Capital flow drill-down
Money in, revenue out: the full breakdown.
Issue 16 · Week 32 of 2026.
The four-category capital scorecard expanded with named transactions, burn-to-revenue context, and per-category trends across recent issues. Each row corresponds to a row in the summary table on the issue page.
Capital flow scorecard
Capital in, revenue out, burn ratio, and movement.
| Category | Capital in | Revenue out | Burn to revenue | Movement |
|---|---|---|---|---|
| Frontier LabsOpenAI, Anthropic, Google DeepMind, xAI | ~$95BPrior ~$95B · Flat | ~$21BPrior ~$21B · Flat | ~4.5x | No frontier lab closed primary financing in-window. The week's only new frontier-adjacent capital moved outward: Alphabet became a founding investor and cloud partner in Discovery Loop, the public benefit corporation founded by Jeff Dean, Sanjay Ghemawat, Oriol Vinyals and Quoc Le, at an undisclosed amount. |
| Hyperscaler-HostedAzure-OpenAI, AWS-Anthropic, Google Cloud-Gemini, Oracle-OCI | ~$250BPrior ~$250B · Flat | ~$70BPrior ~$70B · Flat | ~3.6x | No hyperscaler reported in-window and no capex guidance changed, so the aggregate holds at last week's revised figures. The only market-moving disclosure was governance: Alphabet fell 4% on the DeepMind reorganization with its flagship Gemini model still unreleased against a planned June launch. |
| NeocloudsCoreWeave, Nscale, Crusoe, Lambda, Fluidstack, IREN | ~$17BPrior ~$17B · Flat | ~$5BPrior ~$5B · Flat | ~3.4x | No neocloud financing closed in-window and CoreWeave's $99.4B March 31 backlog remains the last official print, with Q2 results due August 11 — three days after this issue. |
| On-Prem / HybridEnterprise GPU clusters, sovereign and national programs, Cisco / Dell / HPE | ~$101BPrior ~$101B · Flat | ~$36BPrior ~$36B · Flat | ~2.8x | No change to the aggregate, but the argument for self-hosting acquired a measurement it did not have before: Artificial Analysis benchmarked serverless endpoints against a self-hosted reference deployment and found some endpoints preserving well under two-thirds of reference accuracy. |
Category
Frontier Labs.
OpenAI, Anthropic, Google DeepMind, xAI
- Capital in
- ~$95Bvs ~$95B · Flat
- Revenue out
- ~$21Bvs ~$21B · Flat
- Burn / rev
- ~4.5xLower means more capital out than in.
The read
The interesting capital event is an org chart rather than a round. Alphabet simultaneously kept Gemini in-house under an operator, moved AGI strategy to a chair role with no operating load, and funded open-ended automated discovery outside the company as a separate legal entity it does not manage. That is a deliberate decision about which parts of a research agenda belong on the balance sheet, and it is the first time an incumbent has externalized a frontier research thesis while retaining the cloud relationship. The amount is undisclosed, so nothing is booked to capital in.
This week's transactions
| Date | Transaction | Amount | Source |
|---|---|---|---|
| 2026-08-05 | Alphabet named as founding investor and cloud partner in Discovery Loop, a public benefit corporation founded by Jeff Dean, Sanjay Ghemawat, Oriol Vinyals and Quoc Le; investment amount undisclosed | Undisclosed | Sundar Pichai memo; 9to5Google |
| 2026-08-07 | OpenAI paused internal Astra activities not meeting stricter security requirements and delayed the launch after evaluations could not rule out a Critical cyber capability level — a schedule cost rather than a capital event | Not disclosed | OpenAI; The Decoder |
Trend across recent issues
- Capital in
- $95B, $95B, $95B, $95B, $95B, $95B, $95B, $95B
- Revenue out
- $21B, $21B, $21B, $21B, $21B, $21B, $21B, $21B
Category
Hyperscaler-Hosted.
Azure-OpenAI, AWS-Anthropic, Google Cloud-Gemini, Oracle-OCI
- Capital in
- ~$250Bvs ~$250B · Flat
- Revenue out
- ~$70Bvs ~$70B · Flat
- Burn / rev
- ~3.6xLower means more capital out than in.
