Built by Grok 4.6 High

Reports Friday, October 23, 2026. Prepared Thursday, September 24, 2026. This is the fiscal Q3 2026 print, the quarter that ended September 30. It has not been reported. The date is an FMP row with null actuals, last updated the day this page was prepared. It lines up with last year's fiscal Q3 release on October 24, 2025. Company IR, checked the same day, still says no upcoming events are scheduled. No press release has set the date or the session. Last year's Q3 release was after the close. The last print, fiscal Q2 on July 28, was before the open. Do not assume which session this one is. Core Scientific is a calendar-year company. The homepage calendar tags this date 2026Q4 because it stamps the announcement month. That tag is one quarter ahead of the quarter being reported. This page is 2026Q3. The baseline is fiscal Q2 2026, reported July 28.

CORZ | Earnings Preview

Core Scientific reports fiscal Q3 with no revenue range, no EPS range, and no margin range. The only sentence that applies to this print is qualitative. On the July 28 call the CFO said GAAP colocation revenue should take another meaningful step-up, after billing reached 437 MW by mid-July, which he called ahead of expectations. That capacity was already online before the quarter was half over. The operative score is conservative on delivery, not guided on the P&L. It is not an aggressive guide. There is no company number sitting above the street.

The growth rate just inflected, and the cause is mix. Total revenue year over year went from −14.9% in Q3 2025 ($81.103 million versus $95.354 million) to −16.0% in Q4 2025, then +44.9% in Q1 2026 ($115.244 million versus $79.525 million), then +108.8% in Q2 2026 ($164.201 million versus $78.628 million). That is an acceleration of 63.9 percentage points in the year-over-year rate from Q1 to Q2. Colocation did it. Colocation year-over-year growth went from +44.6% in Q3 2025 to +268% in Q4 2025, +805% in Q1 2026, and +1,194% in Q2 2026 ($136.669 million versus $10.560 million). Colocation was 13% of Q2 2025 revenue and 83% of Q2 2026 revenue. Self-mining went the other way and the decline got steeper: −15.7%, −47.2%, −55.2%, then −65.5% ($21.535 million versus $62.424 million). Those colocation percentages are off a sub-$15 million base. They will slow as the base grows even if execution stays clean. Q3 2025 colocation was still only $14.951 million, so this print's year-over-year rate will still look enormous.

Gross margin expansion accelerated. The operating loss did not follow. Consolidated gross margin on the reported percent series expanded +1,600 bps in Q1 2026 (26% versus 10%) and +3,700 bps in Q2 2026 (43% versus 6%). Q3 2025 gross margin was only 5%, so the year-ago comp for this print is easy. Colocation gross margin reached 59% versus 11%, +4,800 bps, but the year-over-year basis-point gain was 400 bps smaller than Q1's +5,200 bps. Self-mining gross margin is −56% versus 5% a year earlier, −6,100 bps. The segment no longer covers its own cost. Gross profit was $70.044 million versus $5.025 million. Operating loss still widened, to −$78.468 million from −$26.284 million. GAAP SG&A was $49.389 million, above the roughly $36 million cash SG&A figure on the call. Dollars of gross profit are the trajectory. GAAP operating income is not there yet.

The street bar is a full quarter of capacity that was already billing, and it is tight versus the only number management quantified. The July 28 release put 437 MW of billing, as of mid-July, at about $635 million of average annualized colocation GAAP revenue. That is about $159 million a quarter if that rate is held flat for three months. It is a spot GAAP annualized figure, not a guide, and it is not Q3 revenue times four. Q2 self-mining plus hosted mining was $21.535 million plus $5.997 million, or $27.5 million, and mining is still being wound down. A flat 437 MW quarter plus a smaller mining book lands in the mid-$170 millions to the low-$180 millions, short of FMP's $198.5 million, unless more megawatts are recognized after mid-July or straight-line GAAP runs above that spot. The street is therefore demanding relative to the published anchor. It does not require AMD revenue, and it does not require the 590 MW CoreWeave contract to be finished. Dalton Phase 2 is still an early-2027 completion.

There is a second arithmetic read, and it is why a beat is still possible without a heroic exit. Realized colocation dollars per average billing megawatt were about $0.44 million in both Q1 and Q2, because megawatts came on during the quarter and GAAP straight-lines the lease escalators. An exit from the summer target (above 450 MW) through a modest further step can put quarterly colocation in the mid-$180 millions before residual mining, which can clear $198.5 million. Use that as the upside case. Do not treat $635 million divided by four as management's number. They did not give one. The deck's separate roughly $1.8 billion annualized colocation figure is the full contracted book, including AMD megawatts that do not bill until 2027.

