Riding the AI wave in a stock that got cut in half
How the Market Wave algorithm returned +63% to +106% on IREN across three runs while buy and hold lost as much as 48%.
Three Market Wave runs on a single stock: IREN, the bitcoin miner turned AI data center operator whose shares round-tripped from the high $20s to a $76.41 record close and back to the high $30s in under a year.
Why this case study, and why now
We have already shown what the Market Wave algorithm does on equities that frustrate investors in different ways. Tesla, a great company whose stock spent years stuck in a wide, jagged range. Nvidia, the strongest secular uptrend in the market that almost nobody felt they could actually buy. Apple, the position everyone owns and nobody manages. In the Apple study we made a forward-looking argument: the AI infrastructure names building the next decade would put their holders through drawdowns severe enough that most would sell into them, and the difference between a good outcome and a great one would come down to whether you had a rule.
IREN is that argument, ten months later, in numbers.
From mid-October 2025, days after a $9.7 billion Microsoft contract sent the stock vertical, IREN ran to a record close of $76.41 and then fell as much as 52% from its highs. An investor who bought the euphoria and held was down nearly half by mid-July 2026. An investor who dollar-cost averaged through it roughly broke even. The Market Wave algorithm, run three ways across that same stretch, made money in every window we tested.
The round trip to nowhere
Strategy+63.4%
- Buy and hold
- -4.6%
- DCA
- -2.1%
The full round trip
Strategy+75.4%
- Buy and hold
- -48.1%
- DCA
- -3.2%
The same crash, worked harder
Strategy+106.2%
- Buy and hold
- -44.2%
- DCA
- -3.4%
All three runs started from the same hypothetical $100,000 in cash and zero shares. The last two run the identical calendar window and differ only in how the algorithm was configured, which is the heart of this study: one asset, one window, one algorithm, two different settings, both of them profitable through a drawdown that cut buy and hold nearly in half.
Backtested results. Past performance does not guarantee future results. For educational purposes only.
The wildest ride in AI infrastructure
IREN Limited, formerly Iris Energy, is a Sydney-founded operator of power-rich data centers in North America. It built its business mining bitcoin. Over the last two years it has repurposed that footprint into AI data centers and an AI cloud business, and in doing so became one of the most closely watched and most violently traded names in the AI infrastructure build-out.
The price history of our test windows reads like a compressed decade:
- Early September 2025The stock trades in the high $20s
- Oct 2025A $9.7 billion five-year Microsoft AI cloud contract sends the stock up roughly 30%, and it keeps going
- November 5, 2025An all-time-high close of $76.41
- December 2025The give-back. Financing worries sweep the AI data center complex and the stock is nearly cut in half from its peak, closing at $39.92 on December 19
- Spring 2026A recovery to roughly $56 by mid-May, helped by $3.6 billion in GPU financing at under 6% and construction progress on the liquid-cooled Horizon data centers serving the Microsoft contract
- July 2026A second collapse, to $33.61, about 52% below the highs, after reports that Meta planned to enter the commercial cloud business raised competition fears across the sector
- Jul 20, 2026A 17% single-day recovery on $2.8 billion of new AI cloud contracts and a raised revenue target
- Early August 2026The stock trades in the high $30s
Here is the catch that matters for this study. Nothing about IREN's business went backward across this window. The Microsoft contract stood, the buildout accelerated, the customer list grew, and the revenue target went up. The stock still fell by half, twice. Owning the AI infrastructure story was never the hard part. Surviving the way the market prices it, week to week, was.
The story was right and the holders still lost
IREN's holders over this stretch were not wrong about the company. They were wrong about themselves. Three failure modes did the damage.
Buying the euphoria and calling it conviction
The Microsoft announcement was real, enormous, and immediately priced in. An investor who bought in mid-October 2025 near $70, three weeks before the record close, was buying a true story at a price that already assumed it. By mid-July 2026 that position was down nearly half, with the story fully intact. Conviction about the business is not a plan for the stock.
Narrative whiplash on a weekly cadence
Inside ten months the consensus on IREN flipped at least four times: transformational Microsoft partner, overextended borrower in a financing squeeze, roadkill in Meta's path, then contracted AI cloud operator with a $4 billion revenue target. Each flip arrived with a double-digit move. An investor reacting to the narrative sold the December low, bought the May recovery, sold the July low, and missed the July 20 repricing. The narrative was a lagging indicator every single time.
The falling-knife freeze
When a stock is 40% off its high and still dropping, buying feels reckless and selling feels like capitulation, so most people do neither. They ride, they watch, and they turn a trading account into a hostage situation.
Monthly dollar-cost averaging deserves its usual credit. By spreading entries across the entire round trip it roughly broke even, in a window where a lump-sum holder lost nearly half. That is the diagnosis working as designed. But notice what "working" meant here: the DCA benchmark ended the ten months slightly underwater, with most of its capital still sitting in cash on its schedule, and with no mechanism to ever take a profit. It avoided the disaster. It captured nothing.
This is the opening a rules-based algorithm is built for: keep the emotional distance that a schedule provides, add back awareness of price and structure, and treat a violent range as raw material instead of a threat.
