June 24, 2026 ·19 min
On Funding AI
~$400B raised in 2026 against a capex bill nearing $750B. Where the cash comes from.
Contents 11 sections
AI: Draining The Cash Away
In 2026, the 5 largest AI spenders are set to raise ~$400B of debt and equity to fund a combined capex bill approaching $750B. This will be the first year that the group’s capex bill outstrips the cash their own operations generate. Even the finest cash return machines ever assembled can’t self-fund the current buildout.
| Company | Fiscal Period 2025 | Operating Cash Flow | 2025 Capex | 2026 Capex Guidance |
|---|---|---|---|---|
| Amazon | Calendar | $139.5B | $131.8B gross (~$128.3B net) | $200B |
| Microsoft | Fiscal (Jun) | $136.2B | $64.6B | $190B |
| Alphabet | Calendar | ~$165B | $91.4B | $180B-$190B |
| Meta | Calendar | $115.8B | $72.2B | $125B-$145B |
| Oracle | Fiscal (May) | $20.8B | $21.2B | $56B |
Funneling operating cash flow + balance sheet health towards the AI buildout will either be value-accretive reinvestment, or empire-building into an arms race with an unknowable payback period.
Google sold equity for the first time in twenty years. Amazon printed the largest corporate bond in history. Meta, which had never issued a bond until 2022, now carries $84B of them on its books. Oracle is spending more than it earns.
I will map out the funding dynamics affecting these players.
Google’s Fundraising Bonanza
For the first time in ~20 years, Google has raised money through selling equity. It’s paired that with the largest debt campaign in its history. Combined, that’s $84.75B of equity issued on top of >$71B of new debt issued over the trailing ~12 months across six currencies. The AI buildout has outgrown even the most cash-generative business in the world.
Google’s Balance Sheet Valuation
I’ll value the tangible assets on the books as of Q1-26 end. Cash stands at $38B. M2M securities are held at $88.8B. Receivables are $63B with another $23.9B in other current assets (includes $6.8B of GFiber held-for-sale).
Non-marketable securities are reported as $107B. Despite the accounting treatment, I consider the Anthropic and SpaceX stakes quite easily marketable (especially post-SPCX IPO). The largest disclosed components include:
- 14% stake in Anthropic (as of Mar-2025)
- 6.11% stake in SpaceX (as of 2025 end, confirmed in Apr-2026 SpaceX filing)
Alphabet owns ~75-80% stake in Waymo and consolidates it as a subsidiary. Waymo’s assets sit with the property lines.
Property and equipment is held at $281B net, dominated by data-center assets. Includes servers and network equipment, land and buildings, and a large construction-in-progress balance reflecting the AI buildout. This breaks down as: technical infrastructure $217.9B, office space $48.8B, corporate/other assets $5.9B, giving property/equipment in service $272.6B, less accumulated depreciation $100.2B, plus assets not yet in service $108.6B.
Netting off against this all are:
- $111B in current liabilities (57% of which are accrued expenses)
- ~$94B in long term debt. ($77.5B at Q1-26 end + $16.5B raised in May-26 EUR and CAD deals)
- $13B in LT income tax payable.
Google’s Equity Raise
The 84.75B equity program is split as:
- ~$18B underwritten common (Class A at $355.2, Class C at $351.8). Closed Jun 4th.
- ~$16.75B 6.25% mandatory convertible preferred (Series A → Class A, Series B → Class C). It converts May-2029. Closed Jun 5th.
- $40B at-the-market (ATM) program. This will be sold gradually from Q3-2026. It’s designed to mimic a “sell to cover” model for employee stock vests to prevent open-market dilution. Instead, Alphabet will settle tax obligations with corporate cash and issue equivalent equity.
- $10B private placement (upfront) to Berkshire at a 6% discount
Of the $84.75B program, the total hard cash raised and settled up to now stands at $44.75 billion (net cash hitting the balance sheet is ~$44.4B). The remaining $40B belongs to the ATM program, which has not yet commenced.
