Fund Tax Architecture: Who Pays, Where and On What

Concept

Fund tax architecture is the design discipline that decides who is the taxpayer at each link of the chain through which investment income travels from a portfolio asset to an investor's account. It is not a choice of a low-tax domicile: the domicile answers only the vehicle-level question, and the vehicle is one of four levels. A fund's legal structure allocates cash — who receives which distribution and when. Tax architecture allocates something different: the character of income (dividend, interest, business profit, capital gain), the timing of its recognition (allocation years can precede cash years), and the identity of the person liable (the vehicle, the investor, or an interposed company). Cash allocation and tax allocation regularly diverge: a limited partner in a US partnership can owe tax on allocated profit in a year when the waterfall pays them nothing, which is why the waterfall mechanics and this page describe two different layers of the same fund.

The chain has a fixed shape. Income arises at an asset in a source country — a portfolio company pays a dividend, a borrower pays interest, a property is sold. The source country takes the first reading: it decides whether to tax the payment at source and whether a treaty reduces that tax. The income then passes through the fund vehicle — and possibly through a feeder (an entry vehicle for one investor group) or a blocker (an interposed corporation that changes who the taxpayer is). The vehicle's domicile takes the second reading: transparent, opaque or exempt. Finally the income reaches the LP, whose residence country takes the third reading — including anti-deferral regimes that can tax the LP before any cash arrives. A parallel, shorter chain runs from the fund to the manager and carry recipients: the management fee is service income where the people work, and carried interest is a profit share whose character depends on the recipient's country.

Three trade-offs organise everything that follows. First, transparency versus shielding: a transparent partnership gives investors one layer of tax and preserved income character, but exposes them to source-country filings and to income types their own status cannot absorb; an opaque corporation shields them and destroys both the single layer and the character. Second, treaty access versus simplicity: interposing a treaty-resident holding company can cut source withholding, but only if the company survives beneficial-ownership and principal-purpose scrutiny — an empty conduit fails and costs money anyway. Third, one architecture versus many investors: the same portfolio produces different tax outcomes for a domestic taxable investor, a tax-exempt institution and a foreign investor, so a single vehicle rarely fits all three, and every additional sleeve (feeder, blocker, parallel fund) buys investor-specific results at the price of entity-level leakage and administration.

The practical test of a finished architecture is that for every material payment in the structure you can name the payer, the recipient, the income character, the withholding applied, the treaty position and the report filed — before the first capital call. The sections below build that map level by level and then run one portfolio through three investor types to show the divergence in numbers.

Four taxpayers in one structure

Every fund structure contains at least four distinct tax positions, and optimising one of them in isolation is the classic design error. The table separates the levels and the mistake most often made at each.

The four levels are legally independent: a change at one level (say, inserting a blocker) shifts liabilities between levels but never makes a level disappear. That independence is why the rest of this page treats classification, source withholding, investor taxation and the manager chain as separate questions with separate answers.

Classification and transparency: the first question

Before any rate matters, each relevant country classifies each entity in the chain as transparent (income is attributed to the owners as it arises) or opaque (the entity is itself the taxpayer, and owners are taxed on distributions or disposal). The classification is made independently by the fund's domicile, by each asset country and by each investor country — and they do not have to agree. A limited partnership treated as transparent at home can be treated as a company by an investor's jurisdiction, and the reverse. These mismatches are not a curiosity: they decide who may claim a treaty on a dividend, who must file in the source country, whether a foreign tax credit reaches the person who bears the tax, and whether anti-hybrid rules deny a deduction somewhere in the chain. The mechanics of entity classification — elections where they exist, equity/debt characterisation, hybrid mismatches — are a discipline of its own; here it is enough that the architecture must be tested in every material country, and that the label used at home proves nothing abroad.

Nor is the home label permanent once investors are in place. Where associated non-resident entities hold in aggregate 50% or more of the voting rights, capital interests or profit rights in a hybrid entity established in an EU Member State, and their own jurisdictions regard that entity as a taxable person, Article 9a of the anti-tax-avoidance directive — inserted by Council Directive (EU) 2017/952 and applied by Member States from 1 January 2022 — treats the entity as a resident of that Member State and taxes it on income not otherwise taxed under the laws of any jurisdiction. The trigger sits in the investor register rather than in the fund's own documents: a Luxembourg SCSp can lose its transparency because of who subscribed. Article 9a(2) exempts a collective investment vehicle, defined cumulatively as a fund that is widely held, holds a diversified portfolio of securities and is subject to investor-protection regulation where it is established — three conditions a concentrated, closely held closed-ended fund can fail on any one of them. That is why the classification map is kept per investor cohort and refreshed as the register changes, not settled once at formation.

