The case for comparing AI with the mortgage boom rests on the financial machinery: financing helps create demand, assets and contracts support more borrowing, and backstops connect participants to the same underlying risk.
In 2007 the asset was a house. In 2026 it’s a GPU. The machine wrapped around it is one we’ve seen before: money that helps fund its own demand, debt stacked on the same cash flows, and promises that pull losses back to whoever made them.
In June I published The Great Compression. The argument was simple: the risks in this economy don’t look dangerous one at a time. They look dangerous as a circle.
Here’s why I keep writing about it. I hate this pattern. Every cycle, the biggest companies, banks and venture funds find the next real thing, build complicated financial machinery around it and make a fortune on the way up. And too often, when it breaks, they get rescued because they’ve become too big to fail. I think AI is heading for the same seat. These companies are now so central to the economy, the stock market and national competitiveness that if something breaks, letting them fail won’t feel like an option.
This is the follow-up, written for founders and for anyone with savings, a job or a business that touches this economy, which is pretty much everyone. I wanted to stop describing the circle and actually draw it. Who pays whom. Who lends to whom. Who has promised to cover whose losses if the assets disappoint.
Here’s what came back. The money in AI is wired a lot like mortgage money was wired before 2008. Investors are also suppliers. Suppliers are also lenders and backstops. Customers can end up as shareholders. When the same names sit on both sides of the trade, a crack in one place rarely stays in one place.
That doesn’t mean the ending has to be the same, and the filings show real cushions too. I kept every one of them in. Every number below links to a filing or an announcement. Where the evidence says maybe, I say maybe. Where it says connected, I’ll show you exactly how.
Everyone is trying to put these pieces together right now. I wanted them in one place: simple enough to get in five minutes, detailed enough to check every number, and clear enough to help you think about how to position yourself before the cycle turns.
— Patrick
The matching diagrams group participants by their financial roles: customers, investors, suppliers, asset owners, lenders and backstops. Follow the arrows to see how financing supports purchases, purchases create revenue, and contracts or guarantees support further investment. Companies can occupy several roles.
2003-2007 / before the collapse
Different securities. Shared mortgage risk.
2025–2026 / companies grouped by financial role
Investor capital becomes demand for products the investors also sell.
Swipe across the pair →
Hover or tap any circle. The same seat lights up in the other machine.
of Baa-rated CDO tranches bought by other CDOs in 2007.
Merrill CDOs sold a tranche into another Merrill CDO, 2003–2007.
89% is a tranche-purchase share. 134 / 142 is a count of Merrill deals, not a dollar-volume share. The AI panel uses different units and contract periods.
Historical sources: FCIC pp. 202–203; Federal Reserve testimony.
The AI parallel is financial interdependence. Crisis severity also depends on leverage, refinancing, contract strength and outside customer cash flows.
Microsoft → OpenAI funding. Cumulative through June 30, 2026. [1]
Microsoft’s OpenAI-related commercial revenue recognized in FY2026. [1]
Revenue includes revenue share and is not cash collected. Funding is cumulative; revenue covers one fiscal year.
Backstops have different triggers. Broadcom’s cited customer is unnamed. [5]
Sources: [1] Microsoft FY26 SEC. [2] Company compute / investment announcements. [3] CoreWeave / NVIDIA SEC. [4] CoreWeave, Meta SEC; Apollo. [5] NVIDIA, Meta, Broadcom SEC. Full links and terms in the accompanying briefing.
This is a shared mechanism, not a forecast of a 2008-scale crisis. Cash-rich sponsors, long-term contracts and amortizing loans can absorb shocks. Evidence through October 5, 2026.
The earlier company comparison and eight detailed loops also appear below the connected network, in one graphic you can switch between. On smaller screens, scroll across the pair. Groups contain selected examples; each company can fill several roles, and every member does not participate in every arrow.
Merrill CDOs sold at least one tranche into another Merrill CDO during 2003–2007. The FCIC, p. 203 documents a concrete system of internal buyers. This is a count of deals, not a share of their dollar volume.
Amazon announced a $50B investment in OpenAI while OpenAI expanded its AWS agreement by $100B over eight years. The February 27 terms begin with $15B of investment and a conditional $35B. The $100B expands an existing $38B AWS agreement. These are distinct obligations, not completed offsetting cash transfers. This is one specific investor–supplier relationship behind the grouped diagram; the company view of the detailed graphics below displays it separately.