What Market Wave actually does
Market Wave is Arch Public's trend and momentum capture algorithm. It does three things with rules instead of emotions:
Ride confirmed directional moves
It enters on rule-based confirmation, not on a headline, a hunch, or a feeling that a price is too high or too low.
Exit on rules, not nerves
It takes profits into strength on defined triggers, banking gains into cash rather than holding every share on principle.
Step aside cleanly when the trend is not there
In chop and indecision it is willing to hold cash and wait, which is exactly when human investors tend to overtrade.
On a stock like IREN, that combination matters more than anywhere we have tested it. A name that swings 30% in a month punishes every emotional response: the euphoric buy, the capitulation sell, the frozen hold. A rule does not know the Microsoft deal is exciting or that a Meta headline is frightening. It sees price reach a defined level and it acts, in both directions, over and over.
The algorithm is configurable. The same logic can run as a scalp recipe working an hourly chart, a fast swing, or a daily swing riding the larger legs. Scenarios 2 and 3 in this study run the identical ten-month window with two different settings, which is the cleanest possible demonstration that the recipe is a dial, not a fixed setting.
Backtested results. Past performance does not guarantee future results. For educational purposes only.
“The round trip to nowhere”
January 1, 2026 to August 6, 2026, seven months · Hourly chart · Scalp recipe
The "round trip to nowhere" test, and the opener. This window skips the 2025 fireworks entirely. It opens with IREN near $40, after the December collapse and before the spring recovery, and it closes seven months later. It contains the run to $56, the crash to $33.61, and the July rebound, and it nets out to a stock that went effectively nowhere: buy and hold returned -4.6%, and the monthly dollar-cost-averaging benchmark returned -2.1%.
This is a scalping cadence, hundreds of small rule-based decisions on an hourly chart that nobody executes by hand. Run that way, Market Wave turned the same seven months into $163,389, a gain of 63.4%.
+63.4%
- Buy and hold -4.6%, outperformance +68.0 percentage points
- DCA -2.1%, outperformance +65.5 percentage points
- Ending value
- $163,389
- Realized profit, banked in cash
- $70,266
- Shares held
- 3,858 at $39.70 avg cost
- Cash
- $17,094
Open full-size chart
Backtested, valued August 6, 2026. Past performance does not guarantee future results. For educational purposes only.
“The full round trip”
October 15, 2025 to August 6, 2026, ten months · Hourly chart · Fast swing
The "full round trip" test. This window starts three weeks before the all-time high, with IREN in the low $70s and the Microsoft euphoria at full volume. It is, deliberately, the worst reasonable entry point in the stock's public history: the moment the maximum number of real investors were actually buying. From there the window absorbs the November peak, the December halving, the spring recovery, and the July collapse and rebound.
Buy and hold from that entry lost 48.1%. The dollar-cost-averaging benchmark, spreading its entries across the whole ride, lost 3.2%. Market Wave, run as a fast swing with entries restricted to confirmed weakness and a trend-breakout trigger, turned $100,000 into $175,428, a gain of 75.4%.
+75.4%
- Buy and hold -48.1%, outperformance +123.5 percentage points
- DCA -3.2%, outperformance +78.6 percentage points
- Ending value
- $175,428
- Realized profit, banked in cash
- $92,430
- Shares held
- 4,142 at $42.02 avg cost
- Cash
- $18,363
Open full-size chart
Backtested, valued August 6, 2026. Past performance does not guarantee future results. For educational purposes only.
“The same crash, worked harder”
October 15, 2025 to August 6, 2026, ten months · Daily chart · Daily swing
The "same crash, worked harder" test, and the headline result. This is the identical calendar window as Scenario 2, October 15, 2025 to August 6, 2026. Nothing about the market changed. We changed the recipe: a much faster wave setting that reads the swings inside the swings, symmetric position sizing, and a wider profit trigger. Where Scenario 2 ran a fast swing, this configuration treated every leg of the round trip as its own trade.
The benchmarks tell the same story as before: buy and hold lost 44.2% on this run's reference pricing, and the dollar-cost-averaging benchmark lost 3.4%. Market Wave turned $100,000 into $206,199, a gain of 106.2%. It more than doubled the account inside ten months, in a stock that spent those ten months being cut in half.
+106.2%
- Buy and hold -44.2%, outperformance +150.4 percentage points
- DCA -3.4%, outperformance +109.6 percentage points
- Ending value
- $206,199
- Realized profit, banked in cash
- $114,631
- Shares held
- 5,435 at $39.48 avg cost
- Cash
- $49
Open full-size chart
Backtested, valued August 6, 2026. Past performance does not guarantee future results. For educational purposes only.