Google’s entire equity raise history is below:
| Event | Security | Shares | Price ($) | Gross proceeds ($) |
|---|---|---|---|---|
| 2004 IPO — 19-Aug-2004 — total offering $1.67B | ||||
| Primary — sold by Google | Class A common | 14,142,135 | $85.0000 | $1,202,081,475 |
| Secondary — selling stockholders | Class A common | 5,462,917 | $85.0000 | $464,347,945 |
| IPO TOTAL | $1,666,429,420 | |||
| 2005 Follow-on — 14-Sep-2005 — total offering $4.18B | ||||
| Follow-on — sold by Google | Class A common | 14,159,265 | $295.0000 | $4,176,983,175 |
| 2026 Equity Capital Raise — priced 2-Jun-2026 — $84.75B (upsized from $80B) | ||||
| Underwritten — Class A common | Class A common | 25,459,689 | $355.1982 | $9,043,235,705 |
| Underwritten — Class C capital | Class C capital | 25,459,689 | $351.8018 | $8,956,764,418 |
| Series A mandatory conv. pref | Dep. shares (1/20th of 6.25% Series A MCP) | 167,500,000 | $50.0 | $8,375,000,000 |
| Series B mandatory conv. pref | Dep. shares (1/20th of 6.25% Series B MCP) | 167,500,000 | $50.0 | $8,375,000,000 |
| At-the-market program | Class A + Class C over time | - | - | $40,000,000,000 |
| Private placement — Berkshire Hathaway | Common stock | - | - | $10,000,000,000 |
| 2026 PROGRAM TOTAL | $84,750,000,123 | |||
Google’s Debt Raise
For me, the most notable bond raise by Google was on Feb 13th 2026, when it raised a GBP £1 billion bond maturing in 100 years (6.125% coupon). The century bond was nearly ten times oversubscribed, with no meaningful restrictive covenants.
Google’s entire debt raise history is below:
| Deal date | Currency | Principal (local) | Coupon | Maturity | Tenor |
|---|---|---|---|---|---|
| 16-May-2011 | USD | $1B | 1.250% | 2014 | 3-yr |
| $1B | 2.125% | 2016 | 5-yr | ||
| Feb-2014 | USD | $1B | 3.625% | 2021 | 10-yr |
| $1B | 3.375% | 2024 | 10-yr | ||
| 09-Aug-2016 | USD | $2B | 1.998% | 15-Aug-2026 | 10-yr |
| $1B | 0.450% | 2025 | 5-yr | ||
| $1B | 0.800% | 2027 | 7-yr | ||
| $2B | 1.100% | 2030 | 10-yr | ||
| 05-Aug-2020 | USD | $1.25B | 1.900% | 2040 | 20-yr |
| $2.5B | 2.050% | 2050 | 30-yr | ||
| $2B | 2.250% | 2060 | 40-yr | ||
| 06-Nov-2025 | USD | $0.5B | Floating (FRN) | 2028 | 3-yr |
| $1B | 3.875% | 2028 | 3-yr | ||
| $2.5B | 4.100% | 2030 | 5-yr | ||
| $1.25B | 4.375% | 2032 | 7-yr | ||
| $3.5B | 4.700% | 2035 | 10-yr | ||
| $2B | 5.350% | 2045 | 20-yr | ||
| $4B | 5.450% | 2055 | 30-yr | ||
| $2.75B | 5.700% | 2075 | 50-yr | ||
| 06-Nov-2025 | EUR | EUR 1B | 2.375% | 2028 | 3-yr |
| EUR 1B | 2.875% | 2031 | 6-yr | ||
| EUR 1B | 3.125% | 2034 | 9-yr | ||
| EUR 1B | 3.500% | 2038 | 13-yr | ||
| EUR 1B | 4.000% | 2044 | 19-yr | ||
| EUR 1B | 4.375% | 2064 | 39-yr | ||
| 13-Feb-2026 | USD | $2.5B | 3.700% | 2029 | 3-yr |
| $3B | 4.100% | 2031 | 5-yr | ||
| $3B | 4.400% | 2033 | 7-yr | ||
| $4.25B | 4.800% | 2036 | 10-yr | ||
| $1.5B | 5.500% | 2046 | 20-yr | ||
| $4B | 5.650% | 2056 | 30-yr | ||
| $1.75B | 5.750% | 2066 | 40-yr | ||
| 13-Feb-2026 | GBP | GBP 0.75B | 4.125% | 2029 | 3-yr |
| GBP 1.25B | 4.625% | 2032 | 6-yr | ||
| GBP 1.25B | 5.500% | 2041 | 15-yr | ||
| GBP 1.25B | 5.875% | 2058 | 32-yr | ||
| GBP 1B | 6.125% | 2126 | 100-yr | ||
| ~03-Mar-2026 | CHF | CHF 905MM | 0.4270% | 2029 | 3-yr |
| CHF 700MM | 0.8900% | 2032 | 6-yr | ||
| CHF 575MM | 1.2525% | 2036 | 10-yr | ||
| CHF 455MM | 1.5823% | 2041 | 15-yr | ||
| CHF 420MM | 1.8675% | 2051 | 25-yr | ||
| 11-May-2026 | EUR | EUR 1.5B | 3.200% | 2030 | ~4-yr |
| EUR 1.75B | 3.450% | 2032 | ~6-yr | ||
| EUR 1.5B | 3.625% | 11-May-2034 | ~8-yr | ||
| EUR 1.75B | 4.100% | 2039 | ~13-yr | ||
| EUR 1.25B | 4.500% | 2045 | ~19-yr | ||
| EUR 1.25B | 4.800% | 2063 | ~37-yr | ||
| 11-May-2026 | CAD | C$1.5B | 3.650% | 15-May-2031 | ~5-yr |
| C$2.0B | 4.000% | 15-May-2033 | ~7-yr | ||
| C$2.25B | 4.350% | 15-May-2036 | ~10-yr | ||
| C$2.75B | 5.000% | 15-May-2056 | ~30-yr |
In May-2026, Alphabet deliberately lined up sterling and Swiss-franc issuance at the very long end to tap insurance and pension demand and diversify funding sources so as not to over-saturate the dollar market.