EPS of −$0.01 is not an operating forecast. Q2 diluted EPS was −$3.32 and the net loss was −$1,155.310 million. The adjusted-EBITDA bridge adds back a $1,045.515 million change in fair value of warrants and contingent value rights. The July 28 release says that mark, reflecting the share price, drove the loss. Q3 2025 EPS of −$0.46 had the same class of item, a $74.864 million charge. Another mark of Q2's size swamps −$0.01 in either direction. It would not mean the colocation step-up failed. Adjusted EBITDA on the Daloopa series was $41.103 million in Q2 2026 versus $21.503 million in Q2 2025, +91%. The company's own release table prints $28.5 million for that year-ago quarter and $8.9 million for Q1 2026, against Daloopa $4.352 million. Direction is up on either Q2 2025 base. The percentage is not. This page uses the Daloopa series and does not blend the two definitions.

Where management and the street still disagree is 2027, not this quarter. Since July 28 the company contracted AMD for 529 MW of critical IT, with revenue from 2027, closed the roughly $444 million Polaris power purchase, added $600 million of undrawn credit, and received conditional ERCOT status on about 1.4 GW in Texas. None of that is Q3 revenue. On September 14 the chief development officer said AMD will take the next roughly 2,000 MW unless the market really goes south. The lease is a reservation right on a further 1,925 MW through December 28, 2028, at certain times and under certain circumstances. Management is repeating a bullish outcome the contract does not require. The street can refuse to underwrite it. Signing was late against their own October 2025 promise. Megawatts, since May, have been early. That split is the credibility check. It does not change the Q3 colocation math. FY2027 FMP revenue of $1,149 million, versus $686 million for FY2026, is where AMD is being counted. The street's growth rate decelerates from +115% in FY2026 to +67% in FY2027. Management has not framed either year with a range.

What to watch on October 23

• Billable megawatts at September 30 versus 395 MW at June 30, versus 437 MW in mid-July, and versus the May checkpoint of more than 450 MW by the end of summer. A hold at 437 MW is the plausible miss. Finishing 590 MW would be a surprise, not the base.
• Colocation revenue versus $136.669 million and versus the roughly $159 million quarterly equivalent of the $635 million spot. Total revenue versus $198.5 million after the mining step-down.
• GAAP colocation margin near 59%, and whether the 80% to 85% CoreWeave cash gross-profit target is reaffirmed. That cash target is not the GAAP margin. Power pass-through sits in both revenue and cost.
• Cash SG&A back toward the low $30 millions. Q2 ran about $36 million, with about $4 million called one-time financing fees. They have refused to call the baseline guidance.
• Whether another warrant, contract-termination, or startup-cost item dominates EPS. Score the operating line separately.
• Any slip of first AMD megawatts past the first half of 2027, or a narrower final ERCOT allocation at Pecos or Hunt. That hits the outer year, not Q3 revenue. Denton, the site already billing, was placed outside Batch Zero.
• Whether the August facility actually releases about $300 million of restricted cash. Cash purchases of property and equipment were −$565.0 million in Q2.

Applied Digital on October 8 and Digital Realty on October 22 are the only similar names before this print, and those dates are FMP, not re-checked on each IR site. CoreWeave, the tenant, reports November 9, after this print. No bitcoin miner in the checked set reports first.

Core Scientific does not guide quarterly revenue, EPS, gross margin, or adjusted EBITDA. The live comment into October 23 is another meaningful step-up in colocation revenue as more capacity bills. The only numeric near-term checkpoint that can be scored, more than 450 MW billable by the end of summer 2026, was set on the May 6 call and was already at 437 MW by mid-July, with four of five CoreWeave campuses described as substantially complete ahead of that timeline. Classification: conservative on delivery timing, not guided on the P&L. The summer print is still open until this quarter closes. A team that is early on megawatts and was late on the second contract is not guiding the income statement aggressively.

The 80% to 85% CoreWeave figure was raised on May 6 from 75% to 80% and repeated on the July 28 deck. It is a cash operating margin on facilities, labor, insurance, and taxes at scale. It is not this quarter's GAAP gross margin of 59%. Cash SG&A is explicitly not guidance. Low $30 million per quarter was called a reasonable baseline, and Q2 ran above it for a stated one-time reason. The May capex envelope of roughly $2 billion assumed no new customer. AMD broke that condition. The replacement language is $11 million to $12 million per MW and about $6 billion for the first roughly 530 AMD megawatts, expected to be project-financed. No new full-year dollar was issued. FY capex is not guided in any way that should be underwritten. September conference color put 2027 deliveries at $12 million to $13 million per MW. That is cost inflation on the outer build, mostly labor and electrical gear, not a Q3 revenue item. Tariffs were framed back in May 2025 as a 5% to 10% capex headwind that would show up in higher lease rates. Nothing since July 28 has turned that into a quarterly P&L range.

