Built by Grok 4.6 High

CRWV | Earnings Review — Q2 2026

Verdict: HIGH-PLATEAU VOLUME, P&L STILL COMPRESSING. Revenue YoY has stopped decelerating — three straight quarters at ~110–112% after a 2025 scale-down from +420% — while the capacity engine re-accelerated (active power 1,500 MW, +219% YoY) and adj. EBITDA dollars doubled to $1,510 million / 59%. Gross margin is the clear decelerating leg (65.9%, −836 bps YoY). GAAP EPS remains a loss at −$1.14 as interest hit $640 million.

Print (small revenue beat, real adj. OI beat): Revenue $2,575 million vs LSEG $2.56 billion (+$15 million / +0.6%) and vs the Q1-issued Q2 band of $2.45 billion–$2.60 billion (in range, $25 million under the high). Adj. operating income $128 million cleared the $90 million high by +$38 million / +42%. CapEx $9,352 million was slightly above the $7.0–$9.0 billion band. Street was already inside the company’s band; clearing LSEG by +0.6% is a residual, not a demand snap.

Guidance (first FY2026 raise — volume/CapEx, not profit): FY revenue mid $12.4–$13.2 billion (+$0.30 billion / +2.4% at the mid) vs CapEx mid $35–$39 billion (+$4.0 billion / +12%). Q3 revenue $3.45–$3.60 billion is +158% YoY at the mid vs Q3’25 $1,365 million — the inflection they have been promising — but Q3 interest $860–$940 million is ~4× adj. OI.

Tone: Offensive vs Q1’s sandbag. “Outperformed our plan across the board.” Q1 was a bookings quarter they refused to translate into a 2026 revenue raise because 2026 was “sold out.” Q2 is the quarter they finally did, because the megawatts showed up (and because they put a ~25% July list-price increase through).

Contradictions (4 live, 2 high): (1) High — sold-out 2026 was the May reason they would not raise FY revenue; in August they raised 2026 MW, CapEx and revenue anyway. (2) High — “insulated / POs in hand” vs CapEx raised for component pricing with FY revenue unchanged. (3) Inference economics “identical” vs 25% list hike as customers “shift to inference.” (4) “Each new build is a smaller part of the whole” vs the largest MW add on record (+50% of base).

Catalysts: Q3 $3.45–3.60B (above ~$3.42B Street) is a delivery print, not a demand print; >$25B early-Q3 commitments into the $104.2 billion backlog; adj. OI vs $860–940M of interest; Fully Connected 2026 (Sep 29–Oct 1).

Capacity, backlog, mix — above the P&L

CoreWeave sells GPU-hours against take-or-pay contracts. Active power is the volume driver; contracted power and backlog are the leading indicators; Customer A is concentration.

| Metric | 2024Q1 | 2024Q2 | 2024Q3 | 2024Q4 | 2025Q1 | 2025Q2 | 2025Q3 | 2025Q4 | 2026Q1 | 2026Q2 | |---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:| | Active power (MW) | — | — | — | 360 | 420 | 470 | 590 | 850 | 1,000 | 1,500 | | YoY | — | — | — | — | — | — | — | +136% | +138% | +219% | | Contracted power (GW) | — | — | — | 1.3 | 1.6 | 2.2 | 2.9 | 3.1 | 3.5 | 3.7 | | YoY | — | — | — | — | — | — | — | +138% | +119% | +68% | | Revenue backlog ($B) | 15.9 | 16.2 | 15.0 | 15.1 | 25.9 | 30.1 | 55.6 | 66.8 | 99.4 | 104.2 | | YoY | — | — | — | — | +63% | +86% | +271% | +342% | +284% | +246% | | Customer A % of rev | 33% | 59% | 65% | 62%* | 72% | 71% | 67% | 67%* | 45% | 36% | | CapEx ($M) | 1,114 | 2,574 | 2,237 | 2,410 | 1,858 | 2,937 | 1,850 | 8,241 | 6,786 | 9,352 | | Revenue ($M) | 188.7 | 395.4 | 583.9 | 747.4 | 981.6 | 1,212.8 | 1,364.7 | 1,572 | 2,078 | 2,575 | | Rev YoY | — | — | — | — | +420% | +207% | +134% | +110% | +112% | +112% |

*2024Q4 and 2025Q4 Customer A % shown as the FY figure. Call: contracted power 4.2 GW as of 11 Aug (not in the Q2 Daloopa print); ~300 MW of the +500 MW Q2 add landed in June.

Driver read: This is a capacity-conversion quarter, not a demand quarter. Active power jumped 1,000 MW → 1,500 MW — YoY accelerating to +219% — so Q2 revenue still under-earns the exit run-rate. Contracted-power YoY is the opposite: slowing to +68% as the installed base scales. Backlog YoY peaked at +342% in 2025Q4 and is rolling over from a very high base. Mix is improving: Customer A fell from 71% to 36% YoY (Customer B 26% + C 10% = 72% top-3). CapEx at $9.4 billion is the tell that H2 volume is already bought.

