. Palantir sells the applied-AI layer into government and enterprise.

We do not rank those here, because this page ranks Canadian-listed companies and because the US names are covered comprehensively everywhere. What is worth covering is the part specific to being Canadian.

How To Buy US AI Stocks From Canada

Three routes, with different costs.

Buy the US listing directly. Any Canadian discount broker will do it. The cost is the currency conversion, usually charged as a spread of around 1% to 2% on the exchange rate rather than as a visible fee, applied every time you move between currencies. On a $10,000 purchase that is $100 to $200 you will not see on the confirmation.

Use Norbert’s gambit to convert first. Buy a dual-listed, interlisted security in Canadian dollars, journal it to the US side of your account, sell it in US dollars, and pay a couple of commissions instead of the spread. Our walkthrough of Norbert’s gambit covers the mechanics, the settlement timing and the mistakes that cost people money. It is worth doing above roughly $5,000 and not worth the trouble below it.

Buy a Canadian Depositary Receipt. CDRs let you buy a fraction of a US share in Canadian dollars on the TSX, with a currency hedge built in. That hedge is not free and its cost is not the number most people assume. We measured it in detail on our Tesla stock in Canada page, using CIBC’s own published ratio and forward-rate series, and found the hedge had cost 2.28% a year against a disclosed 0.60% spread. The same arithmetic applies to the Nvidia and Microsoft CDRs. CDRs are also foreign property for T1135 purposes, which the Tesla page explains.

A currency note that applies to all three. The Canadian dollar’s move against the US dollar is a real component of your return on any US holding, and over a five-year period it can be larger than the difference between two decent stock picks.

Which Canadian Accounts Can Hold AI Stocks?

All of them. The differences are in what each account does to the tax outcome and to your ability to recover from a mistake.

TFSA

Every name on this page and every Canadian-listed AI ETF is TFSA-eligible, and so are US-listed shares. Gains come out untaxed, permanently. The trap specific to this category is that TFSA contribution room is destroyed by losses: if you put $7,000 into an AI stock and it becomes $2,000, you have permanently lost $5,000 of tax-sheltered room. For a group of stocks that routinely fall 30% or more in a year, that is a real cost and an argument for holding your highest-conviction, lowest-drama assets in the TFSA instead. Our TFSA stock rankings and the TFSA contribution room calculator both approach the account from that side, and our TFSA guide covers the rules.

One point of genuine relevance: US-listed shares held in a TFSA are subject to the 15% US withholding tax on dividends and it cannot be recovered. For Nvidia, whose dividend is negligible, this is close to irrelevant. For a dividend-paying US technology holding it is not.

RRSP

The RRSP is the right account for US-listed dividend payers, because Article XXI(2) of the Canada-US tax convention exempts dividends paid into an RRSP from that 15% withholding. Our RRSP stock rankings work through the arithmetic, and the RRSP contribution room calculator and RRSP tax refund calculator cover the contribution side. Everything in an RRSP eventually comes out as ordinary income at your full marginal rate, so the shelter is a deferral rather than an exemption. Our RRSP guide covers the rules in full.

FHSA

An FHSA can hold all of these, and in most cases should not. Money with a dated purpose and a five-year-or-less horizon does not belong in a stock that can fall 40% in a year. Our FHSA investment strategy page sorts holdings by the runway each one needs, and our FHSA guide covers eligibility and the interaction with the Home Buyers’ Plan.

Non-registered

The default once registered room is gone. For this group it is less punishing than for a dividend portfolio, because nothing is taxed until you sell and most of these names pay little or nothing. Capital gains are half-included in income. The capital gains tax calculator will tell you what a sale costs, and the adjusted cost base calculator matters if you buy the same name repeatedly, which most people do.

What Are AI Stocks?

An AI stock is a share in a company whose revenue, costs or competitive position is materially affected by artificial intelligence. That definition deliberately includes companies AI is damaging, because those are shares in AI’s consequences just as much as the winners are, and the market prices both.

The layers, and why they behave differently

• Semiconductors and hardware. Designs and builds the accelerators, servers, storage and optical networking. Highest measured correlation to the AI trade, most cyclical, most exposed to a single customer’s capital-budget decision. Celestica is Canada’s entry here.
• Power and physical infrastructure. Generation, transmission, permitted land, cooling. The current bottleneck. Longer contracts, slower revenue, less volatile businesses, but multi-year construction risk. Capital Power, Hut 8 and Brookfield sit at different points on this layer.
• Data. The proprietary information a general model cannot obtain. Descartes’ logistics network and Thomson Reuters’ legal corpus are both genuine data assets, valued very differently by the market.
• Models. Almost entirely private, and entirely absent from the TSX. Cohere is Canadian and unlisted.
• Applications. Software that uses AI to do a job. The largest layer by company count, the hardest to differentiate, and the layer where AI is as likely to be a threat as an opportunity. Kinaxis, OpenText, Shopify and Constellation all live here.

