How to invest in OpenAI - the proxy basket (MSFT, SFTBY, AMZN, NVDA, ARKVX)

The standard answer to "how do I buy OpenAI stock" is "you can't - buy Microsoft." That answer is two years stale. Since late 2025 the OpenAI exposure map has inverted: an S-1 is confidentially on file with the SEC, retail money has already been let in once (a $3B retail tranche inside the March 2026 round), SoftBank holds roughly 13% of the company through an ADR anyone can buy, Amazon wrote the largest check of the largest private round in history, and two retail funds hold the stock directly.

This piece walks through OpenAI's valuation arc ($157B in October 2024 to $852B in March 2026), the actual IPO timeline (S-1 filed 2026-06-08, listing guided to 2027), and the proxy basket retail can buy today: $MSFT (~27% of the company), $SFTBY (~13%, the leveraged pure-play), $AMZN (up to $50B committed), $NVDA (investor and vendor at once), $ARKVX (9.2% OpenAI, no accreditation required), and $VCX. For accredited investors, secondaries trade OpenAI paper directly: Forge printed $721.85/share on 2026-08-29.

Why OpenAI matters to public-equity investors

OpenAI is the largest frontier AI lab by revenue: roughly $25B expected for 2026, up from about $13B in 2025, against a forecast full-year loss around $14B, with positive cash flow not expected before the end of the decade. The company builds the GPT model family, runs ChatGPT as the largest consumer AI surface in the world, and sells API + enterprise access on top.

For public-equity investors, OpenAI's own P&L is not the point. Two mechanics are:

1. The stake markup. Microsoft's $13B of cumulative investment converted, in the October 2025 restructuring, into roughly 26.8% of OpenAI Group PBC. That position was marked near $135B at the $500B valuation and roughly $228B after the March 2026 round. The same fair-value mechanic QA documented on the Anthropic side (where half of AMZN's and GOOGL's Q1 2026 "AI profits" were stake markups) now runs through MSFT, SoftBank, and, since March, AMZN and NVDA.

2. The spend commitment web. OpenAI has committed roughly $1.15 trillion of infrastructure spend across 2025-2035 to seven vendors. One private company is the single largest customer in the AI supply chain. That makes "OpenAI exposure" a property of half the semiconductors bubble, whether you asked for it or not.

The OpenAI valuation arc

| Event | Date | Post-money / mark | Read | | --- | --- | --- | --- | | Microsoft initial check | 2019 | ~$1B invested | The founding strategic bet | | Microsoft $10B | 2023-01 | ~$29B | The round that funded GPT-4 scale | | Thrive-led primary | 2024-10 | $157B | $6.6B raise | | SoftBank-led round | 2025-03 | $300B | $40B, fully funded by Dec 2025 | | Restructuring to PBC | 2025-10 | ~$500B (employee tender) | MSFT profit-share converts to ~27% equity | | $122B round | 2026-03-31 | $852B | Largest private round ever: AMZN up to $50B, NVDA $30B, SoftBank $30B, retail $3B | | Employee tender | 2026-08 | $852B | $7B at an unchanged mark | | S-1, confidential | 2026-06-08 | targets above $1T | Goldman + Morgan Stanley lead |

Two observations. First, the markup velocity: 5.4x in eighteen months, 2.8x in the last twelve. Second, the round mechanics are increasingly supplier-financed: NVDA's $30B equity check sits next to its GPU sales to the same customer, AMZN's $50B sits next to a $38B AWS commitment flowing back. Bears call the loop circular financing; bulls call it vertical integration by contract. Either way the marks are set by parties with a commercial interest in the number.

The IPO is filed. The clock is real this time

Unlike Anthropic a year ago, OpenAI's IPO is not a rumor: the company confidentially filed its S-1 on 2026-06-08 targeting a valuation above $1 trillion, with Goldman Sachs and Morgan Stanley leading. The original September 2026 window slipped; on 2026-08-19 CFO Sarah Friar told employees the company expects to list in 2027, sooner if growth keeps accelerating. Friar has also said the IPO will reserve an allocation for retail investors, consistent with the $3B retail tranche the banks placed in March.