Within one structure the typical pattern is deliberate: the main fund is a transparent partnership (a Delaware LP, a Cayman ELP, a Luxembourg SCSp), so that investors get a single tax layer and income keeps its character; feeders and blockers are opaque corporations precisely because opacity is the feature being purchased; asset SPVs are local companies taxed where the asset sits. An exempt fund is a third state: the vehicle is formally a taxpayer, but a statutory regime lifts the tax subject to conditions. The exemption is always narrower than it sounds. A Luxembourg RAIF pays an annual subscription tax of 0.01% of net assets and, under Article 45(1) of the law of 23 July 2016, no other Luxembourg tax at fund level — which says nothing about withholding in asset countries or tax in investor countries. Singapore's 13O/13U regimes exempt specified income from designated investments subject to ongoing conditions, not every receipt of any vehicle. A Cayman fund is tax-neutral at vehicle level and precisely there. In each case the exemption removes one layer out of four.

One distinction saves a lot of confusion: tax transparency of an entity (who is taxed on its income) has nothing to do with tax transparency as information exchange — FATCA/CRS reporting applies to transparent and opaque funds alike, and an entity can be fully opaque for income tax while fully reportable for exchange purposes.

Source withholding and treaty access

The source country's tax is collected mostly through withholding: the payer deducts tax before the money leaves. For a fund the operative questions are which rate applies to which payment, and who in the chain is entitled to reduce it. The general mechanics of source taxation are mapped in the withholding-tax guide; what matters here is the fund-specific layer. Taking the US as the worked source country: US-source dividends, interest, rents and royalties paid to non-US persons are FDAP income, withheld under chapter 3 rules at a statutory 30% unless a treaty reduces the rate and the recipient documents its claim on the correct W-8 form. The same payment to a US person is not subject to that regime at all — which is the first illustration of the rule that a rate without a named recipient class is meaningless.

For a transparent fund, the treaty question is asked at the level of the partners, not the vehicle: a Cayman partnership has no treaty network, and each investor's own residence and documentation determine relief on income flowing through it. Interposing a treaty-resident holding company changes the claimant — but the claim only holds if the company is the beneficial owner of the income (it has real power to use it and is not a mere conduit obliged to pass it on) and if the arrangement survives the anti-abuse layer: the BEPS Multilateral Instrument has written a treaty-abuse minimum standard, typically the principal purpose test, into most bilateral treaties, so the applicable text is the treaty as modified by both parties' MLI positions, never the headline rate from a database. The PPT analysis is covered separately; the operational point for fund design is that a residence certificate begins the analysis and substance evidence ends it.

Two operational facts complete the picture. Relief can be applied at source (the custodian applies the treaty rate on payment) or by reclaim (statutory rate withheld, refund requested later): a structure whose relief exists only through reclaims the administrator cannot actually run has the statutory rate as its real rate, plus a financing cost for the delay. And withholding is a documentation chain, not a closing formality — a fund that cannot produce valid investor forms and intermediary statements at the moment of payment loses the relief its memo assumed.

Case note: the Danish beneficial-ownership judgments

Danish companies paid interest and dividends up to Luxembourg and Cypriot intermediates held by private equity funds; Denmark refused the directive exemptions. On 26 February 2019 the Court of Justice (Grand Chamber) ruled in joined cases C-115/16, C-118/16, C-119/16 and C-299/16 on the Interest and Royalties Directive and in joined cases C-116/16 and C-117/16 on the Parent-Subsidiary Directive.

Four holdings carry the analysis. The exemption is confined to beneficial owners, "the entities which actually benefit from that interest economically and accordingly have the power freely to determine the use to which it is put". Authorities and courts must refuse the benefit where there is abuse "even if there are no domestic or agreement-based provisions providing for such a refusal" — no local anti-abuse rule is needed. Abuse requires objective circumstances in which the purpose of the rules is not achieved despite formal compliance, plus the intention to create those conditions artificially; the indications include "the existence of conduit companies which are without economic justification and the purely formal nature of the structure of the group of companies, the financial arrangements and the loans". And the authority need not identify who the real beneficial owner is in order to refuse the claimant.

A fifth holding is aimed at fund vehicles directly: a Luxembourg SCA authorised as a SICAR could not be a "company of a Member State" entitled to the exemption if the interest it received was exempt from Luxembourg corporate income tax. The vehicle's own exemption was the fact that disqualified it.

