## Original Question
Can company-level forecast-vs-actual evidence help us build an early-warning read on AI data-center buildout and downstream semiconductor demand?
## Answer
Yes. CRWV is one of the strongest current additions because it gives a quarterly series for backlog, revenue, active power, contracted power, and capex. It is less granular than APLD on named data-center milestones, but much better on scaled demand conversion.
## Evidence
- Revenue increased from about $982 million in Q1 2025 to about $2.1 billion in Q1 2026.
- Revenue backlog increased from $25.9 billion in Q1 2025 to $99.4 billion in Q1 2026.
- Active power rose from roughly 420 MW in Q1 2025 to more than 1 GW in Q1 2026.
- Contracted power rose from about 1.6 GW in Q1 2025 to about 3.1 GW in Q4 2025.
- Capital expenditures were about $6.8 billion in Q1 2026 after about $8.2 billion in Q4 2025.
## Call
CRWV is a high-quality demand-conversion signal, not a standalone forecast-vs-actual construction ledger. It should sit next to APLD: APLD tests whether named MW get delivered; CRWV tests whether large AI demand converts into revenue, capex, and active power at scale.
## Watch Items
1. Does active power keep rising toward contracted power?
2. Does quarterly revenue continue to convert backlog into recognized revenue?
3. Does capex stay financed without forcing delivery delays?
4. Do customer concentration or contract changes reduce backlog quality?
5. Can we extract site-level delivery dates from filings, transcripts, or presentations?
## Remaining Gaps
The biggest gap is facility-level forecast versus actual delivery. CRWV has strong quarterly financial and power evidence, but this package does not yet map individual data centers, delivery dates, customers, or GPU deployments to each revenue tranche.
## What Would Change The View
- More bullish: active power and revenue keep rising while backlog remains mostly RPO-backed.
- More bearish: backlog grows but active power stalls, capex financing tightens, or revenue conversion slows.
Loop Summary
| iteration | chose_20_words | did_20_words | score_1_10 | score_reason_20_words | next_20_words |
|---|---|---|---|---|---|
| 1 | Create a CRWV source spine from official releases and presentations. | Cataloged eight official CRWV sources spanning Q1 2025 through Q1 2026. | 9 | Official sources cover revenue, backlog, power, capex, financing, and timing. | Extract a normalized quarterly evidence ledger. |
| 2 | Extract revenue, backlog, power, and capex rows with snippets. | Built 29 evidence rows with metric, value, unit, snippet, source, interpretation, and caveat. | 9 | Rows are source-traced and cover the core signal families. | Measure backlog-to-revenue and backlog-to-power conversion burden. |
| 3 | Quantify whether backlog has outrun active power and revenue. | Computed backlog-to-quarterly-revenue and backlog-per-active-MW checkpoints for Q1, Q4, and Q1. | 8 | Useful diagnostic, but site-level delivery rows remain missing. | Score CRWV as a semiconductor-cycle signal. |
| 4 | Score CRWV across demand, revenue, capacity, capex, and chip-cycle usefulness. | Created six signal scores with evidence, reasoning, and next verification requirements. | 9 | Scoring separates strong evidence from remaining verification gaps. | Compare CRWV with APLD instead of averaging them. |
| 5 | Compare CRWV and APLD by disclosure role and research usefulness. | Built a comparison showing APLD as physical execution and CRWV as scaled demand conversion. | 9 | Comparison clarifies how both companies answer different parts of the question. | Identify falsifiers and follow-up tasks. |
| 6 | Define watch items that could change the CRWV read. | Captured priority backlog: site-level rows, Oracle, Equinix, common schema, and NVDA lag test. | 8 | Backlog is actionable, though not yet automated into the loop harness. | Write the one-page research memo. |
| 7 | Write a concise CRWV memo focused on the original research question. | Wrote answer, evidence, call, watch items, and what would change the view. | 9 | Memo gives a clear call while preserving caveats. | Build an auditable dashboard for the public URL. |