Meta
Meta has not issued equity. It now leans on the debt market, despite never having issued a bond until 2022. Excluding Hyperion, Meta has on-balance LT debt of $84B with a WAC of 5.19%.
Hyperion: In Oct-2025 Meta and Blue Owl closed $27 billion of financing for Meta’s Hyperion campus in Richland Parish, Louisiana. This was the largest private-credit transaction ever executed. Deal included:
- $27B in total development costs (A+ rated debt maturing 2049, ~225bps over Treasuries, anchored by PIMCO ~$18B and BlackRock ~$3B)
- and ~$2.5B equity
through an off-balance-sheet SPV, of which Meta holds a 20% equity-method interest and Blue Owl has 80%. SPV debt matures in 2049. Meta contributed land + construction-in-progress assets, and it received a one-time $2.55B distribution + signed a long-term operating lease back with a capped residual-value guarantee for the first 16 years. Blue Owl contributed ~$7B cash.
Meta’s history of debt issuance is shown below:
| Trade / Settle | Size ($B) | Coupon | Maturity | Tenor | Amount ($B) | WAC |
|---|---|---|---|---|---|---|
| Aug 2022 (debut) | 2.75 | 3.500% | 2027 | 5-yr | 10.0 | 4.04% |
| 3.00 | 3.850% | 2032 | 10-yr | |||
| 2.75 | 4.450% | 2052 | 30-yr | |||
| 1.50 | 4.650% | 2062 | 40-yr | |||
| 1 / 3 May 2023 | 1.50 | 4.600% | 2028 | 5-yr | 8.5 | 5.23% |
| 1.00 | 4.800% | 2030 | 7-yr | |||
| 1.75 | 4.950% | 2033 | 10-yr | |||
| 2.50 | 5.600% | 2053 | 30-yr | |||
| 1.75 | 5.750% | 2063 | 40-yr | |||
| 7 / 9 Aug 2024 | 1.00 | 4.300% | 2029 | 5-yr | 10.5 | 5.10% |
| 1.00 | 4.550% | 2031 | 7-yr | |||
| 2.50 | 4.750% | 2034 | 10-yr | |||
| 3.25 | 5.400% | 2054 | 30-yr | |||
| 2.75 | 5.550% | 2064 | 40-yr | |||
| 30 Oct / 3 Nov 2025 | 4.00 | 4.200% | 2030 | 5-yr | 30.0 | 5.14% |
| 4.00 | 4.600% | 2032 | 7-yr | |||
| 6.50 | 4.875% | 2035 | 10-yr | |||
| 4.50 | 5.500% | 2045 | 20-yr | |||
| 6.50 | 5.625% | 2055 | 30-yr | |||
| 4.50 | 5.750% | 2065 | 40-yr | |||
| 30 Apr / 4 May 2026 | 3.00 | 4.550% | 2031 | 5-yr | 25.0 | 5.73% |
| 2.00 | 4.875% | 2033 | 7-yr | |||
| 6.00 | 5.250% | 2036 | 10-yr | |||
| 4.00 | 6.200% | 2046 | 20-yr | |||
| 6.00 | 6.300% | 2056 | 30-yr | |||
| 4.00 | 6.450% | 2066 | 40-yr | |||
| Total senior notes | $84.0B | 84.0 | 5.19% |
Microsoft
Unlike most hyperscalers, Microsoft has largely stayed out of the bond market (has even delevered). It has been financing through off-balance-sheet partnerships. It is converting AI exposure into capex, leases, RPO and strategic equity economics.
Its last new bond issuance was back in Feb-2017. Bonded long-term debt fell -47.2% from the 2017 peak ($76.1B) to $40.2B by 2025 end. This reduction was despite long-term finance-lease liabilities growing from $9.0B (2020) to $43.0B (2025).