Consolidated profitability

| Metric | 2024Q1 | 2024Q2 | 2024Q3 | 2024Q4 | 2025Q1 | 2025Q2 | 2025Q3 | 2025Q4 | 2026Q1 | 2026Q2 | |---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:| | Revenue ($M) | 189 | 395 | 584 | 747 | 982 | 1,213 | 1,365 | 1,572 | 2,078 | 2,575 | | Gross margin % (calc.) | 68.6 | 72.5 | 75.5 | 75.6 | 73.3 | 74.2 | 73.0 | 67.6 | 65.5 | 65.9 | | GM YoY bps | — | — | — | — | +466 | +175 | −252 | −801 | −773 | −836 | | Adj. EBITDA ($M) | 105 | 250 | 379 | 486 | 606 | 753 | 838 | 898 | 1,157 | 1,510 | | Adj. EBITDA margin % | 55 | 63 | 65 | 65 | 62 | 62 | 61 | 57 | 56 | 59 | | EBITDA margin YoY bps | — | — | — | — | +700 | −100 | −400 | −800 | −600 | −300 | | Adj. OI ($M) | 25 | 85 | 125 | 121 | 163 | 200 | 217 | 88 | 21 | 128 | | Adj. OI margin % | 13 | 22 | 21 | 16 | 17 | 16 | 16 | 6 | 1 | 5 | | OI margin YoY bps | — | — | — | — | +400 | −600 | −500 | −1,000 | −1,600 | −1,100 | | Interest, net ($M) | 41 | 67 | 104 | 149 | 264 | 267 | 311 | 388 | 536 | 640 | | Diluted EPS ($) | n/a | n/a | n/a | n/a | −1.49 | −0.60 | −0.22 | −0.89 | −1.40 | −1.14 |

Gross margin (calc.) = (Revenue − COGS) / Revenue. Interest stored negative in Daloopa; table shows the expense as a positive dollar cost. 2024 quarterly EPS is not on the post-IPO share base.

Two clocks. Gross margin peaked at 75.6% in 2024Q4 and has compressed ~800 bps YoY for three quarters — structural mix/depreciation, not noise. Adj. EBITDA margin troughing at 56% in 2026Q1 then inflected to 59% (−300 bps YoY, the smallest hit in four quarters). Adj. OI $128 million / 5% is a sequential recovery from $21 million / 1% and beat the $30–90M guide, but it is still −36% YoY and −1,100 bps vs last year’s 16%. Interest is the EPS killer: $640 million is 25% of sales and 5× adj. OI.

Revenue & adj. EBITDA — absolute levels ($M)

YoY growth trajectory

| Quarter | Revenue YoY | Backlog YoY | Active power YoY | Adj. EBITDA YoY | EBITDA margin YoY | Gross margin YoY | Diluted EPS YoY | |---|---:|---:|---:|---:|---:|---:|---:| | 2025Q1 | +420% | +63% | — | +480% | +700 bps | +466 bps | n/a | | 2025Q2 | +207% | +86% | — | +201% | −100 bps | +175 bps | n/a | | 2025Q3 | +134% | +271% | — | +121% | −400 bps | −252 bps | n/a | | 2025Q4 | +110% | +342% | +136% | +85% | −800 bps | −801 bps | n/a | | 2026Q1 | +112% | +284% | +138% | +91% | −600 bps | −773 bps | +6% | | 2026Q2 | +112% | +246% | +219% | +100% | −300 bps | −836 bps | −90% |

Annual arc (FY2022–FY2025 + LTM)

| Metric | FY2022 | FY2023 | FY2024 | FY2025 | LTM (Q3'25–Q2'26) | |---|---:|---:|---:|---:|---:| | Revenue ($M) | 16 | 229 | 1,915 | 5,131 | 7,590 (calc.) | | Rev YoY | — | +1,346% | +737% | +168% | — | | Adj. EBITDA margin % | −62 | 45 | 64 | 60 | 58.0 (calc.) | | Adj. OI margin % | −135 | 0.3 | 19 | 13 | 6.0 (calc.) | | Interest, net ($M) | 9 | 28 | 361 | 1,229 | 1,875 (calc.) | | Customer A % | 16 | 35 | 62 | 67 | 36 (spot) |

Still a hyper-scale-up, not a mature compounder. Annual YoY decelerated 1,346% → 737% → 168% as the base grew (arithmetic, not demand failure). FY2026 guide mid $12.80B is +149.5% vs FY2025 — still a deceleration at the annual grain even if Q3 quarterly YoY re-accelerates.

Tiny residual revenue beat vs a Street book already at the high end of the company’s own guide; the real beat is adjusted operating income. Pattern: consistent revenue beater, mixed EPS. Revenue-beat magnitude is deteriorating.

