IREN · Forward model · Bear case

The Bear case, 20 quarters out

Model as of 2026-08-31

Each segment is projected from its own operating driver, rolled up into consolidated cash flow, and discounted back to a fair value per share. The assumptions below are editable — change them and every number on this page moves with them.

BASIS. Quarter ended 2026-06-30, labelled 2026 Q2 on this repo's calendar convention - IREN calls it Q4 FY2026 and its fiscal year ends 30 June. The label is checked against the actuals arrays, where the prior June quarter carries mining $180.330m plus AI Cloud $6.963m = $187.293m against the stored 2025 Q2 revenue point of $187.29m. Reported after the US close on 2026-08-27 from the 8-K Item 2.02 Exhibit 99.1, accession 0001878848-26-000051, with the FY2026 10-K indexed the same evening. The previous revision was built on the March quarter hours before this report landed and carried a READ THIS FIRST warning saying so; that warning is discharged. The June quarter is the first in which AI Cloud Services was the larger line - $70.5m against $66.7m of Bitcoin mining, 51% of revenue against 23% in March. SEGMENTS AND RECONCILIATION. Bitcoin mining and AI Cloud Services are IREN's only two reportable segments; both revenue and cost of revenue are disclosed per segment, there are no intersegment eliminations, and the two lines sum to consolidated revenue in all thirteen quarters carried here: $70.500m plus $66.700m equals $137.200m, matching the release's total revenue exactly. No split was invented and nothing was derived to fill a gap. Four historical points remain derived by subtraction from disclosed annual and nine-month figures and are flagged estimated (2023 Q2 and 2024 Q2 on both lines); the AI Cloud zeros for 2023 Q2 to 2023 Q4 are flagged estimated because the segment did not exist. The new quarter's two points are disclosed, not estimated. THE HASHRATE DISCLOSURE IS GONE. This is the most important finding in the roll-forward. Installed EH/s, average operating EH/s, Bitcoin mined, fleet efficiency in J/TH and electricity cost per MWh were all disclosed through 2026-03-31 and are all absent from the 27 August release; the FY2026 10-K carries annual durations only. The Bitcoin vertical is a capacity driver, so both of its state variables are now unobservable. Capacity is frozen at the last disclosed 38 EH/s at 2026-03-31, revenue per exahash-quarter is frozen at the March level of $3.0878m, and utilisation is a plug at 56.85% that reproduces the reported $66.7m - a derived 21.60 EH/s operating. The entire 40% sequential revenue decline is therefore booked to utilisation; the true split between fewer machines and worse Bitcoin economics cannot be recovered from public disclosure. That is tolerable only because this line is being switched off on a date. If the disclosure does not return and IREN were still running mining as a going concern, this vertical would have to be re-cut as a growth driver on segment revenue alone, with no hashrate anywhere in it. THE RUN-OFF IS DATED FOR THE FIRST TIME. The previous revision's note said no dated hashrate glide path had ever been published and the disclosure was 'over time'. On this call the CFO said: 'We currently expect mining operations to be effectively decommissioned by the end of December 2026.' The utilisation glide moves from 20% to 55% of the gap a quarter on that sentence, and it is the only driver level moved because the business changed rather than because the basis moved. A linear decommissioning from the 30 June level to zero on 31 December implies about $33.5m in the September quarter and $11.2m in December; 55% lands $30.0m and $13.5m, a two-quarter total within 2.7% of that. The glide leaves about $10m of tail across CY2027 against a stated zero, which is stated rather than tuned away. THE IMPAIRMENT IS STILL NOT A RUN RATE, AND IT IS BIGGER. $450.4m of impairment in this quarter and $638.8m for the year, 'primarily related to the decommissioning of Bitcoin mining hardware as sites are converted to support AI Cloud growth', plus a $102.1m decrease in the fair value of mining hardware held for sale. The previous revision carried a pre-announced estimate of roughly $520m; the actual charge came in at $450.4m in one quarter. All of it sits below the cash gross margin line and is in neither revenue nor margin, exactly as IREN's own adjusted EBITDA excludes it. Mining's margin is ($66.700m - $24.100m) / $66.700m = 63.87%, down from 68.21%; AI Cloud's is ($70.500m - $9.200m) / $70.500m = 86.95%, up from 86.33%. The release does not break segment cost of revenue into electricity, employee benefits and other direct expenses the way the 10-Q did, so that component detail is gone too. WHAT THE PRINT ARGUES AGAINST AND THIS REVISION DID NOT CHANGE. Three things. First, corporate overhead at 10% of revenue: implied cash corporate cost in the June quarter was $84.7m ($42.6m of mining gross profit plus $61.3m of AI gross profit less the reported $19.2m of adjusted EBITDA), or 61.7% of revenue, and the CFO guided September cash SG&A up another $40m-$50m sequentially. Ten percent is what the function should cost against a revenue base an order of magnitude larger; it understates the near-term dollar burn by more than the previous revision did, and it is left alone. Second, the AI Cloud capex intensity of 440%: it was fitted to $8.33m of ARR per gross MW, and this release restates the ladder in IT megawatts and quotes recent three-year contracts above $20m of revenue per MW (IT) - roughly $13m per gross MW at a 0.65 conversion, 56% more revenue against data centre and GPU capex the CFO says is up only 15-20%. That argues the intensity is now too high. It is left alone and flagged, because the print supersedes the old anchor without publishing a clean replacement per-megawatt capex figure. Third, the AI Cloud sequential growth rate of 150%: the trailing pace over the last three quarters is +135%, +94%, +110%, and the ARR bridge implies 170% for the September quarter. 