The limits matter as much. These are rulings on EU directives, not on bilateral treaties, and they concern years before the MLI principal purpose test applied to those relations. What they change for design is the standard of proof: a company interposed for directive or treaty access has to receive income it is free to keep and use, its substance has to exist before the payment rather than after the assessment, and relief claimed at the top of a chain can fail on facts about the vehicle at the bottom.

The blocker: what it buys and what it costs

A blocker is a corporation deliberately interposed between a transparent fund (or asset) and an investor so that the investor owns shares in a company rather than a share of the underlying activity. Everything a blocker achieves follows from that one substitution. What flows through a partnership to its partners — business income, filing obligations, debt-financed income — stops at the corporation; the investor's world shrinks to dividends and share disposals. That is exactly what certain investors need: a US tax-exempt institution uses a blocker so that debt-financed or operating income is taxed to the corporation instead of becoming its own taxable income; a non-US investor uses one so that a US operating investment does not pull them into US tax returns. The specific US triggers are unpacked in the scenario section below; the blocker as a device is jurisdiction-generic and appears wherever transparent funds meet investors who cannot absorb flow-through income — the typical placements within a Delaware fund structure are described separately.

The price has a fixed anatomy, and each element should be priced before the entity is formed. The blocker pays entity-level tax on its income — for a US domestic blocker, federal corporate tax at the current 21% rate on its taxable income, plus any state tax. Cash leaving the blocker can face a second charge: dividends from a US corporation to non-US shareholders are US-source FDAP, withheld at 30% absent treaty relief; a foreign blocker with US effectively connected earnings faces instead the branch profits tax — 30% of the dividend equivalent amount, reduced by some treaties. Losses inside the blocker do not flow to investors; they wait for the blocker's own future income. Leverage helps only as far as interest deductions survive limitation rules. And the exit route decides a large part of the economics: selling the blocker's shares, selling the asset inside it and distributing, or liquidating produce different taxes for different parties — which is why blocker exits are designed at acquisition, together with the question of scope: a fund-level blocker taxes the whole portfolio to fix one asset, an asset-level blocker multiplies entities but confines the leakage to the position that needs it.

Feeders, parallel funds and AIVs

Blockers change the taxpayer; the remaining sleeve vehicles change the grouping of investors. A feeder fund aggregates one investor cohort and channels it into the master — a corporate feeder doubles as a blocker for that cohort, a partnership feeder stays transparent and only packages reporting, currency or eligibility. A parallel fund invests side by side with the main fund under an agreed allocation policy instead of through it — used when a cohort needs a genuinely different tax owner, leverage profile or asset perimeter, and workable only with strict mechanics: no cherry-picking of deals, blocker taxes borne solely by the cohort that benefits, arm's-length cross transfers. An AIV (alternative investment vehicle) is a per-deal detour: one investment that the main fund should not hold directly is placed in a dedicated vehicle with economics equivalent to the LPA's. The generic rule for all three: each additional entity adds administration, audit, FATCA/CRS classification, banking and directors, so the sleeve must be justified by a cohort or asset that actually needs it — a question of measured cost against measured exposure, exactly as with the blocker.

Three LPs, one portfolio

The divergence the architecture must absorb is easiest to see on a single portfolio. The scenario is deliberately US-bounded — the conclusions below hold for US assets under US rules and do not transfer to other source countries, where the same investor types meet different regimes. All numbers are a fictional teaching example.

A Delaware LP fund holds three US positions: (A) minority equity in listed US corporations paying dividends; (B) a leveraged US rental property held through a partnership, 60% financed by acquisition debt; (C) an operating logistics business organised as an LLC taxed as a partnership. Three LPs each hold 10%: LP1, a US taxable individual; LP2, a US university endowment (tax-exempt); LP3, a non-US corporate investor with no US operations of its own. In the year examined, each LP is allocated $100k of dividends from A, $100k of net rental income from B and $100k of operating profit from C.

The table shows three different funds hiding inside one: the identical portfolio produces three unrelated tax positions, and each LP's conclusion is self-contained.

LP1 (US taxable). The partnership is the natural wrapper: one layer of tax, character preserved, credits and losses flow through. The costs are timing and paperwork — tax follows the K-1 allocation, which can precede distributions (the cash-versus-tax gap that the waterfall page computes on the cash side), plus state filings where the portfolio operates. The one structural mistake available to LP1 is entering through an offshore corporate feeder: a foreign corporation whose income is mostly passive is a PFIC for US persons under the 75% income / 50% asset tests, and the default regime taxes distributions and gains at top rates with an interest charge unless a QEF or mark-to-market election is available. LP1 belongs in the domestic partnership, not in the offshore feeder built for others.