Heatmap — last 8 quarters

This quarter vs consensus / own guide

| Metric | Bar | Actual | Variance | Result | |---|---:|---:|---:|---| | Revenue vs LSEG | $2.56B | $2,575 million | +$15M / +0.6% | BEAT | | Co. Q2 revenue guide | $2.45–2.60B (mid $2.525B) | $2,575 million | +$50M vs mid; −$25M vs high | In range / high end | | Adj. diluted EPS | $(1.20) LSEG | $(1.03) (calc. $567M / 551M sh.) | +$0.17 / +14% | BEAT | | GAAP diluted EPS | n/a (LSEG prints adj.) | $(1.14) | — | Not comparable to Street $(1.20) | | Adj. OI vs co. guide | $30–90M | $128 million | +$38M / +42% vs high | BEAT (above high) | | Interest vs co. guide | $650–730M | $640 million | $10M below low | BEAT (lower interest) | | CapEx vs co. guide | $7.0–9.0B | $9,352 million | +$352M / +3.9% vs high | MISS (volume-driven) |

Beat rates

| Window | Revenue beat rate | Adj. EPS beat rate | Usable? | |---|---|---|---| | L12Q ticker | 6/6 on the public subset; 6 of 12 n/a | 1/3 where LSEG published adj. EPS | No as a 12-quarter rate — IPO Mar 2025 | | L6Q public | 6/6 (100%) | 1/3 (33%) on three comparable prints | Yes for revenue | | L4Q | 4/4 (100%) | 1/3 (33%) | Best available revenue window |

Pattern: Mixed. Not a misser. Not a franchise that crushes Street every line. Top line always clears LSEG; typical recent surprise is +0.6% to +5.5%. This print is the smallest LSEG beat in the public history. Adj. EPS L3Q is 1/3 (Q4’25 M −14%, Q1’26 M −24%, this quarter B +14%). The line that is improving vs own guide is adj. OI — this is the largest % beat vs the high (+42%).

Mgmt variance story: never says “vs Street.” They explain against their own plan. Intrator: “outperformed our plan across the board.” Honest attribution: (a) high-end-of-guide revenue against a Street book that was already there, plus (b) adj. OI and interest both beating their own bands. Agrawal: Q2 added ~500 MW, ~300 MW in June alone — capacity too late to fill a full quarter of revenue, which is why OI cleared the high while revenue only tagged it. CapEx miss framed as “customer deliveries accelerating.”

Headline: First FY2026 raise of the year, and it is a volume/CapEx raise more than a profit raise. Revenue mid +$0.30B (+2.4%) vs CapEx mid +$4.0B (+12%). Q3 is the inflection they have been promising since Q4’25: revenue mid $3.525B is +158% YoY, adj. OI margin guided to ~6.5% after Q1 trough 1% and Q2 5%, with Q4 “low teens.”

New official ranges (issued 2026-08-11)

| Metric | New low | New high | New mid | vs prior (Q1 issue) | |---|---:|---:|---:|---| | Q3’26 revenue | $3.45B | $3.60B | $3.525B | First issuance | | Q3’26 adj. OI | $200M | $260M | $230M | First issuance | | Q3’26 interest | $860M | $940M | $900M | First issuance | | Q3’26 CapEx | $11.5B | $13.5B | $12.5B | First issuance | | FY2026 revenue | $12.4B | $13.2B | $12.80B | Was $12.0–13.0B — raised both ends | | FY2026 adj. OI | $960M | $1,150M | $1,055M | Was $900–1,100M — raised both ends | | FY2026 CapEx | $35B | $39B | $37.0B | Was $31–35B — raised both ends | | FY2026 exit ARR | $18.5B | $19.5B | $19.0B | Was $18–19B — raised both ends |

Implied Q3 adj. OI margin (derived): $230M / $3.525B = 6.5%. Implied FY adj. OI margin: $1,055M / $12.80B = 8.2% vs FY2025 13% — still −480 bps YoY. The “inflection” is sequential, not yet year-on-year. EPS, gross margin, and adj. EBITDA are not guided. YE active power raised to >1.85 GW (from >1.7 GW).

Waterfall — FY2026 raise quality

Implied YoY at the new guide

| Metric | Q3’25 actual | Q3’26 mid | Implied YoY | FY’25 actual | FY’26 mid | Implied YoY | |---|---:|---:|---:|---:|---:|---:| | Revenue | $1,365M | $3,525M | +158% | $5,131M | $12,800M | +149.5% | | Adj. OI | $217M | $230M | +5.9% | $666M | $1,055M | +58% | | Adj. OI margin | 16% | 6.5% (der.) | −950 bps | 13% | 8.2% (der.) | −480 bps | | Interest | $311M | $900M | +190% | $1,229M | not guided | — | | CapEx | $1,850M | $12,500M | +576% | $14,886M | $37,000M | +149% |

Q3 is where revenue re-accelerates (from ~+112% in Q1/Q2 to +158%) while adj. OI is barely up YoY because last year’s 16% margin is being replaced by a 6.5% ramp-cost print. Interest at $900M mid is 3.9× adj. OI — GAAP stays a loss even if they hit the OI guide. Guide low on adj. OI ($200M) is a YoY miss vs $217 million.