If any of the underlying concepts are new, what is a stock covers the basics and how to read a stock quote covers the screen you will be looking at. For the wider TSX picture, our best Canadian stocks hub links the rest of these rankings.

Frequently Asked Questions

What is the best AI stock in Canada?

On this page’s criterion, which is how much of the AI build-out actually reaches the income statement, Celestica ranks first. Its second-quarter 2026 revenue rose 62% to $4.70 billion, its Connectivity and Cloud Solutions segment grew 84% and accounted for 81% of revenue, and the company raised its 2026 outlook to $20.5 billion of revenue and $11.30 of adjusted EPS. It is also the only Canadian name whose day-to-day trading is meaningfully tied to the AI trade, at an R-squared of 30.9% against Nvidia. That concentration is a risk as well as a recommendation.

Are there any real Canadian AI companies to invest in?

Fewer than the lists suggest. Canada’s most significant AI company, Cohere, is private and cannot be bought. TELUS Digital, the closest listed pure play, was taken private by TELUS on October 31, 2025 and delisted. What remains listed in Canada is mostly infrastructure, hardware and enterprise software with AI features rather than AI-native businesses. Celestica, Hut 8 and Capital Power are real AI exposure of a physical kind. The software names are real companies with real AI products and, so far, no visible AI revenue.

Can I hold AI stocks in my TFSA?

Yes. Every stock and ETF on this page is TFSA-eligible, as are US-listed AI shares. The consideration specific to this category is that a loss inside a TFSA permanently destroys contribution room, and these are volatile holdings. US-listed dividends inside a TFSA also suffer a 15% withholding tax that cannot be recovered, which matters little for names that pay no dividend.

What is the best AI ETF in Canada?

By size, the CI Global Artificial Intelligence Fund (TSX: CIAI), which held $1.1 billion of net assets as at September 11, 2026 with a management fee of 0.55% and an MER of 0.76% as at June 30, 2026. The Evolve Artificial Intelligence Fund (TSX: ARTI) is an actively managed alternative with a 0.60% management fee and a far smaller asset base. Both hold mostly US mega-caps rather than Canadian companies, so neither is a way to own the names ranked on this page.

Is there a Canadian equivalent of the AIQ ETF?

Not an exact one. AIQ is a US-listed index fund tracking a global AI and technology basket. The closest Canadian-listed options are CIAI and ARTI, both actively managed rather than index-tracking, and both holding a similar set of large US technology companies. A Canadian can also simply buy AIQ itself in a US-dollar account, at the cost of the currency conversion described above.

Is Nvidia still worth buying for Canadians?

That is a question about Nvidia rather than about Canada, and this page does not rank US securities. What is specific to Canadians is the mechanics: you can buy the US listing and pay a conversion spread, use Norbert’s gambit to convert more cheaply, or buy the Canadian Depositary Receipt in Canadian dollars and accept a currency hedge whose real cost we measured at more than three times its disclosed spread on the Tesla CDR. There is also an indirect route: Celestica builds hardware into the same buildout, and its measured relationship to Nvidia is the closest of any TSX-listed company.

How do I buy US AI stocks from Canada?

Open an account at any Canadian discount broker, fund it in Canadian dollars, and either convert to US dollars or buy a Canadian Depositary Receipt on the TSX. Converting through your broker costs roughly 1% to 2% as a spread on the exchange rate. Norbert’s gambit reduces that to a couple of commissions and is worth the effort above about $5,000. CDRs avoid the conversion entirely but embed a currency hedge you are paying for whether or not you want it.

Why are Canadian software stocks falling if AI is a boom?

Because the market has concluded that AI reduces what packaged enterprise software is worth. Constellation Software fell from roughly C$5,060 in June 2025 to a low near C$2,196 in January 2026 on the argument that a general-purpose model can rebuild a niche vertical application cheaply. The company’s own disclosure does not yet show that happening: organic growth in maintenance and other recurring revenue, the line that would erode first, has been positive in all nine quarters it publishes and was 4% in the second quarter of 2026. Whether that holds is the single biggest open question in Canadian technology investing.

---

Disclaimer: The content on bestcanadianstocks.ca is for informational and entertainment purposes only and does not constitute financial advice. Past performance is not indicative of future results. Always consult a qualified financial advisor before making investment decisions. Share prices and moving averages are computed from Yahoo Finance closes and reflect the September 21, 2026 close. Every company financial figure comes from that company’s own filing, named and linked beside it. The AI exposure figures are our own regression of each stock’s daily log return on Nvidia’s daily log return over the 504 trading sessions ended September 11, 2026, and R-squared is the share of a stock’s daily variance the AI trade explains rather than a forecast. Fund fees and assets are from the fund manager’s own published fund page.