The financing network also includes the credit layer: equipment lease debt at xAI, borrowing at CoreWeave and contingent support from suppliers and tenants. BIS documents investment–commercial overlap, while NVIDIA’s residual-value framework adds another support mechanism. The similarity is the financial feedback mechanism. The ability to absorb losses determines how damaging a reversal becomes.
The connected network gives each company one shared position, so capital, compute contracts, asset financing and contingent support meet in the same image. Orange shows equity, blue shows commercial contracts, purple shows credit, and dashed paths distinguish warrants, guarantees and conceptual links. Follow the return paths to see how investors can also become suppliers, customers or backstops.
One connected ecosystem. Capital, purchases, credit and guarantees overlap.
Arrows show the direction of financing, purchases or support. Grey supply links are economic relationships. Node sizes and line widths do not measure exposure. One step means a disclosed direct relationship on this map. It shows connection, not a forecast of losses.
Microsoft recognized FY2026 commercial revenue from OpenAI, including revenue share. [1]
A $6B OpenAI receivable was outstanding at June 30.
CoreWeave: $35.6B total debt at June 30. Its $8.5B facility is included in that financing picture, rather than added to it. [18, 21]
The facility relies on a specific Meta take-or-pay agreement. The separate $21B Meta expansion is not established as its exact collateral contract.
Hyperion: ~$27B development. El Paso: ~$14B development, with $12.5B debt financing announced. [19]
The ~$28B and ~$13B value thresholds decline over time and cover conditional shortfalls. They are not cash already paid or extra project funding.
NVIDIA describes up to 25% residual-value support for selected financing opportunities. The aggregate exposure is undisclosed. [22]
Can cash generated outside the financing circle carry the commitments made inside it?
Outside demand exists. Its share of this network’s funding is not measured here.
Capital → Purchases → Revenue & contracts → Asset-backed credit → More capacity
The 2008 similarity is the reinforcing financial mechanism and shared underlying exposure.
xAI’s $5.4B transaction includes Apollo’s $3.5B; June lease debt was $13.33B across three arrangements. [16]
Download this infographic PNG Full resolution preserves every label. Scroll horizontally to inspect the network.
Outside-customer cash is not quantified in this map; gray demand links are conceptual.
These are the earlier detailed graphics, combined into one. Switch between the company view and the deal view below. Both keep the individual investments, commercial commitments, credit arrangements and conditional support. The opening role-based comparison and connected network provide additional ways to read the same financing relationships.
OpenAI + Claude + Grok + Meta / selected 2025–2026 connections
Orange: capital Blue: commercial payments / compute
Broadcom’s cited passage leaves the customer unnamed.
The full company comparison, including its disclosed figures and financing qualifications.
Download company comparison PNG
The same firms supply capital, sell compute, finance assets and support future values. Follow the economic return paths.
Orange: capital / credit Blue: commercial flows Red dashed: contingent support Dashed orange: equity rights
GPU utilization, customer payments, campus rents and resale values matter across equity, contracts, loans and guarantees. A common demand or asset-value shock could test several participants at once.
AIP partners include BlackRock/GIP, Microsoft, MGX, NVIDIA and xAI. Targets are not deployed funding.
These are selected economic connections, not audited dollar tracing or a prediction of an identical crisis. Strength of outside demand, cash collection, leverage and contract terms determine how a reversal spreads.
Sources [1–8], exact terms and additional backstop sources: accompanying Patrick Frank research briefing.
The original eight-loop infographic.
Download eight-loop infographic PNG
Microsoft adds direct evidence of an investor earning commercial revenue from its investee. Meta adds contracts and contingent promises that support financing for the assets it will use. These connections extend the comparison beyond one model developer.
The FY2026 SEC note also records a $6B OpenAI receivable. Revenue recognized and cash collected differ. These disclosures establish an investor–customer relationship; they do not trace the original investment dollars back to Microsoft. The October Azure agreement announced $250B of incremental services, a separate multiyear contract rather than annual revenue.