H1 tracking: H1 revenue $4.653B is 36.4% of FY mid; H1 adj. OI $149M is 14.1% of FY mid; H1 CapEx $16.14B is 43.6%. Implied Q4 at FY mid: revenue ~$4.62B (+194% YoY vs Q4’25 $1,572 million); adj. OI ~$676M (margin 14.6%, which is exactly “low teens”). The FY raise is Q4-loaded. Missing the power-delivery schedule in Q3/Q4 is how this guide breaks.

Other forward statements (Q2 call)

| Item | Q2 statement | |---|---| | YE 2026 active power | >1.85 GW (was >1.7 GW) | | Contracted power | 3.7 GW at Q2-end; 4.2 GW as of the call; plus ~1.5 GW options/LOIs → ~6 GW visibility vs ≥8 GW by 2030 | | Backlog | $104.2B; >$25B early-Q3 commitments not in that figure; >50% already in delivery, >2/3 by YE | | Exit-2027 ARR | Not restated. Last stated at Q1: >$30B, >75% contracted. Daloopa still stores ~$30B from Q4’25 | | New-contract contribution | +5 to 10 ppt vs recent-quarter adds; a lot of it Vera Rubin | | List pricing | ~25% increase across SKUs in July + component pass-through | | Managed inference ARR | $1M → >$100M; exit-2026 ≥$250M | | Non-GPU ARR | Storage/CPU/networking/software >$400M | | Long-term adj. OI margin | Daloopa still 25–30% from Q4’25; not restated this call |

Tone: Q2 vs Q1

| Dimension | Q1’26 (May 7) | Q2’26 (Aug 11) | Direction | |---|---|---|---| | Self-grade | “Transformational” on bookings. Revenue beat the high; they reaffirmed FY $12–13B. | “Exceptional.” “Outperformed our plan across the board.” First time they raise FY revenue, adj. OI, CapEx, exit ARR, and YE power together. | More confident, less sandbagged | | Margin | Q1 = “trough.” Sequential expansion; “inflect as we cross from Q2 to Q3.” | “Q2 marked the quarter in which we saw margins inflect expanding sequentially.” | Promise kept on the sequential step; still far from 16% adj. OI in Q2’25 | | Sold-out | “Largely sold out of 2026” — the reason FY was not raised | Same constraint, sharper; then they raised 2026 MW and revenue anyway | See Contradictions C1 | | Pricing | “Prices increasing” — no % | Quantified ~25% SKU increase in July | From qualitative to a number | | CapEx | Low-end raised $1B for component inflation | Mid +$4B, framed as accelerating deliveries and recent wins | Less defensive, more “we are choosing to spend” |

How the guide breaks (in order): (1) Power/delivery slip in Q3/Q4 — FY mid implies Q4 ~$4.62B. (2) Interest outruns adj. OI — Q3 $900M vs $230M. (3) Gross-margin compression does not stop — derived GM 65.9% vs 74.2% a year ago, not guided. (4) CapEx prints above the high again without a matching revenue beat.

8-quarter revenue & EPS trajectory. Accel = change in the YoY rate vs the prior quarter’s YoY rate, in basis points (second derivative — not QoQ revenue). Daloopa has no 2023 quarterly series, so 2024Q3/Q4 YoY are n/a.

| Metric | 2024Q3 | 2024Q4 | 2025Q1 | 2025Q2 | 2025Q3 | 2025Q4 | 2026Q1 | 2026Q2 | |---|---:|---:|---:|---:|---:|---:|---:|---:| | Revenue YoY % | n/a | n/a | +420.3 | +206.7 | +133.7 | +110.3 | +111.7 | +112.3 | | Rev Accel (bps) | n/a | n/a | n/a | −21,350 | −7,305 | −2,338 | +137 | +63 | | EPS YoY % | n/a | n/a | −140.3† | +63.0† | +87.9† | n/a | +6.0 | −90.0 | | Revenue ($M) | 584 | 747 | 982 | 1,213 | 1,365 | 1,572 | 2,078 | 2,575 | | Diluted EPS ($) | −1.82 | n/a | −1.49 | −0.60 | −0.22 | −0.89 | −1.40 | −1.14 |

†EPS YoY through 2025Q3 is distorted by the March 2025 IPO share-count step-up (diluted shares 249 million → 487 million). Treat as noisy, not operating. NI to common is the cleaner earnings trajectory: four of the last four quarters show YoY widening of the GAAP loss.