150% sits between them and is left where it is. THE GUIDANCE THIS MODEL IS MEASURED AGAINST. IREN publishes no revenue or EPS guidance. The nearest thing is the CFO's ARR bridge: about $0.5bn of ARR exiting the June quarter, $1bn operating on 27 August after Microsoft accepted Horizon 1 on 13 August and carrying to the end of the September quarter, more than $4bn contracted by the end of the December quarter with the revenue effect landing 'predominantly in the March quarter'. Day-weighting $0.5bn over 44 days and $1.0bn over 48 across the 92-day September quarter gives average ARR of $761m and a revenue equivalent of $190.2m; the model's first projected quarter puts AI Cloud at $176.3m, a residual gap of -7.3%, and total revenue at $206.3m. On the full year, fiscal 2027 revenue is $2,465.1m and CY2027 is $6,028m, exiting at a $9.66bn annualised run rate against a 0.8GW (IT) ladder worth more than $16bn of ARR at the disclosed pricing - about 60% of it, the same deliberate under-shoot the previous revision carried. The two anchors are not consistent with each other and the model is matched to the near quarter, not tuned to both. CAPITAL. The $11,899,054,000 of twelve-month commitments that sized the corporate programme is superseded by explicit FY2027 capex guidance of approximately $25bn to $30bn, covering exactly the four quarters 2026 Q3 to 2027 Q2. The AI vertical's revenue-proportional capex spends $6.402bn across those quarters, so the programme carries $21.098bn and the four quarters total the $27.5bn midpoint. That is the single largest change in this revision and it more than doubles the near-term outflow. Against it the company puts $14bn of existing cash and committed GPU financing and prepayments, a targeted further $8bn of GPU financing and prepayments not yet raised, and an unencumbered data centre portfolio to borrow against. Free cash flow is negative in every quarter until 2030 Q4. NET CASH, SAME DEFINITION, RECHECKED. The previous value reproduces exactly: cash and restricted cash $2,213.274m less debt $3,687.832m less finance leases $274.256m = -$1,748.814m at 2026-03-31. Applying the identical definition at 2026-06-30 gives $7,619.5m of cash and restricted cash from the cash-flow statement, less $7,592.944m of debt ($169,370,000 current plus $7,423,574,000 non-current), less $243.796m of finance leases, equals -$217.240m. The cash figure is to one decimal in millions because the release prints it that way; the debt and lease figures are 10-K exact. Two things this definition deliberately does not do: it does not net the $1,842.6m of deferred revenue, which is customer money against undelivered service and is arguably the largest liability on the page, and it does not exclude the $1.7bn of restricted cash, most of which is set aside for Microsoft GPU capex. Both would make the number materially worse. Shares are 394,058,648 as of 2026-08-14, the most recent count disclosed, against 380,193,608 at 30 June; convertible underlying shares, the 30m NVIDIA investment rights at $70 and 14.8m of RSUs are charged nowhere in the fair value. SEASONALITY. Not recomputed, and still absent from both verticals. The previous revision's test - ratios to a centred four-quarter moving average - found a window-to-window spread that met or exceeded the entire seasonal signal on both lines. Adding one quarter does not change that, and the case is now stronger rather than weaker: the Bitcoin line is being decommissioned on a date and the AI line is a ramp off a base that doubled in a quarter. What moves these lines is the Bitcoin price, IREN's own decommissioning schedule and the date a data centre passes customer acceptance - none of them a calendar. WHAT THE MODEL PRINTS. Base case: 2026 Q3 revenue $206.3m (AI Cloud $176.3m, mining $30.0m); fiscal 2027 $2,465.1m; CY2027 $6,028m; terminal quarter 2031 Q2 at $10,474m with mining at zero; free cash flow negative in every quarter until 2030 Q4; fair value $51.01 against a $40.53 close. Bear -$31.58, bull $186.81, Roberts $169.92. The spread is enormous and that is the honest output of a company spending $27.5bn in a year to earn $2.5bn.

Where each case comes from

From cash flow to fair value

Capex outside the verticals

Each vertical's capex is a share of its own revenue, which is the right shape for capacity that scales with sales. These are not that: fixed-size programmes with their own schedule that spend whether or not any segment grows. They are added on top, and they are why free cash flow dips in the years below.

The projected path

Every row is projected. A year-over-year change is shown only where the quarter it compares against exists — an em dash means there is no comparable quarter, not a flat year.

Model revisions

Assumptions are marked to reality as each quarter prints. Every change is appended here, with the fair value the model produced at the time, so the model's own history stays visible.