FMP quarterly analyst-estimates has no row for September 30 or December 31. The Q3 figures are the earnings calendar row. Annual EBITDA consensus of −$1,378 million, with a low-to-high of −$2,787 million to +$289 million, is not a usable bar and is not used as one.

First-half 2026 revenue is $115.244 million plus $164.201 million, or $279.4 million. The FY2026 FMP revenue mean of $686.4 million leaves about $407 million for the second half. After the Q3 estimate, the residual for Q4 is about $208 million. That residual is street arithmetic. It is not an FMP Q4 row and it is not guidance. FY2026 EPS of −$4.48 is almost entirely the first half already reported, −$1.06 and −$3.32. It is the warrant mark, not a view that the second half loses several more dollars.

Guide columns are Not Guided except the forward build-cost row, which is a delivery cost and not Q3 revenue. Consensus is FMP. Internal is n/a on every line. There is no model in this run. Percent difference versus a guide midpoint is n/a because there is no midpoint.

Current quarter — Q3 2026

The build-cost row is forward delivery for the AMD sites. First AMD megawatts are the first half of 2027. It is not Q3 revenue. Year-ago drivers, so the missing guide can still be compared with a print: colocation $14.951 million, self-mining $57.438 million, hosted mining $8.714 million, total $81.103 million, gross profit $3.905 million, colocation gross margin 26%, operating loss −$57.425 million, adjusted EBITDA −$2.432 million.

FQ+1 — Q4 2026

The Q2 call did not give a Q4 revenue, EPS, margin, or megawatt range. Mining was described as winding down through year-end. Dalton Phase 2, the last CoreWeave campus, was left on an early-2027 completion. FMP has no Q4 2026 revenue or EPS row. The roughly $208 million residual is not entered as consensus.

FY2026 still on the books, and FY2027

No formal FY2026 or FY2027 revenue, EPS, or margin range was given on the May 6 call or the July 28 call. The June 2025 merger-proxy projections are not a live guide. The CoreWeave transaction was terminated, and management did not reaffirm those figures. FY2026 is still a CoreWeave billing year. AMD shows up in the FY2027 mean, not in this print. About 265 MW of the initial AMD book is targeted for 2027 and the rest for 2028.

Eight reported quarters

Q3 2026 is blank on purpose. Growth rates are versus the same quarter a year earlier. Earlier year-over-year cells are blank where the compare was not scored in this pull. Gross margin and colocation margin are the reported percent series. Adjusted EBITDA is the Daloopa series, which matches the release at Q2 2026 and does not match the release's Q1 2026 or Q2 2025 columns. Hosted-mining, colocation-margin, and adjusted-EBITDA year-over-year rows use those same cited levels. An EBITDA cell is n/m where the sign changed, so a percent would not be a growth rate.

License fees, not power pass-through, are the economic colocation line. In Q2 2026 license fees were $98.8 million of the $136.669 million, up 1,310% year over year from $7.0 million. Power pass-through was $35.1 million, about a quarter of colocation revenue, and should not be read as incremental margin. Colocation gross profit was $79.983 million.

Annual trajectory

Daloopa annual history in this pull starts in 2022. A fifth year was not returned. Total-revenue growth went from −21.5% in 2023 to +1.6% in 2024 and then −37.5% in 2025. The decline is the mining book shrinking faster than colocation scaled. The 2026 quarterly pattern is the reverse. Colocation is now large enough to lift the total. FY2025 mix was still mining-heavy: self-mining $229.207 million, colocation $65.424 million, hosted mining $24.388 million.

The tone has moved from "we still have to prove a second customer" to "the second customer is signed, and the next 2 GW is ours unless the market breaks." The operating message for this quarter is narrower than that rhetoric. Confidence is high on construction and, as of July 28, newly high on having a second contract. It is deliberately low on any quarterly P&L number. That split has been stable. What changed is whether the second contract existed.