Total debt stands at $125.4B in Q1-26.
| Date | Action | Amount | Nature | Notes |
|---|---|---|---|---|
| Feb-2017 | Last new-money bond | ~$17B | New issuance | - |
| May-2020 | Debt exchange offer | ~$9.25B swapped | Liability management (no new cash) | Traded high-coupon notes for new 2.525% ‘50 and 2.675% ‘60 notes + cash |
| Mar-2021 | Debt exchange offer | (similar) | Liability management (no new cash) | Further coupon/maturity refinancing |
| 2023 (Activision) | Bridge loan commitment | $22.5B | Committed, undrawn | $69B deal funded with cash. No bonds issued |
| Sep-2024 | AI Infrastructure Partnership launched | ~$30B equity → up to $100B w/ debt | Equity into off-balance-sheet JV | Fund-level debt stays off Microsoft’s GAAP statements. BlackRock, GIP, MGX + Microsoft. |
| Oct-2025 | AIP acquires Aligned Data Centers | $40B | ~70% debt at fund level | Largest data-center deal ever. Debt not on Microsoft’s books |
The 2020 and 2021 transactions were exchange offers: In May 2020 Microsoft let holders swap higher-coupon notes (4.0% due 2055, 4.5% due 2057, 3.95% due 2056 and others) for a new series of 2.525% notes due 2050 plus a cash payment, capped at $6.25B of new 2050 notes, with a parallel pool creating 2.675% notes due 2060. That lowers interest cost and pushes out maturities; it raises no new money. March 2021 repeated the exercise.
When Microsoft bought Activision Blizzard for $69B (closed Oct-2023), it arranged a roughly $22.5B bridge facility but ultimately funded the deal with cash on hand and never tapped it.
In Sep-2024 it co-founded the AI Infrastructure Partnership with BlackRock, GIP and MGX, seeking to unlock $30B of equity capital that could mobilize up to $100B including debt financing, with Nvidia and later xAI as partners.
Its first major deployment was the $40B acquisition of Aligned Data Centers in Oct-2025 (the largest data-center deal ever) with about 70% debt leverage sitting at the fund level, so Microsoft contributes equity and gets AI exposure while the debt doesn’t appear on its books.
Oracle
Oracle has levered itself higher than anyone else on the list. Its senior debt is 2x since 2019. Unlike Meta or Microsoft, it carries the AI borrowing directly on balance sheet.
Oracle’s debt issuance history since COVID is below:
| Date | Instrument | Amount | Structure / tranches | Stated purpose |
|---|---|---|---|---|
| Feb 2020 | Senior notes outstanding | ~$51.6B | On balance sheet | Years of debt-funded buybacks |
| Apr 2020 | Senior notes | $20.0B | 6 tranches: $3.5B 2.5% ‘25, $2.25B 2.8% ‘27, $3.25B 2.95% ‘30, $3B 3.6% ‘40, $4.5B 3.6% ‘50, $3.5B 3.85% ‘60 | General corporate (buybacks, refi) |
| Mar 2021 | Senior notes | $15.0B | 6 tranches: $2.75B 1.65% ‘26, $2B 2.3% ‘28, $3.25B 2.875% ‘31, $2.25B 3.65% ‘41, $3.25B 3.95% ‘51, $1.5B 4.1% ‘61 | Refinance maturities + buybacks |
| Jun 2022 | Bridge term loan | $15.7B | 364-day delayed-draw | Fund $28.3B Cerner acquisition |
| Aug + Nov 2022 | Term loans | $5.7B | ($4.4B + $1.3B) Bank term loans | Cerner (prepay bridge) |
| 2022 | Senior notes | — | multi-tranche | Term out the Cerner bridge; Cerner maturities |
| Sep 2024 | Senior notes | ~$6.25B | 2029, 2034, and $1.25B 5.5% ‘64 tranches | Refinance 2024–25 maturities + commercial paper |
| Sep 2025 | Senior notes | $18.0B | 6 tranches: $3B 4.45% ‘30, $3B 4.8% ‘32, $4B 5.2% ‘35, $2.5B 5.875% ‘45, $3.5B 5.95% ‘55, $2B 6.1% ‘65 | AI / OCI data-center buildout |
| Late 2025–2026 | Vantage senior loan + further debt | ~$38B + ~$18B | Senior loan (data centers) + bonds | Stargate / OpenAI data centers |
Amazon
Amazon has not raised equity yet. It is a net stock repurchaser.
It is funding the largest corporate capex program ever attempted (2026: $200B, up 2x YoY) from operating cash + a long-term debt machine that is beginning to look more like that of a sovereign issuer than a corporate one.