The read
A 4% move on an org chart is worth more attention than its size suggests, because it prices execution risk rather than spend. The market spent the last month rewarding hyperscalers whose AI capex came with a legible monetization path. Alphabet's reorganization is the first event this cycle where investors marked down an AI position on leadership and shipping cadence rather than on capital intensity, which is a different and more durable kind of scrutiny than the capex debate.
This week's transactions
| Date | Transaction | Amount | Source |
|---|---|---|---|
| 2026-08-05 | Alphabet shares fell 4% after Pichai moved Hassabis to Chair and Alphabet Chief Scientist, appointed Kavukcuoglu SVP over Gemini development and frontier research, and confirmed the departure of Jeff Dean and three senior Gemini leaders | -4% share reaction | The Indian Express, citing Reuters |
Trend across recent issues
- Capital in
- $187B, $187B, $187B, $187B, $210B, $230B, $250B, $250B
- Revenue out
- $62B, $62B, $62B, $62B, $62B, $62B, $70B, $70B
Category
Neoclouds.
CoreWeave, Nscale, Crusoe, Lambda, Fluidstack, IREN
- Capital in
- ~$17Bvs ~$17B · Flat
- Revenue out
- ~$5Bvs ~$5B · Flat
- Burn / rev
- ~3.4xLower means more capital out than in.
The read
The category has now gone two full issues priced on an estimate rather than a print, which is the longest stretch this publication has carried it. August 11 resolves the single number that matters: whether contracted backlog converts to revenue at the pace the category's financing assumes. Nothing this week changes the question, and the specialized-inference silicon story adds a new one for later — decode workloads migrating to model-specific chips would erode the GPU-hour demand that backlog implicitly assumes.
This week's transactions
| Date | Transaction | Amount | Source |
|---|---|---|---|
| 2026-08-08 | CoreWeave Q2 2026 results remain scheduled for August 11 with no in-window financing or backlog disclosure; the $99.4B March 31 figure is still the last official print | Not disclosed | CoreWeave investor relations |
Trend across recent issues
- Capital in
- $12B, $12.7B, $13.5B, $13.5B, $17B, $17B, $17B, $17B
- Revenue out
- $5B, $5B, $5B, $5B, $5B, $5B, $5B, $5B
Category
On-Prem / Hybrid.
Enterprise GPU clusters, sovereign and national programs, Cisco / Dell / HPE
- Capital in
- ~$101Bvs ~$101B · Flat
- Revenue out
- ~$36Bvs ~$36B · Flat
- Burn / rev
- ~2.8xLower means more capital out than in.
The read
Self-hosting has historically been justified on control, data residency and cost, all of which are arguments a CFO can push back on. This week it gained a quality argument with a number attached. If the reference deployment is the accuracy ceiling and some endpoints deliver 22% where the reference delivers 37%, then running your own weights is no longer only a governance preference — it is a measurable capability decision. The counterweight is that most enterprises will not match a lab's recommended precision or serving configuration either, so self-hosting buys the option to reach the reference, not the reference itself.
This week's transactions
| Date | Transaction | Amount | Source |
|---|---|---|---|
| 2026-08-04 | Artificial Analysis published the Endpoint Accuracy Index benchmarking serverless endpoints against self-hosted reference deployments of official weights at the lab's recommended precision, initially covering GLM-5.2, gpt-oss-120b and DeepSeek V4 Pro | Not disclosed | Artificial Analysis |
| 2026-08-04 | Liquid AI released LFM2.5-2.6B under the LFM Open License v1.0, whose Section 5(b) states commercial use above a $10M revenue threshold is not licensed under the agreement, while the release page described deployment 'without restrictions' | Not disclosed | Liquid AI; LFM Open License v1.0 |
Trend across recent issues
- Capital in
- $94B, $94B, $94B, $94.5B, $101B, $101B, $101B, $101B
- Revenue out
- $36B, $36B, $36B, $36B, $36B, $36B, $36B, $36B