On February 26, 2025, the priority was diversification: CoreWeave below half of critical IT load by the end of 2028. Prospects were doing more due diligence than in the first half of 2024, and 2025 critical IT delivery had been pushed to about 250 MW. On May 7, 2025, the message was still proof of execution. First Denton megawatts were the credibility event. Tariffs were a capex headwind to be recovered in lease rates, not a reason to cut the build. On March 2, 2026, Adam Sullivan scored himself against four promises from the October 30, 2025 update and did not clear the customer one. About 500 MW was under exclusivity. Jim Nygaard called 2025 a mining year and put the colocation inflection in "the coming months." That was the defensive call. On May 6, 2026, Sullivan said he was very confident versus three months earlier because more capital was in the ground. Nygaard declared the inflection reached, raised the cash-margin target, and set the summer megawatt checkpoint. The logo was still missing. Exclusivity had expired.

July 28 is the tone shift that matters for this print. The missing logo arrived as AMD: about 530 MW firm on 15-year agreements, 2.5% annual escalators, more than $14 billion of potential base contracted revenue on the company's figures, plus a reservation up to about 2.5 GW of leasable capacity. Total leased customer power went to about 1.1 GW. Nygaard's Q3 line was a step-up, not a range, and the mid-July megawatts were ahead of expectations. The May comment of 243 MW plus another 200 MW in the coming months lands almost exactly on 437 MW. That operating promise was kept. The October customer promise was kept a quarter late and larger than the March exclusivity sketch. What was not provided, again, was a dollar guide.

The overconfidence risk showed up after the quarter. Russell Cann on September 14 said the next deal is the 2,000 MW AMD option and that AMD will take it unless the market goes south. The 8-K reservation is conditional through December 28, 2028. That is the line the street can choose not to underwrite. It is not a Q3 earnings item. C-suite is stable into this print. Sullivan, Nygaard, Brown, and Charbonneau hosted both 2026 calls. Nygaard joined in 2025, so the CFO track record is still short. Mark Adams joined the board on July 29. That is an addition, not a departure. No guidance withdrawal and no CEO or CFO change.

Assumptions under the commentary, all of which can still be checked on this print:

• GAAP colocation is straight-lined over the CoreWeave lease, so reported dollars run ahead of cash rent early in the term. A step-up can be real billing and still not equal cash.
• The Q3 step-up assumes CoreWeave keeps accepting halls and that the mid-July 437 MW stays in billing. It does not assume a new customer, a change in the 80% to 85% target, or any AMD revenue.
• The Q2 10-Q still said colocation revenue was entirely CoreWeave. Contracted capacity is no longer single-customer. Recognized revenue is, until AMD bills. That same 10-Q puts CoreWeave at about 77% of first-half revenue. There is no Daloopa id for the 77% figure.
• Mining is assumed to keep shrinking. Q2 self-mining gross profit was −$12.2 million. A bitcoin move that keeps more miners online would add revenue the street may not be modeling. A faster shutdown would subtract it. Either is second-order next to colocation.
• Financing shifted from "the client pays the capex" on the original CoreWeave deal to "we fund the AMD build." The August credit facility is the corporate piece of that assumption. It does not change the megawatt schedule.

What would make the conservative delivery read wrong is a stall below the summer target, so the step-up is only the July run-rate and mining falls faster than colocation rises. What would make $198.5 million look easy is an exit well above 450 MW. The print to watch is billable megawatts and colocation revenue versus $14.951 million a year ago, not EPS.

The catalyst that can move this print is not a new customer. It is whether CoreWeave megawatts that were already billing by mid-July show up as a full quarter of GAAP colocation, and whether any further CoreWeave megawatts commenced before September 30. No company release between July 28 and September 24 updated billable megawatts past 437 MW. The October print is the next disclosure. Other public aggregators that day were about $188 million to $195 million of revenue. They are not used. FMP at $198.5 million is the high end of that band, which is another way of saying the primary bar is full.

CoreWeave, the tenant, raised its own year-end active-power target on August 11 to more than 1.85 GW and later described short-dated compute contracts. That is GPU-cloud pricing, not this lease rate. The $635 million on 437 MW is about $1.5 million per megawatt-year of infrastructure. It is evidence the tenant's customers are still taking capacity. It does not add a Core Scientific megawatt in Q3. The competitive set on the company's own slide is leasing power to the same buyer set. Nothing in that set changes Q3 revenue. It is why a slip in the 2027 dates would matter.

What changes the read on the night: billable megawatts at or above about 450, with colocation above the roughly $159 million spot, confirms the step-up inside $198.5 million. A hold at 437 MW with colocation closer to $150 million puts the FMP revenue estimate at risk even with a clean execution story. A GAAP colocation margin well below 59%, or a walk-back of 80% to 85% cash, says the ramp is less profitable than the last two quarters. Another large GAAP loss with revenue in line misses −$0.01 without saying anything new about colocation. A restricted-cash release well short of about $300 million reopens the funding question against Q2 cash capex of −$565.0 million.