Microsoft is also an investor and supplier to Claude’s developer: its partnership with Anthropic and NVIDIA pairs up to $5B of Microsoft investment with $30B of Azure compute commitments. The April OpenAI amendment keeps Microsoft as a primary cloud partner and revenue-share recipient through 2030 with a cap, while permitting OpenAI products on other clouds.
estimated development cost; funds 80% / Meta 20%
~$28B declining RVG threshold
$12.31B initial lease commitment; leases begin 2029
announced debt financing within a ~$14B development
~$13B declining RVG threshold
Announced July 2026; funds 80% / Meta 20%
Meta’s June filing discloses its nonconsolidated Hyperion venture, leases and residual-value guarantee. El Paso’s announcement repeats the lease-and-guarantee structure. The guarantees cover conditional shortfalls in property value after nonrenewal or termination and other conditions. Development costs, debt, rent and guarantee thresholds overlap; they are not a funding total. El Paso is labelled as announced because the latest quarterly disclosure describes closing conditions.
Meta separately announced $21B of CoreWeave capacity through 2032. DBRS’s rating explanation says an $8.5B CoreWeave GPU facility is underwritten by a specific Meta take-or-pay agreement. A contractual payment stream supports credit that buys the equipment used to provide service. The separate $21B expansion is not established as that facility’s exact collateral contract. The facility fully amortizes and has protections for delivery and availability; this gives the comparison concrete financing evidence and concrete differences from runnable mortgage-era funding.
AMD’s Meta agreement adds a different incentive: up to 160M warrant shares, contingent on GPU purchases, stock-price and other conditions. Meta’s initial 1GW purchase commitment is binding; full vesting requires 6GW. This is separate from OpenAI’s AMD warrant, and represents potential supplier equity rather than cash funding.
Microsoft also participates in AIP alongside BlackRock/GIP, MGX, NVIDIA and xAI. AIP’s current stated aim is $30B of equity capital, with potential total investment of $100B including debt. These are fund-wide targets, not Microsoft’s contribution or capital already deployed.
Why the financing pattern does not settle the crisis outcome: Microsoft generated $182.9B of corporate operating cash flow in FY2026 against $115.9B of cash property-and-equipment spending, according to its results. That is a material loss-absorbing resource. The argument is strongest when it identifies shared funding mechanisms and correlated exposures while examining who can keep paying under stress.
Claude is Anthropic’s product; Grok is xAI’s; Gemini is Google’s. The financing comparison belongs at the company and contract level. Adding them shows a repeated structure across competing labs: suppliers provide capital, developers commit to compute, and outside financiers help carry the assets.
Reuters reports, from Anthropic’s confidential IPO prospectus, that it plans to spend at least $111.1B with Google, $110B with Amazon and $31.4B with Microsoft under long-term infrastructure obligations over the next seven to ten years, regardless of usage. All three are also Anthropic investors. The money they put in and the money they are owed sit on opposite sides of the same relationship.
Reported from a prospectus that is not yet public. These commitments overlap the AWS and Azure figures above and are not added to them. Reuters reporting
Google’s own AI spending adds the scale question. Alphabet’s Q2 results show $44.92B of capex against $39.07B of operating cash flow: quarterly free cash flow was −$5.86B, while trailing-year free cash flow remained +$53.27B. These are corporate figures, not Gemini-only costs. The investor–supplier loop is Google’s Anthropic relationship; its internal Gemini development is a different funding channel. The same release shows Alphabet raised $30.5B from new common stock, $19.1B from mandatory convertible preferred stock and $20.3B from senior notes in the quarter. These are company-wide financing figures, not Gemini-only funding. Even Alphabet went to outside capital in the quarter its spending ran ahead of operating cash.
Why this strengthens the 2008 comparison: competing developers can depend on the same chip suppliers, cloud providers and financing pools. Credit and guarantees can turn equipment demand into claims held by investors. A common demand or collateral shock could travel through several balance sheets. This supports a common-risk cycle hypothesis; the outcome depends on external demand and the capacity to absorb losses.