| 8 | Render a public dashboard with tables and visual summaries. | Generated dashboard HTML with evidence tables, comparison, scores, and loop summary. | 8 | Readable and auditable; charting remains simple HTML rather than plotted images. | Validate artifacts with machine checks. |
| 9 | Write and run a validator covering evidence, loops, memo, and dashboard. | Added validation for required files, source rows, evidence rows, scores, and word limits. | 9 | Validator catches common fake, tiny, or malformed loop outputs. | Publish the package and summarize quality. |
| 10 | Publish the run and provide a quality read. | Copied dashboard to public index and recorded validation status. | 9 | Package is complete enough for the next company expansion. | Add Oracle and Equinix before running twenty unattended loops. |
Core Evidence Ledger
| company | ticker | period | metric | value | unit | source_id | source_url | source_snippet | interpretation | caveat |
|---|---|---|---|---|---|---|---|---|---|---|
| CoreWeave | CRWV | Q1 2025 | revenue | 982 | USD_millions | crwv_q1_2026_presentation | source | Revenue chart shows Q1'25 $982 and Q1'26 $2,078. | Revenue more than doubled year over year before Q1 2026. | Presentation rounds to millions and is not a capacity metric. |
| CoreWeave | CRWV | Q2 2025 | revenue | 1213 | USD_millions | crwv_q1_2026_presentation | source | Revenue chart lists $1,213 for Q2'25. | Sequential revenue growth continued after Q1. | Does not identify which data centers drove the revenue. |
| CoreWeave | CRWV | Q3 2025 | revenue | 1365 | USD_millions | crwv_q1_2026_presentation | source | Revenue chart lists $1,365 for Q3'25. | The revenue base kept scaling with deployed capacity. | Revenue includes usage and contracts, not solely new builds. |
| CoreWeave | CRWV | Q4 2025 | revenue | 1572 | USD_millions | crwv_q1_2026_presentation | source | Revenue chart lists $1,572 for Q4'25. | Revenue conversion was already large before the Q1 backlog step. | Quarterly result is not a forecast-vs-actual delivery test. |
| CoreWeave | CRWV | Q1 2026 | revenue | 2078 | USD_millions | crwv_q1_2026_presentation | source | Revenue: $2.1 billion, up 112% YoY. | CoreWeave is already a scaled revenue conversion signal. | Company rounds headline revenue to $2.1 billion. |
| CoreWeave | CRWV | Q1 2025 | revenue_backlog | 25.9 | USD_billions | crwv_q1_2026_presentation | source | Revenue Backlog: $25.9 billion in Q1'25. | Backlog base was already substantial one year ago. | Backlog is subject to delivery and service availability. |
| CoreWeave | CRWV | Q2 2025 | revenue_backlog | 30.1 | USD_billions | crwv_q2_2025_press_release | source | Revenue backlog was $30.1 billion as of June 30, 2025. | Q2 backlog increased from Q1. | Backlog includes RPO plus estimated future committed-contract revenue. |
| CoreWeave | CRWV | Q3 2025 | revenue_backlog | 55.6 | USD_billions | crwv_q3_2025_press_release | source | Revenue backlog was $55.6 billion as of September 30, 2025. | Backlog almost doubled in Q3. | Large contract wins can make backlog lumpy. |
| CoreWeave | CRWV | Q4 2025 | revenue_backlog | 66.8 | USD_billions | crwv_q4_2025_press_release | source | Revenue backlog was $66.8 billion as of December 31, 2025. | Q4 backlog remained sharply above Q2. | Backlog remains dependent on capacity delivery. |
| CoreWeave | CRWV | Q1 2026 | revenue_backlog | 99.4 | USD_billions | crwv_q1_2026_presentation | source | Revenue Backlog: $99.4 billion, up 284% YoY. | Backlog accelerated again in Q1 2026. | The backlog figure can fluctuate with contract timing. |
| CoreWeave | CRWV | Q1 2026 | remaining_performance_obligations | 98.8 | USD_billions | crwv_q1_2026_presentation | source | Remaining performance obligations $98.8. | Most Q1 2026 backlog is now formal RPO. | RPO still needs service availability to convert. |
| CoreWeave | CRWV | Q1 2026 | other_committed_contract_revenue | 0.6 | USD_billions | crwv_q1_2026_presentation | source | Other amounts ... committed customer contracts $0.6. | Q1 backlog is less dependent on estimated non-RPO amounts than Q1 2025. | Still subject to delivery and service availability. |
| CoreWeave | CRWV | Q1 2025 | active_power | 420 | MW | crwv_q1_2025_press_release | source | approximately 420 MW of active power and approximately 1.6 GW contracted power. | Baseline active power before the 2025 expansion. | Approximate company disclosure. |