Inflection map

| Quarter | Signal | What happened | |---|---|---| | 2025Q1 | Revenue YoY peak | First public print: $982 million vs $189 million = +420%. Easy-comp peak, not a new acceleration. | | 2025Q2 | Primary deceleration | YoY halved to +207% (−21,350 bps). Largest Accel move in the window — pure base math. | | 2025Q3–Q4 | Cascade, then trough | +134% then +110% — slowest YoY of the post-IPO window. Absolute dollars still stair-stepping up every quarter. | | 2026Q1 | Deceleration stops | +112% (+137 bps) — first positive Accel of the public window. Rate stopped falling; it has not re-accelerated in any economically meaningful way. | | 2026Q2 | High-level hold / capacity conversion | +112% (+63 bps) on a record $2,575 million. Two quarters at ~112% after a four-quarter slide from 420% → 110%. Not a re-acceleration of the franchise; a plateau at still-extraordinary triple-digit growth. EPS YoY −90%: loss −$1.14 vs −$0.60. |

Net verdict: Revenue is in a high-plateau after a completed deceleration. That is a better setup than a still-falling rate. The Q3 guide encodes a large re-acceleration (+153% to +164% YoY, +4,050 to +5,150 bps Accel) if the June megawatts convert on time. Miss that guide and the plateau is the story. EPS/NI is deteriorating because the capital structure required to deliver that revenue (debt to fund $9.4B of Q2 CapEx) is consuming the operating profit. Do not read EPS Accel as a franchise signal until net interest stops growing faster than EBITDA.

Key drivers (Q2 call)

1. Capacity conversion, back-end loaded — ~500 MW in Q2, ~300 MW in June alone. That is why sequential revenue did not match the power add (Wells Fargo’s question). Q3 guide is the conversion test.
2. Demand still exceeds supply; pricing moving up — “Near-term capacity remains effectively sold out.” July ~25% list-price increase; new contracts 5–10 ppt above recent cohorts; A100 recontracted into 2029.
3. Backlog is the leading indicator — $104.2 billion (+246% YoY), excluding >$25 billion of early-Q3 commitments.
4. Mix into software / inference — Managed inference booked ARR $1M → >$100M; YE target ≥$250M. Storage/CPU/networking/software already >$400M ARR.
5. Customer diversification is real, concentration is not gone — Customer A 71% → 36%; B and C now ≥10%. Top-3 still 72%. Existing-customer expansion still 93%.
6. The EPS drag is interest on a scaling debt stack — Q2 interest $640 million vs $267 million; Q3 guide $860–940M. Total debt $31.4 billion vs $11.1 billion (+184% YoY). WACC of debt −~300 bp over the past year (~$1.1B annualized savings) does not change the dollar growth.

What consensus believed for this quarter vs what printed, and what moves the tape into November.

| # | Catalyst | Timing | Street bar | Management language | Read | |---|---|---|---|---|---| | 1 | Q3 print + FY bar | ~2026-11-16 | Pre-print TheFly Q3 ~$3.42B. Company mid $3.525B is ~+3%. Post-print FactSet $3.541B has moved onto the mid. | Q3 $3.45–3.60B; adj. OI $200–260M; interest $860–940M. “Low teens in Q4.” | Highest-conviction near-term setup. A $3.50B print beats Aug 11 TheFly and screens in-line vs post-print FactSet. Watch June’s 300 MW in the run-rate; adj. OI vs $200–260M and vs Q3’25 $217M (guide low is a YoY miss). | | 2 | >$25B early-Q3 commitments | Scored at Q3 print | Street treated $104B as visibility, not a 2026 P&L unlock. Models do not yet have the $25B. | Backlog $104.2B “does not include the over $25 billion of net new customer commitments.” >50% already in delivery; >2/3 by YE. | A Q3 backlog that does not step toward ~$125B+ would say the early-Q3 number was a press-release, not a booking. Only 40% of backlog is inside 24 months. | | 3 | Power ramp: >1.85 GW YE | Q3–Q4 2026 | Into Q2, Street modelled a linear path from 1,000 MW toward >1.7 GW. | YE >1.85 GW. June 300 MW “come through in Q3 and Q4.” | Physical conversion catalyst. An in-range Q3 print with power still “sold out” confirms the June lag; a miss with 1.5 GW already on would say utilization / mix, not MW, is the constraint. | | 4 | July ~25% list + 5–10 ppt new deals | Mix into 2H ASPs | Street had no explicit ASP series. | “Approximately 25% increase across SKUs.” New Q2 contracts 5–10 ppt above recent. “A lot of that is coming in, in the Vera Rubin SKU.” | List +25% on a sold-out stack is not realized ASP +25% — 5-year take-or-pay does not reprice mid-term. If Q3 adj. OI margin does not step toward high-single / low-teens, the hike is stuck in backlog. | | 5 | Managed inference ARR | Intra-2026 | No inference line in models. $250M is ~1.3% of exit ARR mid $19B. | $1M → >$100M; YE ≥$250M. Growth “constrained only by our near-term capacity.” | Re-rating option, not a 2026 EPS driver. Bull: Hopper/Ampere roll-off feeds a higher-ASP product DDTL 5.5 will now finance. Bear: rounding error next to $9.4B of quarterly CapEx. | | 6 | Vera Rubin NVL72 | H2 2026 | Street wanted first-to-rack proof (GB200 analog) and that Rubin ASPs/margins are higher. | “First cloud provider to bring up and validate NVIDIA’s Vera Rubin NVL72.” Margin expansion “right from the start.” | Generation-cycle catalyst. Q3 mix comments that do not mention Rubin revenue would push the 5–10 ppt to 2027. Hudson River Trading (8/20 IR) is the first named post-print Rubin offtake. | | 7 | Customer A/B/C mix | 10-Q each quarter | Street’s structural overhang: Microsoft-proxy 50%+ book. | A 36%, B 26%, C 10%. Named wins: Caterpillar, Isomorphic, Grammarly, Leidos. | Highest-quality fundamental catalyst in the print. Residual: top-3 still 72%; 93% still expansion of the same book. Next test: does A stay ≤40% as Rubin ramps? | | 8 | DDTL 5.5 / cost of capital | Closed 2026-08-10 ($2.6B) | July dislocation (spread +125 bp) was the risk. This deal SOFR+550 vs May DDTL 5.0 SOFR+450 is +100 bp. | “First to include shorter duration customer contracts.” Priced in “one of the most dislocated weeks for credit this year”; still took full size. | Unlock: 2–3 year enterprise and inference now financeable. Cost: this specific deal is wider than May, and Q3 interest ~$900M is ~3.3× Q3 adj. OI mid. | | 9 | Fully Connected 2026 | 2026-09-29 – 10-01, Moscone South | No numeric consensus. First post-print product podium before the Q3 call. | “You will hear from our leadership and customers alike.” | Highest-leverage event before November. A Caterpillar / HRT / Grammarly on-stage with utilization or token metrics converts logos into a modelable line. | | 10 | Supply chain / CapEx quality | Intra-year | After Q1, Street haircut margins for component inflation. | “Value of output has outpaced the value of the input increases.” Solidigm LTA. FY CapEx $35–39B. | If Q3 CapEx prints above $13.5B without a MW/revenue beat, inflation is back in the driver’s seat. |