As of 2019 end, Amazon had $23.4B of long term debt (including $16B of 2017 Whole Foods purchase debt). Since then, it has ramped up. Q1-26 long-term debt stood at $119B.
GAAP total debt including lease obligations reached ~$210B as Q1-26 end, up from $153B at 2025 end.
Let’s see how this happened. Amazon’s issuance history since 2019 is compiled below:
| Settled | Instrument | Amount | Structure |
|---|---|---|---|
| As of 2019 | — | $23.4B | Carried legacy notes (2014, 2017 Whole Foods) |
| Jun 2020 | Senior notes (USD) | $10.0B | 6 tranches: 0.400% ‘23; 0.800% ‘25; 1.200% ‘27; 1.500% ‘30; 2.500% ‘50; 2.700% ‘60 |
| May 2021 | Senior notes (USD) | $18.5B | 8 tranches: 0.250% ‘23; 0.450% ‘24; 1.000% ‘26; 1.650% ‘28; 2.100% ‘31; 2.875% ‘41; 3.100% ‘51; 3.250% ‘61 |
| Apr 2022 | Senior notes (USD) | $12.75B | 7 tranches: 2.730% ‘24; 3.000% ‘25; 3.300% ‘27; 3.450% ‘29; 3.600% ‘32; 3.950% ‘52; 4.100% ‘62 |
| Dec 2022 | Senior notes (USD) | $8.25B | 5 tranches: $1.25B 4.700% ‘24; $1.25B 4.600% ‘25; $2.0B 4.550% ‘27; $1.5B 4.650% ‘29; $2.25B 4.700% ‘32 |
| Jan 2023 | 364-day term loan | $8.0B | SOFR + 0.75%; short-term |
| Nov 2025 | Senior notes (USD) | $15.0B | 6 tranches: $2.5B 3.900% ‘28; $2.5B 4.100% ‘30; $1.5B 4.350% ‘33; $3.5B 4.650% ‘35; $3.0B 5.450% ‘55; $2.0B 5.550% ‘65 |
| Mar 13, 2026 | Senior notes (USD) | $36.9B | 11 tranches: 2 floating (SOFR+0.44% ‘28, +0.59% ‘29) + 9 fixed, 3.850%–6.050%, to 2076 |
| Mar 16, 2026 | Senior notes (EUR debut, separate) | €14.47B (~$16.8B) | 8 tranches: €1.75B floating (EURIBOR+0.35%) ‘28 + 7 fixed, 2.800%–4.850%, to 2064 |
| ~Jun 8, 2026 | Senior notes (CAD/Maple debut) | C$14.0B (~$10B) | 5 tranches, 2029–2056, ~3.400%–5.000% |
| Jun 8, 2026 | Delayed-draw term loan (DDTL) | $17.5B | Senior unsecured; draw thru Sep 30, 2026; each draw matures +3yr; SOFR +0.625–0.875% |
In Jun-2020 Amazon raised $10B across six tranches at record-low rates. A 0.4% 3-year note was a record low for a corporate bond at the time.
Its May-2021 deal of $18.5B in eight tranches was its largest bond issue in history then, followed by a April 2022 sale of $12.75B in 7 parts. In Jan-2023, Amazon took an $8B unsecured 364-day term loan, citing the ‘uncertain macro environment’.
AI prevalence ramped up the pace. The Nov-2025 $15B deal was its first dollar bond since 2022. Since Nov-2025, they’ve raised ~$82B vs ~$41B across the 2020-2022 stretch.
March 2026 stole the show: Amazon set a record for the largest corporate bond sale in history, raising $53.8 billion across 19 tranches and two currencies over two days:
- a $37 billion US dollar leg and
- a €14.5 billion euro component
This beat Verizon’s $49 billion record that had stood since 2013. It was the fourth-largest US corporate bond sale on record and the biggest not tied to an acquisition. Part of the rationale was funding their OpenAI equity investment.
Amazon kept going. In Jun-2026, it issued a record CAD $14bn Canadian “maple” bond (five tranches, 2029–2056). This was the largest corporate bond ever in that market (beating Alphabet’s CAD $8.5B from May-2026).
This was immediately followed by a $17.5B delayed-draw term loan (Citigroup as admin agent) available through end-September, each draw maturing three years out at SOFR +0.625–0.875%. Quite a nice top-up for cash flow needs.