| CoreWeave | CRWV | Q2 2025 | active_power | 470 | MW | crwv_q2_2025_press_release | source | ended the quarter with approximately 470 MW of active power. | Physical active power rose sequentially. | Approximate company disclosure. |
| CoreWeave | CRWV | Q3 2025 | active_power | 590 | MW | crwv_q3_2025_press_release | source | bringing the total to approximately 590 MW. | Q3 added active power while backlog surged. | Approximate company disclosure. |
| CoreWeave | CRWV | Q4 2025 | active_power | 850 | MW | crwv_q4_2025_press_release | source | bringing the total to more than 850 MW. | Q4 added a larger active-power tranche. | Disclosure says more than 850 MW; row uses 850 as a floor. |
| CoreWeave | CRWV | Q1 2026 | active_power | 1000 | MW | crwv_q1_2026_one_pager | source | surpassed 1 GW of active power. | CoreWeave crossed a meaningful physical deployment threshold. | Disclosure says surpassed; row uses 1,000 MW as a floor. |
| CoreWeave | CRWV | Q1 2025 | contracted_power | 1.6 | GW | crwv_q1_2025_press_release | source | approximately 1.6 GW of contracted power at quarter end. | Future-power commitments exceeded active power by almost 4x. | Contracted power is not operating power. |
| CoreWeave | CRWV | Q2 2025 | contracted_power | 2.2 | GW | crwv_q2_2025_press_release | source | increased total contracted power approximately 600 MW to 2.2 GW. | Contracted supply kept expanding ahead of active power. | Contracted power can slip or be reconfigured. |
| CoreWeave | CRWV | Q3 2025 | contracted_power | 2.9 | GW | crwv_q3_2025_press_release | source | Expanded total contracted power to approximately 2.9 GW. | Contracted-power queue rose materially in Q3. | Contracted power is a supply pipeline metric. |
| CoreWeave | CRWV | Q4 2025 | contracted_power | 3.1 | GW | crwv_q4_2025_press_release | source | Expanded total contracted power to approximately 3.1 GW. | Supply commitments remained far above active power. | Not every contracted MW implies near-term GPU installation. |
| CoreWeave | CRWV | 2030 target | active_power_target | 8 | GW | crwv_q1_2026_one_pager | source | well on our way to more than 8 GW by 2030. | Management frames a multi-year power expansion target. | Long-dated target, not delivered capacity. |
| CoreWeave | CRWV | Q1 2025 | capital_expenditures | 1.858 | USD_billions | crwv_q1_2026_presentation | source | Capital expenditures $1,858 for Q1 2025. | Capex base was already high before 2026 growth. | Company-defined capex adjusts for construction in progress changes. |
| CoreWeave | CRWV | Q2 2025 | capital_expenditures | 2.9 | USD_billions | crwv_q1_2026_presentation | source | Capital expenditures chart lists $2.9 for Q2'25. | Infrastructure spend increased with demand. | Rounded presentation figure. |
| CoreWeave | CRWV | Q3 2025 | capital_expenditures | 1.9 | USD_billions | crwv_q1_2026_presentation | source | Capital expenditures chart lists $1.9 for Q3'25. | Capex stayed elevated despite sequential variation. | Rounded presentation figure. |
| CoreWeave | CRWV | Q4 2025 | capital_expenditures | 8.2 | USD_billions | crwv_q1_2026_presentation | source | Capital expenditures chart lists $8.2 for Q4'25. | Q4 capex spiked ahead of 2026 delivery/revenue. | Large quarterly capex can reflect timing of equipment and leases. |
| CoreWeave | CRWV | Q1 2026 | capital_expenditures | 6.786 | USD_billions | crwv_q1_2026_presentation | source | Capital expenditures: $6.8 billion. | Q1 capex remained very high while backlog approached $100 billion. | Capex is spend, not delivered revenue capacity. |
| CoreWeave | CRWV | Q1 2026 | net_cash_used_in_investing | 7.708 | USD_billions | crwv_q1_2026_press_release | source | Net cash used in investing activities (7,708). | Actual cash investment confirms capital intensity. | Cash-flow statement includes investing items beyond capex. |
| CoreWeave | CRWV | Q1 2026 | net_cash_provided_by_financing | 3.914 | USD_billions | crwv_q1_2026_press_release | source | Net cash provided by financing activities $3,914. | Financing is a key part of the growth model. | Financing availability is a major stress variable. |
Backlog Conversion Diagnostics
| period | revenue_usd_m | backlog_usd_b | active_power_mw | backlog_to_quarterly_revenue_x | active_power_backlog_usd_m_per_mw | read |
|---|---|---|---|---|---|---|