What Q2 closed: sequential adj. OI ($21 million → $128 million) and Customer A mix (71% → 36%). Those were the two questions that made CRWV a “Microsoft GPU pass-through” in 2025. They are less true in the Q2 10-Q.

What Q3 opened: conversion math. June 300 MW has to show up in $3.45–3.60 billion. The $25B has to show up in backlog. Adj. OI has to clear $200 million and last year’s $217 million. Interest will still dwarf AOI. This is a spread business: GPU yield minus (D&A + SOFR+spread).

Call: 2026-08-11 · Michael Intrator (CEO), Nitin Agrawal (CFO) · IR cut the queue after Brett Knoblauch (Cantor).

Tally: 6 Well Answered / 6 Deflected. The Street used Q&A to pressure-test one idea: the print and the raise are real, but the economics of the next layer (recontracting older GPUs, managed inference vs take-or-pay, 5–10 ppt “better” contribution, 2027 ARR) are being described, not sized.

Scoreboard

| # | Analyst | Topic | Badge | |---|---|---|---| | 1a | Samik Chatterjee — J.P. Morgan | Typical recontract term; fleet up for renewal | Deflected | | 1b | Samik Chatterjee — J.P. Morgan | Solidigm and broader long-term supply agreements | Well Answered | | 2a | Brad Zelnick — Deutsche Bank | Managed-inference capacity vs take-or-pay | Deflected | | 2b | Brad Zelnick — Deutsche Bank | Unpack the 5–10 ppt better contribution margin | Deflected | | 3a | Irvin Liu for Amit Daryanani — Evercore | Is $18–19B still the 2027 exit-ARR target? | Deflected | | 3b | Irvin Liu for Amit Daryanani — Evercore | NIMBY; >3 GW by YE27 and 8 GW by 2030 | Well Answered | | 4a | Raimo Lenschow — Barclays | Inference CPU/storage; can existing DCs evolve? | Deflected | | 4b | Raimo Lenschow — Barclays | Edge vs neo-cloud vs hyperscalers (Meta) | Well Answered | | 5a | Michael Turrin — Wells Fargo | Linearity of the 500 MW add; June 300 MW run-rate | Well Answered | | 5b | Michael Turrin — Wells Fargo | Vera Rubin monetization vs prior generations | Well Answered | | 6a | Brett Knoblauch — Cantor | Recontract vs spot vs inference; lead time | Deflected | | 6b | Brett Knoblauch — Cantor | Does DDTL 5.5 change target contract duration? | Well Answered |

Highest-content answers

• Power linearity (Turrin): Q2 added ~500 MW; 300 MW in June alone — more than any prior full quarter. Back-end loaded, “come through in Q3 and Q4.” Binary timing question answered. The guide is the conversion.
• Vera Rubin (Turrin): “Seeing the margin expansion right from the start.” Of the 5–10 ppt step-function, “a lot of that is coming in, in the Vera Rubin SKU.” First cloud to bring up NVL72.
• NIMBY / GW path (Evercore): Moratoriums move where, not whether. Contracted 4.2 GW as of the call plus ~1.5 GW options/LOIs → “close to about 6 GW already.” “None of those numbers will be impacted by the regulatory pushback as of today.”
• DDTL 5.5 (Knoblauch): Not just cheaper Hopper recontract — a new customer set. “Enterprise wants 2- or 3-year contracts, not 5.” Until 5.5, that market was not easily accessible.