The most useful public test is whether operating cash generation covers current capital spending. Some builders spend far more than their operations generate, while stronger partners can finance expansion from established businesses. The figures below show that difference; they do not isolate cash from independent AI customers.
| Company / period / spending definition | Operating cash | Capital spending | Difference | Coverage |
|---|---|---|---|---|
| CoreWeaveH1 2026Cash PPE and capitalized internal-use software | $3.663B | $14.117B | −$10.454B | 25.9% |
| MicrosoftFY 2026Cash PPE; finance-lease principal excluded | $182.935B | $115.948B | +$66.987B | 157.8% |
| MetaH1 2026Cash PPE 49.113 plus finance-lease principal 1.805 | $64.088B | $50.918B | +$13.170B | 125.9% |
| AmazonTTM Jun 2026PPE net of sales and incentives; finance-lease principal excluded | $161.403B | $169.007B | −$7.604B | 95.5% |
| SpaceX consolidatedH1 2026Gross cash PPE; rebates and finance-lease principal excluded | $3.466B | $28.476B | −$25.010B | 12.2% |
CoreWeave is the clearest financing-dependence example here. Its H1 cash capital spending exceeded operating cash by $10.454B. Its filing reports $13.985B of net financing inflow; operating cash also included a $1.365B increase in deferred revenue, incorporating customer advances for future service.
Microsoft and Meta generated more operating cash than the stated spending measures. Amazon’s negative $7.604B difference is its company-defined trailing-year free cash flow. Microsoft’s difference precedes finance-lease adjustment and other uses. These cash engines make the network more resilient, even as the capital commitments grow.
SpaceX’s filing makes Grok’s infrastructure spending visible, but does not provide AI-only operating cash flow. Its AI segment also includes X advertising, subscriptions and cloud infrastructure. Revenue and operating loss are not cash flow. Separately, company-wide net financing inflow was $100.291B, including $85.675B of net IPO proceeds. That funding is a material cushion; the spending gap alone does not imply imminent distress.
What remains unmeasured: public primary cash-flow statements were not located for OpenAI or Anthropic. Funding rounds, revenue run rates and signed compute contracts cannot establish how much outside customer cash covers their bills. That missing denominator is central to testing the circular-financing thesis.
Construction and equipment need funding before years of future compute payments arrive. Disclosed debt schedules make that pressure measurable. Contract values establish demand commitments; payment dates determine how well the financing bridge works.
| Relationship / amount type | Timing disclosed | Payment detail / limits |
|---|---|---|
| OpenAI → AWS+$100B | 8-year commitment announced Feb 2026 | Annual payments undisclosed |
| Anthropic → AWS>$100B | 10-year commitment announced Apr 2026 | Annual payments undisclosed |
| OpenAI → Azure$250B incremental | Services commitment announced Oct 2025 | Services maturity and annual schedule undisclosed |
| Meta → Hyperion venture$12.31B initial leases | Starts 2029; 4-year initial term per property | 16-year contingent support; no complete RVG schedule |
| Meta / El Paso venture$12.5B announced financing | Capacity expected online 2028 | Exact lease start and annual payments undisclosed |
| CoreWeave DDTL 4.0$8.5B facility capacity | Draws through Jun 2027; matures Mar 2032 | Monthly amortization; not an $8.5B balloon |
| NVIDIA → CoreWeave$6.3B initial order | Residual capacity purchase obligation to Apr 13, 2032 | Subject to delivery, availability and termination terms |
The Azure revenue-share endpoint in 2030 is not an established maturity for the separate $250B services commitment. Eight- and ten-year AWS terms do not reveal annual installments. El Paso’s expected opening does not establish its exact lease commencement. Dividing headline totals evenly across years would invent a payment schedule.
The 2008 similarity is dependence on future cash flows, asset values and continuing finance. The financing tenor still matters: CoreWeave’s specific DDTL 4.0 facility amortizes against a Meta contract, rather than requiring an $8.5B refinancing at maturity. A multiyear project loan behaves differently from the short-term funding runs that amplified mortgage losses.
Hyperion is owned 80% by Blue Owl-managed funds and 20% by Meta. Outside bond financing supports the venture, while Meta supplies leases and a conditional property-value guarantee. The documented guarantee creates a route by which an asset-value shortfall can become cash owed by Meta to the venture.
Assume less capacity is needed and property value falls.
Termination or nonrenewal occurs; all other conditions are assumed met.
Property fair value is below the applicable, declining guarantee threshold.
A capped payment can bring the exposure back to the sponsor.
The values below are invented assumptions, not actual property thresholds, appraisals or forecasts. Other contractual conditions are assumed satisfied.