The proxy math changes at listing: every vehicle below converts from "exposure to a private mark" into "exposure to a printed price," and the proxy premium (or discount) compresses. Until then, the basket is the trade.

The proxy basket - public-equity exposure to OpenAI

Every name here trades from any retail brokerage; the stack QA actually uses to hold them is documented at /stack.

1. Microsoft (MSFT) - ~27% of the company, a ~$228B mark

The deepest proxy, and the one the market has priced the longest. Microsoft invested roughly $13B across 2019-2023; the October 2025 restructuring converted its profit-share rights into approximately 26.8% of OpenAI Group PBC. At the $852B mark that position is worth roughly $228B, a 17x multiple on cash deployed, and it now moves MSFT's reported income the way Anthropic's marks move AMZN's. OpenAI also committed an incremental $250B of Azure spend as part of the restructuring, so the operating business and the stake compound the same narrative.

Caveats: the March round diluted the percentage (no updated cap table has been published), Azure lost compute exclusivity in the restructuring, and OpenAI's stake is still under 7% of MSFT's own market cap: this is exposure, not concentration. Track MSFT live: /stocks/msft.

2. SoftBank (SFTBY) - the ~13% pure-play, with leverage attached

The most OpenAI per dollar of market cap on any exchange. SoftBank's cumulative commitment reaches about $64.6B for roughly 13%: the $40B 2025 round fully funded by December 2025, plus $30B across 2026 in three $10B tranches (April, July, and October 1 still pending). The stake is large relative to SoftBank itself, and it is levered: a $40B bridge loan financed the 2026 tranches, a $10-20B bond issuance was announced 2026-08-26 to refinance it, and a separate $10B loan is secured against the OpenAI stake itself.

That is the read: SFTBY (the US ADR of 9984.T) is a leveraged holding company whose OpenAI position now competes with ARM as its defining asset. You get the stake, plus ARM, plus Vision Fund marks, plus yen exposure, plus the debt. Not in the QA universe yet, so no live page to link; sizing it small is the point.

3. Amazon (AMZN) - the largest check of the largest round

New to the OpenAI cap table in March 2026, with the round's biggest commitment at up to $50B, alongside OpenAI's $38B AWS compute commitment flowing the other way. The structural kicker: AMZN is now the only public company with major equity in both frontier labs, holding the largest external Anthropic position (a $70B+ mark) and a top-three new OpenAI position at once. The two-lab proxy is one ticker. For the Anthropic half of that story, see How to invest in Anthropic. Track AMZN live: /stocks/amzn.

4. Nvidia (NVDA) - investor, vendor and bottleneck at once

NVDA put $30B into the March round, on top of a separate agreement to invest up to $100B progressively as OpenAI deploys 10GW of NVIDIA systems, first phase on Vera Rubin in H2 2026. The circularity is the caveat: equity dollars go in, GPU purchase orders come back out. NVDA is less an OpenAI proxy than the tollbooth OpenAI's roadmap runs through, which is its own thesis. Track NVDA live: /stocks/nvda.

5. ARK Venture Fund (ARKVX) - direct exposure, no accreditation

The retail vehicle with the most direct OpenAI line: 9.2% of the fund as of mid-2026, its second-largest position behind SpaceX (13.8%), and it holds Anthropic too, so one ticket carries both labs plus the pre-IPO cohort (Stripe, Tenstorrent). Caveats: interval-fund liquidity (quarterly repurchase windows, not daily), fees, and NAV marks that lag primary rounds.

6. Fundrise Innovation Fund (VCX) - the other retail sleeve

VCX holds OpenAI alongside its Anthropic and SpaceX positions, with no accreditation requirement. Weights shift by disclosure cycle and the same NAV-lag caveats apply as covered in the Anthropic piece. Useful as a small slice of the basket, not as the core.