Deflections that matter

| Analyst | Asked | Got instead | Why it matters | |---|---|---|---| | Chatterjee (JPM) | Typical recontract term and MW / $ of fleet rolling | A100-to-2029 anecdote; “very limited part of the fleet”; inference YE ARR ~$250M | Without a vintage / roll-off schedule the recontracting “upside” cannot be put in a model. $31.4 billion of debt is underwritten on contracted cash flows; residual value is the equity kicker and it is unquantified. | | Zelnick (DB) | Rules for allocating scarce MW to inference vs 5-year take-or-pay | $1M → $100M ARR anecdote; “control over the silicon” | If inference is still a rounding error of capacity, the mix-shift thesis is 2027+. | | Zelnick (DB) | Split of the 5–10 ppt (duration vs price vs SKU vs attach) | “Difficult to deconstruct” | July ~25% list + component pass-through + Rubin mix + shorter duration are different durability stories. Unbridged to the 5% adj. OI margin (−1,100 bps YoY). | | Liu / Evercore | Confirm 2027 exit ARR still $18–19B | 2026 exit ARR raised to $18.5–19.5B | Year swap. The 2025Q4 medium-term 2027 ARR of ~$30 billion was left untouched. That is the number that decides whether FY CapEx of $35–39 billion earns its cost of capital. | | Lenschow (Barclays) | Can existing DCs take inference’s extra CPU/storage? | “You build AI infrastructure”; “will move seamlessly” | If inference needs different halls, the $11.9 billion CIP and the FY CapEx raise have a mix risk the Street cannot see. | | Knoblauch (Cantor) | Mix of recontract vs spot vs inference, and how many months before roll-off | “One we’re working through” | Second attempt at Chatterjee. Process opacity is now the tell. |

What the next call has to close: (1) MW or $ of 2026–2028 contract expiries; (2) inference as a % of active power, not just ARR; (3) a bridge of the 5–10 ppt claim to reported adj. OI; (4) 2027 ARR, even as a range.

4 genuine incompatibilities across six transcripts (FY2025 Q1 → FY2026 Q2). Labeled guidance raises excluded unless two statements about the same fact cannot both be true.

C1 — High: “sold out of 2026” was why they would not raise FY revenue

C2 — High: “insulated / POs in hand” vs CapEx raised for components

C3 — Medium–High: inference economics “identical” vs the margin catalyst

C4 — Medium: “smaller part of the whole” vs the largest MW add on record

Not carded: labeled FY raises; YE power 1.7 → 1.85 GW (labeled); exit ARR path; Q1 trough / sequential expansion path (timing of the word “inflect” slid one quarter; the path is the same); 5-year vs shorter-duration barbell; WACC “300 bp” recycled across different windows; 107% vs 207% YoY on the Q2’25 call (ASR; Daloopa is 207%).

Q2 is an AI-infrastructure demand and supply-chain call, not a classical macro print. Management did not cite CPI, consumer confidence, unemployment, or a Fed path. Use this call for AI BOM, power/grid, and HPC credit — not for the household economy.

Macro (what is usable)

| Theme | What management said | Read-through | |---|---|---| | Credit / rates | ~$18B raised in Q2; >$32B debt+equity to date. DDTL 5.5 priced in “one of the most dislocated weeks for credit this year” — still took full size. WACC of debt −~300 bp / ~$1.1B annualized savings. Inaugural Eurobond. | Credit is open for contracted GPU/HPC paper. Constructive for other neo-clouds and IG-adjacent AI-infra borrowers that can show take-or-pay contracts. Offset: interest still +140% YoY to $640 million — cheaper spreads, much more principal. Do not infer a Fed path. | | Input inflation | July ~25% list-price increase + component pass-through. “Value of output has outpaced the value of the input increases.” | Producer-price / AI-capex inflation, not CPI. Better for GPU, HBM, SSD, networking, and power-equipment vendors that can hold price. Worse for AI-cloud buyers facing higher $/GPU. Treat 25% list as a spot-price signal, not audited ASP. | | Consumer | Not discussed. Inference framed as enterprise / lab monetization (Grammarly, You.com, coding agents). | No consumer-macro signal. Do not use this call for retail, BNPL, or leisure. | | Industrial / enterprise | Caterpillar will deploy Vera Rubin as its “data factory” for physical AI on autonomous construction equipment. Life sciences (Isomorphic); quants (Flow Traders, IMC); IBM RL/agents; Leidos for defense. | Positive industrial-AI capex read — a Fortune industrial is training physical-AI models at Rubin scale. Cross-reads better for industrial software, robotics, construction-tech, and GPU suppliers tied to physical AI. | | Power / NIMBY | Moratoriums “are not going to impact the demand… they are going to impact where this infrastructure gets built.” Pay for grid upgrades so they do not hit the rate base. | Demand intact, siting harder. Better for utilities and grid OEMs that can charge AI customers for upgrades. Worse for DC development in hostile municipalities. Do not read moratoriums as a demand cliff. | | Supply chain | “Land, power and shell through GPUs and networking through memory, all of which is being challenged.” Solidigm LTA as the illustration. | Confirms the multi-year AI BOM shortage (HBM/SSD/networking/labor, not only GPUs). Better for SK Hynix/Solidigm, Micron, Samsung memory, NVDA, networking. Worse for would-be neo-cloud scale-ups without allocation. |