| Q1 2025 | 982 | 25.9 | 420 | 26.4 | 61.7 | Demand runway is already long relative to quarterly revenue. |
| Q4 2025 | 1572 | 66.8 | 850 | 42.5 | 78.6 | Backlog scaled faster than active power, raising execution burden. |
| Q1 2026 | 2078 | 99.4 | 1000 | 47.8 | 99.4 | Backlog is huge; the next question is active-power and revenue conversion. |
Signal Scores
| signal_dimension | score_1_10 | evidence | why_score | what_to_verify_next |
|---|---|---|---|---|
| backlog_growth | 9 | Revenue backlog rose from $25.9B in Q1 2025 to $99.4B in Q1 2026. | Strong forward demand signal with formal RPO support. | Track future backlog cancellations, duration, and customer concentration. |
| revenue_conversion | 8 | Revenue rose from $982M in Q1 2025 to $2,078M in Q1 2026. | Large current revenue confirms conversion is happening, not only forecasted. | Tie revenue growth to delivered active power and utilization. |
| physical_capacity | 8 | Active power rose from roughly 420 MW to over 1 GW. | Physical delivery bridge is visible but not site-level granular. | Extract facility-level openings and customer capacity allocations. |
| capex_signal | 8 | Q1 2026 capex was about $6.8B after $8.2B in Q4 2025. | Spend signal is large and directly linked to AI infrastructure scaling. | Separate GPUs, data center infrastructure, leases, and construction timing. |
| semiconductor_cycle_lead | 7 | Backlog, contracted power, active power, and capex are observable quarterly. | Better than single-company MW, but GPU order timing is still indirect. | Compare capex and active-power additions against NVDA data-center revenue. |
| overall_signal_quality | 8 | CRWV combines demand backlog, capex, and physical power series. | It complements APLD by adding scale and quarterly financial conversion. | Add Oracle and Equinix to distinguish company-specific financing from cycle signal. |
APLD vs CRWV Comparison
| company | ticker | best_signal | current_evidence | strength | weakness | role_in_research |
|---|---|---|---|---|---|---|
| Applied Digital | APLD | named-campus forecast versus ready-for-service actuals | PF1 Building 1 forecast Q4 2025 and delivered 100 MW in November 2025. | Physical build execution is visible by campus and MW. | Small revenue base and limited delivered MW so far. | Use APLD to test forecast accuracy and construction execution. |
| CoreWeave | CRWV | backlog, capex, revenue, active power, and contracted power | $99.4B backlog, $2.1B Q1 revenue, over 1 GW active power. | Scaled quarterly demand and financial conversion signal. | Less facility-level detail and financing/deployment lags can distort timing. | Use CRWV to test whether AI demand converts into spend and active capacity. |
| Research system | APLD+CRWV | APLD physical milestones plus CRWV scaled demand conversion | APLD has named MW delivery; CRWV has backlog/revenue/capex/power series. | Different disclosure styles triangulate the AI data-center cycle. | Still missing Oracle and Equinix for hyperscaler and colocation context. | Treat each company as a different lens, not as interchangeable averages. |
Decision Backlog
| priority | next_work | why | acceptance_test |
|---|---|---|---|
| 1 | Extract CRWV site-level disclosures from 10-Q, transcript, and presentation footnotes. | Facility-level timing would turn CRWV from macro signal into forecast-vs-actual ledger. | At least five site or region rows with dated delivery claims and source snippets. |
| 2 | Add Oracle cloud capex and data-center capacity commitments. | Oracle is closer to hyperscaler customer demand and may confirm CRWV backlog quality. | Quarterly ORCL capex, RPO/cloud backlog, and data-center capacity rows. |
| 3 | Add Equinix xScale and leasing indicators. | Colocation disclosures show whether broader data-center supply follows AI demand. | xScale backlog, openings, capex, and leasing rows across eight quarters. |
| 4 | Build a common schema for backlog, capex, active MW, contracted MW, and revenue. | Cross-company scoring needs comparable metric families without averaging unlike bases. | One validated CSV schema populated by APLD, CRWV, ORCL, and EQIX. |
| 5 | Add NVDA data-center revenue lag comparison. | The research goal is an early-warning semiconductor-cycle gauge. | Differenced cross-correlation and honest short-history labels. |