Window is July 28 through September 24, after the fiscal Q2 report and before this print. The earnings release, the call, and the 10-Q results are excluded. Nothing in the window is a revised guide, a new billing-megawatt figure, or a reported Q3 number. The pattern is a second customer and more power, not a change in what is billing now.

Left out on purpose: same-day sell-side target changes, routine Form 4 activity under a December 2025 10b5-1 plan, small director open-market buys, and a second-hand sector initiation. None of those is a company action that resets the quarter. The FMP snapshot, revenue about $198.5 million and EPS about −$0.01, is untouched by the news. The incremental information is about 2027 and about the balance sheet.

Mixed, and the two lines are telling different stories. On the 11 quarters inside the last 12 that have an FMP estimate, revenue beat 4 of 11 and diluted EPS beat 4 of 11. The last four are 1 of 4 on each line. The revenue surprise has already turned. Seven misses in a row, from Q3 2024 through Q1 2026, broke in Q2 2026 at +16.4% ($164.201 million versus a $141.1 million estimate). GAAP EPS did the opposite. The last two dollar misses are −$1.04 and −$3.26. The Q1 miss lines up with an impairment of property, plant and equipment of $266.5 million. The warrant bridge that quarter was only $30.8 million. The company says the Q2 loss was the warrant liability, not the colocation ramp.

The count says another miss. The magnitude says the miss regime already ended. Surprises went −33.0%, then −1.5%, then +16.4%. The street spent 2025 too high on mining. It was slightly high in Q1 2026 and low in Q2 2026, which is what a late mark-up of colocation looks like. Q3 can still miss $198.5 million if billing stalls near 437 MW and the mining step-down is larger than the model. It should not be scored as if the company were still the seven-quarter revenue misser. The EPS base rate should not be carried forward as an operating call. The Q3 estimate is again about zero. On the reported percent series in the table, gross margin expanded versus the year-ago quarter for three prints: +2,100 bp in Q4 2025, +1,600 bp in Q1 2026, and +3,700 bp in Q2 2026. The basis-point gain narrowed, then widened on the last print. Q3 2025 is not a fourth point, because the year-ago reported percent is not in the pull. That reported series is the line that has been improving. GAAP diluted EPS is the line that has been getting noisier.

There is no adjusted-EPS series in Daloopa, and the company does not put an adjusted EPS on the release. A beat is actual above the FMP estimate. Revenue surprise is (actual − estimate) / estimate. EPS surprise is shown in dollars, because the estimate sits near zero and a percent is not a magnitude. Four quarters disagree between FMP epsActual and the face of the release. The heatmap uses the face. The one that changes the call is Q4 2025: the March 2, 2026 release prints diluted EPS of $0.42 and net income of $216.0 million. FMP stores −$0.42. Scoring the FMP sign would turn a beat into a miss. It is not used. Q3 2023 has actuals and no estimate, so it is on the grid and out of the rate. Q3 2026 is not on the grid.

Green is a beat, red is a miss, yellow is a near-zero surprise, gray is no FMP estimate. Filled cells are the larger surprises. The Q2 2026 revenue cell is green. The Q2 2026 EPS cell is red. Those two cells are the setup. Read left to right. Revenue is green through Q2 2024, red for seven quarters, pale in Q1 2026, and green again in Q2 2026. The two largest revenue misses are Q3 2025 and Q4 2025, when the street was still modeling a mining base and colocation was not yet most of revenue. EPS does not follow that path. Its largest misses are Q2 2024 and Q2 2026, both large losses against a near-zero estimate, with a large Q1 2025 beat in between that was a warrant gain, not an operating beat. Q1 2025 adjusted EBITDA was −$6.071 million in the same quarter diluted EPS was $1.25.

Average revenue surprise was +7.7% in the first four scored quarters, −11.4% in the middle four, and −11.4% in the last four. The last-four average equals the middle four only because −27.6% and −33.0% are averaged with −1.5% and +16.4%. The path inside the last four is the fact that matters. Q2 was the largest dollar revenue beat since Q1 2024. Average EPS dollar surprise went from +$0.12 in the middle four to −$1.03 in the last four. Using the EPS beat rate as a read on colocation execution runs the sign the wrong way.

Q2 2026 is why the streak should not be read as "they miss everything." Revenue was +16.4% and +108.8% versus the year-ago quarter. Colocation was +1,194%. Gross profit of $70.044 million is the margin expansion. Adjusted EBITDA of $41.103 million is the operating result. The EPS miss got worse because the warrant liability got larger. The revenue beat got better because billed colocation got larger.