Three windows, one algorithm
| The round trip to nowhere | The full round trip | The same crash, worked harder | |
|---|---|---|---|
| Window | Jan 2026 to Aug 2026 | Oct 2025 to Aug 2026 | Oct 2025 to Aug 2026 |
| Recipe | Scalp recipe | Fast swing | Daily swing |
| Chart | Hourly | Hourly | Daily |
| Strategy return | +63.4% | +75.4% | +106.2% |
| IREN buy and hold return | -4.6% | -48.1% | -44.2% |
| vs buy and hold | +68.0 pts | +123.5 pts | +150.4 pts |
| Monthly DCA return | -2.1% | -3.2% | -3.4% |
| vs DCA | +65.5 pts | +78.6 pts | +109.6 pts |
| Strategy ending value | $163,389 | $175,428 | $206,199 |
| Realized profit | $70,266 | $92,430 | $114,631 |
| Shares held at end | 3,858 | 4,142 | 5,435 |
| Average cost basis | $39.70 | $42.02 | $39.48 |
| Cash on hand at end | $17,094 | $18,363 | $49 |
All windows start from $100,000 in cash and zero shares.
Backtested results. Past performance does not guarantee future results. For educational purposes only.
The recipe is the product
The instinct is to look at three different numbers and assume one run is the "right" one and the others are noise. They are not. They are the same algorithm answering three different questions.
Scenario 1 asked: what can rules do with a stock that goes nowhere?
Answer: $70,266 in realized profit and a 63.4% return across seven months in which buy and hold lost 4.6%. A round trip is only "nowhere" if your plan is to hold. If your plan is to trade confirmed weakness against confirmed strength, a round trip is two trends.
Scenario 2 asked: what happens if you start at the worst possible moment?
Answer: 75.4% from an entry three weeks before the all-time high, against a 48.1% loss for holding. The fast swing survived the entry nobody should survive, because it never treated the entry as a commitment. It treated it as the first data point.
Scenario 3 asked: how much is the volatility itself worth?
Answer: 106.2%, with realized gains exceeding the starting capital. IREN's violence, the thing that made it unownable for humans, was the raw material. The daily swing harvested it leg by leg.
A dollar-cost averager, by contrast, owns exactly one decision: buy on schedule. On IREN that discipline was enough to avoid catastrophe, and it deserves that credit. It was not enough to make money. Averaging blind into a round trip produces an average price, and an average price in a stock that ends where it started produces approximately nothing.
The algorithm's edge is not prediction. It is the willingness to sell strength and buy weakness, mechanically, in a stock where doing either by feel required courage nobody reliably has.
Backtested results. Past performance does not guarantee future results. For educational purposes only.
The AI build-out will pay the disciplined, not the devoted
IREN signed $12.5 billion of contracts, raised its targets, and built its data centers. The stock still halved, twice. Both things were true at once, and they will keep being true, for IREN and for every name in the cohort.
- 01
A great story is not a strategy
Every IREN holder in this window had the story right. The Microsoft deal was real, the pivot worked, the revenue is contracted. Holding on story alone still lost 48% from the October entry. The market does not pay you for being right about the business. It pays you for having a plan for the price.
- 02
Volatility is a cost to holders and an asset to rules
The same swings that took buy and hold down by half generated $70,266, $92,430, and $114,631 in realized, banked profit across our three runs. Where a human feels a 30% swing as fear, a rule registers it as distance, and distance is what a trading system converts into cash.
- 03
The recipe is the product
The identical ten months produced +75.4% on the fast swing and +106.2% on the daily swing, both starting from the worst entry on the chart. How hard the algorithm works a position is a choice you make once, calmly, in advance, instead of a hundred times under pressure.
The AI infrastructure build-out is not close to over, and neither is the volatility that comes with pricing it in real time. You do not have to predict either one. You have to be systematically present for both.
Backtested results. Past performance does not guarantee future results. For educational purposes only.
The plan matters more than the pick
Arch Public's equities strategies are live, bringing rules-based automation to individual stocks and ETFs across the AI infrastructure complex and beyond. The architecture:
Zero custody. Your assets never leave your account.
Full control. Pause, adjust, or stop any strategy at any time.
Rules, not reactions. The algorithm executes the plan, so neither the euphoria of a $9.7 billion headline nor the fear of a 50% drawdown makes the call.
IREN is the clearest case we have published for why the plan matters more than the pick. The pick was fine: the company delivered. The holders still lost half, the dollar-cost averagers broke even, and the same ten months run through a rules-based algorithm produced +75.4% on one recipe and +106.2% on the other, with more cash banked than the account started with.
The question worth asking is the one this whole series keeps arriving at: when the next 50% drawdown comes to a stock you believe in, what exactly is your plan, and who is going to execute it at 2 AM when you will not?
Buy and hold, the same crash, worked harder
-44.2%
The Market Wave Algorithm for Equities, daily swing
+106.2%
$114,631 banked as realized profit
Two comparisons matter here.
The benchmark rows. In every column, both passive approaches finished underwater. This is the first study in the series where that is true, and it is exactly why we ran it. On Alphabet and Apple the debate was how much return a rule adds to a winner. On IREN the debate is simpler: the rule was the difference between profit and loss.
Columns two and three share an identical window and differ only in recipe. The fast swing made 75.4% and the daily swing made 106.2%, banking $114,631 in realized gains along the way. Same stock, same ten months, same algorithm. What you ask the algorithm to do is a decision, and either decision beat both benchmarks by a wide margin.