Highest-signal counterparties

1. NVIDIA (NVDA) — First cloud to bring up Vera Rubin NVL72; A100 (2020 architecture) recontracted into 2029; “a lot of” the 5–10 ppt step-up is the Rubin SKU. Better for NVDA: first-wave Rubin demand and pricing power, Hopper/Ampere residual value (anti-obsolescence), CRWV as a reference-architecture channel. Worse for “GPU glut 2026.” CRWV did not mention AMD.
2. Solidigm (SK Hynix NAND) — Multi-year priority access to enterprise SSD capacity. Better for Solidigm / SK Hynix NAND and constructive for the AI SSD/HBM complex (MU, Samsung) if storage is the new bottleneck after GPUs. Worse for CRWV peers without allocation.
3. Caterpillar (CAT) — Vera Rubin for physical AI on autonomous construction equipment. Better for CAT’s autonomy narrative and for NVDA industrial design-wins. Not a CAT equipment-cycle (housing/non-res) read — this is R&D/infra spend.
4. Jane Street / Flow Traders / IMC — Quant vertical compounding (Q1: ~$10B backlog, Jane Street +$6B plus $1B strategic investment). Confirms HFT/quant AI spend is scaling, not a 2022-style quant winter. Customer so constrained it bought equity.
5. Leidos (LDOS) — CoreWeave Federal into defense / national-security missions. Better for LDOS as integrator; named, not quantified.
6. Hyperscalers / Meta (META) — “The hyperscalers use us, the labs use us.” Customer A mix 71% → 36% is the hard diversification datapoint (letters not mapped this run). Barclays put Meta’s edge-AI news on the call the day before; Intrator declined to treat edge as a TAM destroyer. Q1 $21B Meta agreement was not re-quantified. Competitive: META building more of its own / edge silicon is a long-term substitution risk.
7. Neo-clouds — Intrator: Q2 power add larger than “any single neo-cloud operates in total today.” Nitin: competition up, yet “demand, pricing and margin all expanding.” Worse for subscale neo-clouds. Expanding prices despite more competitors is the industry read — if corroborated on IREN/NBIS prints, the group has pricing power; if not, CRWV is idiosyncratic.
8. Google Cloud (GOOGL) — Interconnect “beginning with Google Cloud” (IR PR). Co-opetition, not a disclosed GPU offtake. Isomorphic Labs (Alphabet affiliate) is a new life-sciences customer.

Ranked external uses of this call

1. NVIDIA residual value + Rubin pricing (A100-to-2029; Rubin carrying the 5–10 ppt)
2. BOM tightness beyond GPUs (Solidigm SSD LTA; memory/labor/storage)
3. HPC credit as a real asset class (full-size deal in a dislocated week; Eurobond; 300 bp cheaper WACC) — does not cancel +140% YoY interest dollars
4. NIMBY ≠ demand destruction; grid-upgrade / utility read is positive if AI customers pay
5. Enterprise/industrial conversion (CAT + IBM + quants + Leidos) that makes Customer A 71%→36% believable
6. Missing classical macro — do not force Fed/CPI/household inferences

CoreWeave printed a capacity-conversion quarter: revenue YoY has stopped decelerating at ~112%, active power re-accelerated to +219%, adj. EBITDA dollars doubled, and sequential adj. OI inflected off the Q1 trough — above a $30–90M guide that had already marked Q1 as the bottom. That is the operating promise from Q1, delivered.

It is not a quality-of-earnings print. Gross margin is still −836 bps YoY. GAAP EPS deteriorated 90% as $640 million of interest (and Q3 guided ~$900M) outran scale. The first FY raise of 2026 is a CapEx raise (+$4.0B mid) with a token revenue raise (+$0.30B). Revenue-beat magnitude vs Street has faded to +0.6%. And the May reason they would not raise FY revenue — “sold out of 2026” — was retired in August without being named as retired.

Q3 is the test: $3.45–3.60 billion is a delivery number, not a demand number. Hit it, and 2026Q2’s +63 bps Accel was the pause before the step-up. Miss it, and the plateau plus the CapEx/interest stack is the story. Data sourced from Daloopa.