LP2 (US tax-exempt). Exempt status covers investment income — dividends, interest and capital gains are excluded from UBTI — but the partnership look-through treats LP2 as conducting the fund's businesses itself. Stream C is therefore taxable in full, and stream B is taxable in proportion to the acquisition debt. LP2's decision is quantitative: pay tax on its UBTI share directly, or invest through a corporate blocker that absorbs streams B and C at the corporate rate while converting LP2's return into dividends. The blocker is worth its 21%-plus-friction only if the UBTI otherwise generated is large enough; for a portfolio that is mostly stream A, it is pure cost.

On the same fictional numbers the choice is arithmetic. LP2's direct UBTI for the year is stream C in full ($100k) plus the debt-financed share of stream B (60% of $100k = $60k) = $160k; an exempt organisation that is a corporation is taxed on UBTI at corporate rates, so $160k × 21% ≈ $33.6k before the $1,000 specific deduction and any state tax, and LP2 files Form 990-T because gross unrelated income exceeds $1,000. Route B and C through a blocker instead and the corporation is taxed on all of both streams — $200k × 21% = $42k — after which the dividend up to a US exempt shareholder is not UBTI at all. On these facts the blocker costs $8.4k more than paying the tax directly. Move one variable — acquisition debt on the property from 60% to 100% — and the direct charge rises to $42k and the two routes meet. Below that crossover the blocker is not bought for tax saved but for what it removes: the 990-T filing, the debt-financed portion of the gain on a later sale, the risk that leverage rises mid-hold, and the effect of §512(a)(6), under which each unrelated trade or business is computed separately, so a loss in one cannot shelter income in another.

LP3 (non-US). The picture splits by income character, which is why no single answer to "what does a foreign investor pay on a US fund?" exists. Passive streams are a withholding matter: dividends at 30% or the treaty rate; much US-source interest can qualify for the portfolio interest exemption; gains on non-real-property stock are generally outside US tax for a foreign investor without a US business. Streams B and C are the opposite: they make LP3 a participant in a US trade or business, with a US return to file and §1446(a) withholding — 21% of allocated ECTI for corporate partners, 37% for non-corporate — taken on allocation whether or not cash is distributed. Real estate adds FIRPTA, and a secondary sale of the fund interest itself adds 10% withholding under §1446(f). A blocker sitting between LP3 and streams B–C replaces all of that with one corporate taxpayer — at the blocker price set out above. Whether LP3's home country then taxes the same income again, and whether its own anti-deferral rules attribute the blocker's profit currently, is a question of LP3's residence law, not of US law; a controlling stake in a low-taxed passive blocker is exactly the fact pattern CFC regimes exist for.

Case note: where the secondary-sale rule came from

A Greek mining company held an interest in a US LLC taxed as a partnership and had that interest redeemed. The IRS treated the gain as effectively connected on the strength of Revenue Ruling 91-32; the Tax Court disagreed, and the D.C. Circuit affirmed on 11 June 2019 (No. 17-1268), declining to defer to a ruling resting on "a single unreasoned sentence" and holding that the US-office rule looks to where the redemption itself occurred rather than to where the appreciation had been generated. Congress had already reversed the outcome prospectively: the Tax Cuts and Jobs Act of 2017 §13501(a)(1) enacted §864(c)(8), under which a foreign partner's gain on the sale of an interest in a partnership engaged in a US trade or business is effectively connected to the extent of that partner's distributive share of the gain that would have been effectively connected had the partnership sold all of its assets at fair market value on the date of the sale. The §1446(f) withholding in the map below is the collection device attached to that rule.

Two things follow for the architecture. The tax on a secondary sale is a function of the partnership's asset mix on the sale date rather than of anything the seller did, so a fund that acquires a single US operating business changes the exit position of every foreign LP already inside it. And the certification workflow before settlement is not administrative hygiene: it enforces a rule Congress wrote specifically to overturn a taxpayer win.

The same land, a different source country

The conclusions above are US conclusions. Move the fund's real estate to the United Kingdom and the policy is unchanged — non-residents are taxed on gains from land — while the taxpayer, the collection point and the available escape hatch all change.

The charge reaches through entities. Since 6 April 2019 a non-UK resident is taxable not only on a direct disposal of UK land but on an indirect one: the disposal of an interest in an entity that is UK property rich — 75% or more of the market value of its gross qualifying assets deriving from UK land, under TCGA 1992 Sch 1A Part 2 — where the seller also has a substantial indirect interest, a 25% investment held at any point in the two years before the disposal, which HMRC treats as insignificant only where it was held for 10% or less of that window (75 days or fewer). Because the 75% test runs on market value at the disposal date, the fund's valuation record becomes a tax input and not only an investor-reporting one.