Qualifying exit + all other conditions met: max(0, $20B − $16B) = $4B.
| Assumed property threshold | Assumed property fair value | Qualifying lease exit | Hypothetical maximum payment |
|---|---|---|---|
| $20B | $22B | Yes | $0B |
| $20B | $22B | No | $0B |
| $20B | $16B | Yes | $4B |
| $20B | $16B | No | $0B |
| $20B | $10B | Yes | $10B |
| $20B | $10B | No | $0B |
With a qualifying exit and all other conditions satisfied: payment = max(0, assumed property threshold − assumed property fair value). A $20B assumed threshold and $16B assumed value yield a $4B illustrative payment. A value at or above the threshold produces zero.
Without a qualifying exit: no payment under this modeled exit trigger, even if value is below the assumed threshold.
The actual disclosed aggregate threshold is approximately $28B and declines over time. Meta’s filing does not publish a complete property-by-property schedule or all eligibility conditions, and says payments were not considered probable at June 30. The guarantee covers long-lived campus property and infrastructure, rather than serving as a GPU-value guarantee.
Weak usage by itself does not trigger a payment. Replacement tenants, property sales, declining thresholds and continued lease payments can reduce the risk. The public disclosures do not establish how payment or remaining losses would be distributed among lenders and equity investors.
The mortgage-era parallel is risk that returns through a promise. Outside investors help finance an asset, but a sponsor’s guarantee can connect that asset’s downside back to its balance sheet. This is a specific transmission mechanism, not evidence that a lender has already lost money or that an AI crisis is inevitable.
of Baa-rated CDO tranches were bought by other CDOs in 2007.
of AI-to-AI disclosed deal value involved an investor with a commercial relationship.
NVIDIA residual-value support for selected financing opportunities.
These measure different things; they are evidence of related mechanisms, not comparable rates. FCIC documents securities buying one another. BIS documents overlap between investment and commercial relationships. NVIDIA describes conditional loss support.
In each cycle, capital buys the asset, the activity validates the growth narrative, and financial structures make the next round easier. The danger is a feedback loop that strengthens in both directions.
Common mechanism: stronger financing can create the activity used to justify stronger financing. The historical loop draws on the FCIC and Federal Reserve; the AI loop combines BIS findings with disclosed financing structures. Real outside customers also contribute demand.
Left: what happened in the mortgage crisis. Right: a possible AI transmission path, not a forecast. The common issue is correlated cash flow and collateral. AI funding terms differ from the short-term funding runs described by the Federal Reserve.
The strongest comparison is that participants can help finance one another’s purchases. That can reinforce growth before independent cash generation catches up.
CoreWeave combines those relationships with borrowing supported by infrastructure and customer contracts. Each arrangement can be reasonable individually; together they connect demand, collateral and loss absorption.
SEC: completed $2B equity investment · SEC: $6.3B initial capacity order · CoreWeave: $8.5B facility, Exhibit 99.1. NVIDIA’s residual-capacity purchase obligation is subject to delivery, availability and termination terms. The order is not $6.3B of cash already paid. The facility is a borrowing limit, not a separate amount to add to reported debt.
CoreWeave’s June 30 filing reports these obligations separately. The facility’s investment-grade ratings apply to that financing, not the entire company. Ratings are a useful underwriting signal; they do not eliminate dependence on customer cash flow and asset values.
Institutional financing broadens the pool of capital. Supplier support for future asset values can make that capital more willing to participate, while leaving the parties exposed to a common compute-market downturn.
Third-party capital mobilization aim, over time
Memoranda of understanding; subject to final agreements
NVIDIA’s platform announcement is an ambition for third-party financing, not $500B already raised or invested by NVIDIA. Its residual-value framework provides support in some cases, up to 25% of an opportunity. The total guarantee exposure is undisclosed; multiplying $500B by 25% would be misleading.
A concrete example of institutional GPU credit: Blue Owl announced $2.4B for IREN on August 28, split between a senior secured loan and notes, drawn alongside hardware delivery. This demonstrates asset-backed financing; it does not by itself establish a circular equity arrangement.
A lender, borrower and supplier may still rely on the same utilization, customer payments and equipment values. Residual-value support is a contingent loss-sharing mechanism; it is legally different from the CDS protection used before 2008.
NVIDIA’s six-partner platform initiative aims to mobilize over $500B. It expands the potential funding base; the amount is a plan, not capital already raised.
NVIDIA announcementNVIDIA describes residual-value support for selected opportunities. This is relevant to who absorbs losses if compute assets disappoint.
NVIDIA frameworkBlue Owl’s IREN package combines senior secured lending and notes. Draws follow hardware delivery, illustrating the connection between equipment purchases and credit.