OpenAI
OpenAI’s capital is majorly equity raised in private rounds. It has gone from Microsoft’s initial $1B in 2019 to a $122B round that closed in Mar-2026. Cumulatively, reporting put OpenAI’s total raised at $57B+ by mid-2025, before the $122B round dwarfed everything prior.
| Date | Type | Amount | Valuation | Notes |
|---|---|---|---|---|
| 2021 | Secondary | — | ~$14B | Employee sale (a later secondary ~$20B) |
| Jan-23 | Primary | ~$10B | — | Microsoft. ~$13B total 2019–23 |
| Apr-23 | Secondary | ~$0.3B | $27–29B | Sequoia, a16z, Thrive, K2, Founders Fund |
| Feb-24 | Secondary | — | ~$86B | Thrive-led tender |
| Oct-24 | Primary | $6.6B | $157B | Thrive (Microsoft, Nvidia, SoftBank). convertible notes |
| Oct-24 | Debt | $4.0B (undrawn) | — | 9-bank revolver. +$2B option. At 6%. |
| Mar-25 | Primary | $40B | $300B | SoftBank ($30B) + $10B syndicate; tranched |
| Oct-25 | Secondary | — | ~$500B | Employee sale |
| Feb–Mar 2026 | Primary | $122B (closed) | $852B | Amazon $50B; Nvidia $30B; SoftBank $30B; +$3B retail |
| Mar-26 | Debt | ~$4.7B (undrawn) | — | Revolver expanded; global bank syndicate |
It closed the latest round on Mar 31st 2026 with $122B committed capital at an $852B post-money valuation. This is the largest private fundraise in history. It built on a $110B tranche announced Feb 27th at $730B pre-money ($840B fully diluted) valuation (the round stayed open and CFO Sarah Friar confirmed an additional $12B came in before close).
Amazon committed $50B in the round, of which $35B is contingent on OpenAI going public or meeting technological milestones. SoftBank’s $30B is payable in tranches across April, July, and October 2026, backed by a $40B bridge loan it secured March 27. By late March, OpenAI had received roughly $25B in immediate capital ($15B Amazon + $10B SoftBank tranche one).
For the first time, OpenAI raised more than $3B from individual investors through bank channels, and disclosed an expanded but undrawn revolving credit facility of about $4.7B (OpenAI’s only conventional debt).
OpenAI’s infrastructure commitments total > $1.15 trillion across seven vendors for 2025–2035: Broadcom $350B, Oracle $300B, Microsoft $250B, Nvidia 10 GW equivalent, AMD $90B, Amazon AWS $38B, and CoreWeave $22B.
| Partner | Officially Disclosed Commitments |
|---|---|
| Microsoft Azure | OpenAI contracted an incremental $250B of Azure services. Signed alongside OpenAI’s October 2025 for-profit conversion, when OpenAI dropped Microsoft’s cloud right-of-first-refusal. |
| Amazon | Initial $38B AWS agreement. Later expanded by $100B over 8 years, including ~2GW of Trainium capacity. |
| CoreWeave | Contracts up to ~$22.4B through 2029. $11.9B initial (Mar 2025), + $4B (May 2025), + $6.5B (Sep 2025). OpenAI holds ~7% equity stake. |
| NVIDIA | At least 10GW of NVIDIA systems (“millions” of GPUs). NVIDIA intended to invest up to $100B in OpenAI progressively as each GW is deployed. This is now replaced by the $30B equity check. |
| AMD | 6GW AMD Instinct GPU agreement. Initial 1GW MI450 deployment in H2-2026; OpenAI warrant for up to 160MM AMD shares, vesting with deployment milestones + share price milestones (final tranche at $600). For AMD, this is ~$100B potential revenue over time. |
| Broadcom | 10GW custom AI accelerator collaboration; deployment targeted from H2-2026 through end-2029. Terms undisclosed. The Financial Times estimates $350B–$500B. |
| Oracle / Stargate | OpenAI says Stargate with Oracle adds 4.5GW and total Stargate development exceeds 5GW, toward a $500B / 10GW four-year White House commitment. Now ~7GW planned and >$400B over three years. The Abilene flagship (1.2GW) is live. |
The Funding Circularity
OpenAI is raising unprecedented private capital while simultaneously signing multi-year compute, GPU, and custom-silicon agreements with the same ecosystem that is financing, supplying, and hosting it. Some of these arrangements are firm dollar contracts, while others are GW-scale supply agreements with undisclosed economics.
The mechanism: a small cluster of chipmakers and clouds invest in the AI labs that then spend the money buying those same investors’ chips and cloud capacity, so the cash loops among a handful of interconnected firms.
As mentioned, OpenAI’s infrastructure commitments total roughly $1.15 trillion across seven vendors for 2025–2035: Broadcom $350B, Oracle $300B, Microsoft $250B, Nvidia $100B, AMD $90B, Amazon AWS $38B, and CoreWeave $22B.