Collection is the mirror image of FIRPTA. There is no transferee withholding: the non-resident seller reports and pays within 60 days of completion for completions on or after 27 October 2021, and must file even where there is no tax to pay or the disposal produced a loss. The US takes the money first and refunds against a return; the UK takes the return first and enforces afterwards — which moves the fund's operational problem from cash trapped in withholding to investor filings the administrator has to chase.

The UK also offers a lever the US does not. A UK property rich collective investment vehicle can make an exemption election under paragraph 12 of Schedule 5AAA, whose effect "is to move the point of taxation on gains on UK land from the relevant fund to its investors" — bought with reporting rather than with tax: information on investor disposals in the two years before the election (para 14) and a report for every period of account, delivered within 12 months of the end of that period (para 15), with revocation available to HMRC for serious failures. Where a US structure buys the investor-level result with a blocker and its corporate tax, a UK structure can buy it with an election and a reporting duty. Same policy, different price, different failure mode — which is why a conclusion carried across a border is a guess until the local rule has been read.

The manager and carry chain

The manager's tax position is a separate circuit that shares no answers with the investor chain. The management fee is service income of the management company, taxable where the company is resident and where its people actually perform the functions — a fee booked to an entity in a convenient jurisdiction while the deal team works elsewhere invites transfer-pricing reallocation of the profit and can create a permanent establishment where the people negotiating actually sit, and the choice of manager jurisdiction is a licensing and substance decision before it is a tax one. Fees also live in the indirect-tax world, and there the rule is delegated: Article 135(1)(g) of the VAT Directive exempts "the management of special investment funds as defined by Member States", so whether the same fee to the same fund family is exempt, taxable or outside the scope turns on which Member State's definition applies and where the service is supplied. Exemption is not automatically good news: an exempt supply carries no input-VAT recovery, so unrecovered VAT on the manager's own costs is final money rather than a timing item.

Carried interest is economically a profit share and legally an allocation through the carry vehicle, and its tax character depends on the recipient's country — there is no universal answer. In the US, carry allocated on an applicable partnership interest keeps long-term capital gain treatment only where the underlying gains meet the more-than-three-year holding period of §1061; gains from shorter holdings are recharacterised as short-term. The UK moves qualifying carry into an income-tax regime from April 2026, and other jurisdictions range from capital treatment to full employment income — the country-by-country map is maintained in the carried interest guide. Two design rules survive every jurisdiction: keep the three cash flows — fee for services, carry as profit share, GP commitment return as return on invested capital — separately documented, because commingling them invites the least favourable characterisation of the mix; and remember that a carry vehicle in a neutral jurisdiction does not relocate the services of a partner who works in London or New York — when carry recipients move country mid-fund, the grant, vesting and payment years each acquire their own residence analysis.

The full money and tax flow map

A finished architecture can be written down as a payment map: every material cash flow with its taxpayer, source treatment and obligation. The table shows the map for the US-bounded scenario above; a real fund maintains one of these per source country.

Read vertically, the map is the audit answer to "who pays what where"; read horizontally, each row is a control the administrator must actually be able to operate — the fund is the withholding agent for several of them, and an obligation the data system cannot support is a penalty waiting for a payment date.

Reporting, tax advances and the data layer

Three obligations attach to the architecture as a whole rather than to any single flow. First, information reporting: every entity in the chain — fund, feeders, blockers, GP, carry vehicle — needs its own FATCA and CRS classification, and the two regimes classify differently, so one copied answer is a standing error; the exchange-of-information layer is mapped in the tax transparency guide, and EU-nexus structures add DAC6 hallmark analysis at design time and again at each material amendment. Second, tax advances: when the fund pays withholding on behalf of a specific partner, LPA drafting treats it as an amount distributed to that partner — the approach reflected in the ILPA Model LPA's tax-distribution and carry provisions — so that one investor's tax never silently reduces another's return; the same logic gives carry holders tax distributions when allocations precede cash, and it only works if the LPA says so before the first such payment. Third, the data obligation: allocations, withholding by payment, investor status forms with effective dates, debt balances and tax lots must live in the administrator's system, because K-1s, Forms 8804/8805, FATCA/CRS files and treaty relief claims are all generated from the same records — an architecture whose reporting depends on data nobody captures has a compliance gap designed into it.