Secondary markets (accredited investors only)

For accredited investors, OpenAI paper trades directly: Forge quoted $721.85/share on 2026-08-29 and Hiive $737.24 on 2026-08-27. Mechanics: OpenAI's transfer restrictions are notoriously tight, so much of the flow is SPV interests layered on top of actual units rather than the units themselves; minimums typically run $25-100K, fees 3-5%, and the accreditation bar is $1M net worth excluding primary residence or $200K+ income. The spread between the secondary print and the $852B round mark is a live sentiment gauge: watch whether it widens or compresses into the listing.

The supplier web - OpenAI exposure without the equity

The $1.15T of 2025-2035 commitments makes a second basket possible: companies whose revenue, not equity, is the OpenAI exposure.

| Vendor | Commitment | Status (mid-2026) | | --- | --- | --- | | $AVGO | ~$350B, custom accelerators | Design phase complete, first deployment H2 2026 | | $ORCL | ~$300B over 5 years (Stargate) | GB200 racks live in Abilene, hundreds of thousands of GPUs operational | | MSFT | ~$250B Azure | Part of the restructuring agreement | | NVDA | ~$100B (10GW systems) | Vera Rubin first phase H2 2026 | | $AMD | ~$90B, 6GW Instinct | First 1GW rolling out H2 2026; OpenAI holds warrants for up to 160M AMD shares | | AMZN (AWS) | ~$38B | Signed alongside the equity check | | $CRWV | ~$22B | Incremental capacity contracts |

Two reads. First, the AMD line is the strangest structure in the map: the customer holds warrants on the supplier, so OpenAI upside leaks into AMD equity by contract, roughly 10% of the company if milestones hit. Second, these are commitments against a counterparty forecasting a $14B annual loss: the hyperscalers and semiconductor names carrying them own the counterparty risk as well as the revenue. Live pages: /stocks/avgo, /stocks/orcl, /stocks/amd, /stocks/crwv.

OpenAI vs Anthropic - the symmetry, updated

A year ago the pattern was "MSFT is to OpenAI what AMZN is to Anthropic." The 2026 update is messier and more interesting. Both labs now have live listing tracks (Anthropic targeting H2 2026, OpenAI guided to 2027). Both are supplier-financed at the margin. And the clean pairing broke: AMZN now sits on both cap tables, NVDA invests in the lab it supplies, and SoftBank turned itself into the closest thing to a listed OpenAI tracking stock. A basket of MSFT + AMZN + $GOOGL + NVDA + ARKVX captures effectively the entire public-equity expression of frontier-lab exposure across both companies; the marginal decision is how much SFTBY leverage you want on top.

What to watch

• The S-1 going public. The confidential filing becomes a real prospectus with share classes, lockups and the retail allocation Friar promised. That document reprices every proxy in this piece.
• 2027 vs late-2026. Friar's "sooner if growth accelerates" is the only conditional that matters; a pulled-forward listing compresses the proxy premium fast.
• SoftBank's October 1 tranche and the bond pricing. The final $10B installment and the $10-20B refinancing print are a public stress-test of the private mark. A hard bond book is a read-through on how credit markets value the stake.
• H2 2026 deployment milestones. NVDA's Vera Rubin first phase, AMD's first 1GW, AVGO's first custom silicon. Misses pressure the whole supplier web; hits validate the $1.15T commit stack.
• The Forge/Hiive spread vs $852B. Secondary prints above the round mark say the retail bid is ahead of the institutions; below it, the pre-IPO froth is cooling. Bubble shifts and rule-based alerts on the proxy names are part of /pro.

Live data on the proxy basket: /stocks/msft · /stocks/amzn · /stocks/nvda · /stocks/amd · /stocks/avgo · /stocks/orcl · /stocks/crwv - price, ETF holdings, bubble correlation, bot positions.

Bubble context: /bubbles/hyperscalers - the cluster the cloud-platform proxies belong to and how it's moving.

QuantAbundancia is educational research. Nothing here is investment advice. See /disclosures.