The reverse leg is what makes it circular: Nvidia said it would invest up to $100 billion in OpenAI, which plans to fill its data centers with Nvidia chips. Amazon held talks to invest at least $10 billion alongside a deal for OpenAI to buy AWS compute; and OpenAI took a 7% stake in CoreWeave while agreeing to buy billions of capacity. AMD took it furthest, granting OpenAI warrants for up to ~10% of AMD in exchange for purchase commitments.
More than $800B of these arrangements circle a small cohort, which can inflate apparent demand and make revenue look organic when much of it is the same dollars going around.
The issue is lower-quality demand visibility. A substantial portion of the backlog is economically linked to financing, warrants, supplier investment, or cloud-credit commitments, so reported revenue may be less independent than in a normalized customer-demand state.
Apple
Apple has issued no equity. It is among the largest share repurchasers in history. Since just 2019, it has spent $576.2B repurchasing common stock.
Its debt balance has been going down since COVID. Long term debt stood at $113B in Mar-2022 and $82.7B in Mar-2026.
Apple’s cost of debt re-rated from 1.84% (Feb-2021) and 2.12% (Aug-2021) at near-zero rates to 4.35% (May-2023) and 4.32% (May-2025). It’s annual issuance shrank from $14B to $4.5B. Apple only borrows when cheap debt beats its cost of capital for funding buybacks.
Since 2019, it has issued $54B in gross debt ($51.3B + €2.0B). It’s detailed issuance breakdown is below:
| Date | Instrument | Gross proceeds (billions) | Ccy | Net proceeds (billions) | # Tranches | Maturities | WAC |
|---|---|---|---|---|---|---|---|
| Sep 2019 | Senior notes | 7.0 | USD | 6.96 | 5 | 2022–2049 | 2.2% |
| Nov 2019 | Green bond | 2.0 | EUR | 1.99 | 2 | 2031–2034 | 0.25% |
| May 2020 | Senior notes | 8.5 | USD | 8.42 | 4 | 2023–2050 | 1.59% |
| Aug 2020 | Senior notes | 5.5 | USD | 5.45 | 4 | 2025-2060 | 1.77% |
| Feb 2021 | Senior notes | 14.0 | USD | 13.9 | 6 | 2026–2061 | 1.84% |
| Aug 2021 | Senior notes | 6.5 | USD | 6.46 | 4 | 2028–2061 | 2.12% |
| May 2023 | Senior notes | 5.3 | USD | 5.2 | 5 | 2026–2053 | 4.35% |
| May 2025 | Senior notes | 4.5 | USD | 4.47 | 4 | 2028–2035 | 4.32% |
SpaceX
SpaceX completed the largest IPO in history on June 12, 2026 (selling 555.6 million shares at $135 apiece, raising $75B at $1.75T). Also, as of yesterday, it priced a $25B inaugural bond.
SpaceX’s total issuance history (equity + debt) since 2015 is shown below:
| Date | Type | Amount ($B) | Implied valuation ($B) | Notes |
|---|---|---|---|---|
| 2015 (Series G) | Primary | $1.0 | $12 | Google + Fidelity buy ~10%; ~$1B at ~$10–12B |
| May 2019 | Primary | $1.0 | $33 | Series J tranche |
| Aug 2020 | Primary | $1.9 | $46 | Series N — largest single primary round |
| Feb 2021 | Primary | $1.2 | $74 | Sequoia, Coatue, Fidelity |
| Jun 2022 | Primary | $1.7 | $127 | Led by Mirae Asset |
| Dec 2022 | Secondary / tender | $140 | ||
| Jan 2023 | Primary | $0.8 | $137 | a16z — LAST primary equity round before IPO |
| Oct 2023 | Debt | Conventional bank debt | ||
| Jun 2024 | Secondary / tender | $210 | Approx. range | |
| Dec 2024 | Secondary / tender | $1.3 | $350 | Up to ~$1.25B; ~$500M buyback included |
| Jul 2025 | Secondary / tender | $1.0 | $400 | ~$1B deal |
| Dec 2025 | Secondary / tender | $2.6 | $800 | Up to $2.56B; CFO Johnsen letter, 12-Dec-2025 |
| Feb 2026 | M&A (all-stock) | $1,250 | Acquired xAI: SpaceX ~$1,000B + xAI ~$250B = ~$1.25T combined. 0.1433x exchange; largest private merger ever. | |
| 12 Jun 2026 | IPO (primary) | $75.0 | $1,770 | Nasdaq: SPCX. 555.6M shares @ $135. Largest IPO ever (vs Aramco ~$25–30B). Opened $150, Day-1 close $160.95 (~$2.1T). Goldman lead, 21-bank syndicate. |
| 23 Jun 2026 | Debt | $25.0 | Inaugural senior unsecured notes, 5 tranches. Settles 26-Jun-2026. |
The new $25B debt deal has the following tranches:
| Tranche | Coupon | Maturity | Amount ($B) |
|---|---|---|---|
| Senior Notes due 2031 | 5.350% | 2031 | $7.0 |
| Senior Notes due 2033 | 5.650% | 2033 | $6.0 |
| Senior Notes due 2036 | 5.875% | 2036 | $6.0 |
| Senior Notes due 2046 | 6.600% | 2046 | $2.5 |
| Senior Notes due 2056 | 6.650% | 2056 | $3.5 |
| Total issuance | 5.855% | $25.0 |
Changes in FCF
A company’s worth its discounted sum of cash flows to its owners. An owner only gets sustainably distributable cash flows when the company makes more than it spends, not if it has a funding gap. Let’s assess these companies’ funding cash gap, defined here as:
Funding gap = capex − operating cash flow
This is a crude and oversimplified method relying on GAAP. The proper treatment would be to reconstruct each company’s owner earnings to great granularity. That depth, across a group with $17.8T of combined market cap, goes beyond the scope of this note (Alphabet $4.2T, Amazon $2.55T, Meta $1.44T, Microsoft $2.79T, Oracle $0.48T, Apple $4.35T, SpaceX: $2T).