Blue Owl announcementReuters reports a one-year delay at Oracle / Blue Owl’s Project Jupiter. Blue Owl says financial commitments remain. It shows timing risk, not a confirmed default.
Reuters reportThe added authorization allocates potential cash to shareholders. Its role differs from investment in a customer or support for a credit transaction.
NVIDIA announcementReuters reports about 80% of Anthropic’s decade-long infrastructure plan is non-cancelable or payable regardless of usage. Basis: a confidential prospectus seen by Reuters; earlier contracts overlap. Reuters puts the plan at at least $518B over a decade, with six partners.
Reuters reportingThe new study gives the comparison an empirical foundation: investors and commercial partners often overlap. Its observations cover 2021–2025, not current cash tracing.
BIS BulletinReuters reports a financing capacity of up to $42B for Anthropic infrastructure. Capacity is not cash drawn; this amount is distinct from the earlier lease backstop.
Reuters reportingReuters, citing FT, reports discussions to place roughly $8B of NVIDIA chips in a debt-funded vehicle and lease them back. It is a proposal, not a closed transaction.
Reuters / FT reportingThe September 28 announcement is capital allocated to shareholders, with execution expected through fiscal 2028. Authorization is not execution. It belongs as a cash-out branch, rather than an arrow funding customer purchases. If executed, repurchases use cash that could otherwise absorb losses, but this headline does not prove circular financing or an impending liquidity shortfall.
| Shared mechanism | Mortgage boom | AI buildout |
|---|---|---|
| Demand from inside the ecosystem | Other CDOs bought 89% of Baa-rated CDO tranches in 2007. | AI investors can also be commercial partners; suppliers help fund customers. |
| Assets support more borrowing | Mortgages and securities supported leveraged financing; falling values tightened credit. | GPU infrastructure and customer contracts can support secured loans. |
| Ratings help attract capital | Structured-credit ratings encouraged investor confidence in mortgage exposure. | CoreWeave’s specific asset-and-contract-backed facility has A3 / A (low) ratings. |
| Backstops connect balance sheets | CDS and sponsor liquidity promises moved risk, sometimes returning it to concentrated providers. | Capacity commitments, supplier backstops and Meta’s conditional project-value guarantees connect participants to infrastructure performance. |
| A common shock travels through the network | Mortgage losses, collateral haircuts and funding runs reinforced one another. | Weaker utilization or asset values could affect borrowers, lenders and suppliers together. This is a risk scenario. |
A new asset class is acquiring familiar financial machinery: demand financed from within its ecosystem, borrowing against assets and expected cash flows, and promises that link losses across participants. That supports the argument for a familiar financing cycle around a different asset.
The mortgage crisis was amplified by runnable short-term funding, forced sales and thin loss absorption. The Federal Reserve reports broker-dealer repo liabilities grew 2.5 times in four years; asset-backed commercial paper fell about $200B in August 2007. A long-dated AI project loan is not an overnight repo position.
AI also has real external customers and cash-generating suppliers. OpenAI’s March round closed with $122B of committed capital, but committed financing is different from recurring operating cash flow. Whether the cycle ends in manageable overinvestment or a wider crisis depends on utilization, refinancing terms, collateral values and who can absorb losses.
The BIS also points to late-1990s telecom vendor financing as a close commercial precedent. The 2008 comparison is most persuasive for leverage and risk transmission, rather than a claim that every contract or funding structure is identical.
I wrote this so founders and everyday investors can see the machine, not to scare anyone. Nobody knows when or if it cracks, and anyone selling you a date is guessing. Here’s how I use this map.
On timing: Last cycle, New Century, a leading subprime lender, filed for bankruptcy in April 2007. Lehman Brothers didn’t fail until September 2008, 17 months later. Cycles like this run longer than skeptics expect and end faster than believers expect. That’s why I watch the signals, not the calendar.
Research and commentary, not investment advice. Talk to a licensed advisor about your own situation.
Can cash generated outside the financing circle carry the commitments made inside it?
That’s the whole argument in one sentence. If the answer is yes, the circle is a flywheel and this boom pays for itself. If the answer is no, the circle is the problem, and every promise inside it gets tested at the same time.
Same machine. New asset. Watch the cash.
Source-backed snapshot through October 5, 2026. Source actions beside each section retain the reviewed evidence and qualifications.