| Company | Revenue ($B) | FCF Margin | ||||
|---|---|---|---|---|---|---|
| 2023 | 2024 | 2025 | 2023 | 2024 | 2025 | |
| Oracle (FY) | $50 | $53 | $57 | 17.0% | 22.3% | -0.7% |
| Amazon | $575 | $638 | $717 | 6.4% | 6.0% | 1.6% |
| Meta | $135 | $165 | $201 | 31.9% | 31.7% | 21.7% |
| Alphabet | $307 | $350 | $403 | 22.6% | 20.8% | 18.2% |
| Microsoft (FY) | $212 | $245 | $282 | 28.1% | 30.2% | 25.4% |
| Apple | $383 | $391 | $416 | 26.0% | 27.8% | 23.7% |
| SpaceX | $8.7 | $14 | $18.7 | - | - | - |
The one trend that is common is everyone’s FCF cushion is vanishing at different speeds.
| Name | FCF (2023 to 2025) | Status |
|---|---|---|
| Oracle (FY) | +$11.7B → −$0.4B → −$23.7B | Already underwater. Structural deficit widening, BBB-rated |
| Amazon | +$32B → +$33B → +$7.7B | Cushion nearly gone. capex ~$200B vs $139B OCF |
| Meta | +$44B → +$52B → +$43.6B | Compressing. 2026 capex guide ($125-145B) likely flips it |
| Alphabet | +$69B → +$73B → +$73B | Still self-funds, but FCF margin ~18%→~5% as 2026 capex doubles to ~$180B |
| Microsoft (FY) | +$60B → +$74B → +$71B | Least stressed. Mitigating cash burn |
Oracle is already funding externally. Amazon is approaching the edge of internally funded capex. Meta is moving into structured financing. Alphabet has pre-funded with equity despite massive cash generation. And Microsoft remains the best-positioned, but the FCF multiple is deeply compressed vs pre-AI.
Closing
Building AI is extraordinarily expensive and advanced chips are scarce, so pairing long-term buying commitments with financing is how you lock in supply.
The demand that justifies the whole outlay is, by the companies’ own disclosures, partly circular. It’s vendor-financed, warrant-linked, and cloud-credited among the same seven firms.
Maybe the circular funding relationships feed a virtuous circle lining up suppliers, builders and customers. This has occurred in the past. Railroads, telecom and earlier compute build-outs all used suppliers financing customers to bootstrap genuine markets.
The contagion risk exists. For example: if Microsoft’s AI monetization disappoints, it could cut Azure spend, hitting Nvidia’s revenue, denting CoreWeave’s valuation, and circling back to OpenAI’s funding capacity. The strain will show up first in the infrastructure providers’ balance sheets (rising debt + lease commitments).
I saw an interview with Goldman’s ex-CEO Lloyd Blankfein (on a show called My First Million) talking about the recent AI fundraising, including by Google. He (unsurprisingly) agreed with the fundraising. He said the raise from Google and other hyperscalers were prudent as it was future funding management in a good equity environment.
He said: “You raise capital when it’s available, not when you need it”. When the market turns or a company faces an unexpected systemic shock, liquidity dries up instantly. I agree with the prudent nature of the CFOs to raise this cash now. I just don’t know which ones will earn a return on deployed cash worthy of the immense outlays with no current end to the spend in sight.