Saudi Public Investment Fund (PIF)
Public Investment Fund, Riyadh; Governor Yasir Al-Rumayyan, chaired by Crown Prince Mohammed bin Salman. The Vision 2030 sovereign vehicle and primary backer of HUMAIN.
PIF is the most aggressive sovereign AI-infrastructure bet, but most of it is private and does not show in the 13F: its subsidiary HUMAIN (launched May 2025) is building Saudi 'AI factories' with Nvidia (up to 600,000 GB300-class GPUs over 3 years), AMD, Qualcomm, AWS and SpaceX (xAI), plus a $1.5B Groq inference commitment and a $3B stake in Elon Musk's SpaceX (xAI). Its tiny public US 13F sleeve is dominated by Lucid (EV), Electronic Arts and Uber — a legacy concentrated book, not the AI story. The real AI/semi/compute/power exposure lives off-13F in HUMAIN, direct GPU procurement, and data-center power buildout (1.9GW by 2030 target).
- Not open — Not an open-ended vehicle — analysis only, not an available allocation.
- Partial 13F — The disclosed book is longs-only and ~45 days lagged — a slice, not the strategy.
Top holdings
Largest US 13F position — Legacy strategic stake; mobility/autonomy
Large — Gaming; part of PIF's Savvy Games AI/gaming thesis
Controlling stake — ~$9.5B cumulative PIF commitments since 2018; +$550M April 2026 placement
Wholly-owned subsidiary — AI value-chain co; Nvidia/AMD/Qualcomm GPUs, SpaceX (xAI) + Groq deals — the actual AI-trade exposure, off-13F
~$3B via HUMAIN — Series E; 500MW Saudi data center JV
Recent moves
Q1 2026 13F (filed 15 May 2026): consolidated US sleeve to just 4 names (Uber, EA, Lucid, Mapleton), ~$12.0B. Off-13F: HUMAIN expanded the Nvidia partnership (up to 35,000 GB300 systems approved by Washington) and added AMD/Qualcomm/AWS/SpaceX (xAI) at the Nov 2025 US-Saudi Investment Forum; +$550M into Lucid (Apr 2026).
Loading disclosed positions…
Our take
Why it matters: PIF is the sovereign most willing to write nine- and ten-figure checks straight into the AI compute/power stack — Nvidia GPUs, SpaceX (xAI), Groq, AMD — making it a bellwether for Gulf AI capex. Edge/limits: almost none of that shows in the 13F, which is a stale, concentrated legacy book (Lucid/EA/Uber); reading PIF off its 13F badly understates the AI exposure. Honest caveat: HUMAIN/SpaceX (xAI)/Groq figures are announced commitments and GPU-allocation approvals, not audited deployed capital, and the AI thesis carries concentration + geopolitical export-control risk. China-adjacent framing aside, these are Saudi/US-aligned deals — but the dollar amounts should be read as headline commitments, not booked positions.
The 13F is a decoy. The Q1 2026 US sleeve is ~$12.0B in three material names — Uber ~44% ($5.23B, 72.8M shares), EA ~42% ($5.05B, 24.8M shares), Lucid ~14% — plus a de minimis Claritev stub; the quarter's only activity was exiting Allurion.
A legacy book, not the AI thesis, and it will shrink ~40% mechanically: the $55B EA take-private (PIF/Silver Lake/Affinity, ~99% shareholder approval Dec 2025) is still pending CFIUS clearance as of Jul 2026, its outside date extended from Jun 30 to Sep 28. Whenever it closes, PIF's second-largest listed position converts to an unreported private stake — which anyone reading the next 13F as 'PIF dumped EA' will mistake for a decision. Second: separate announced commitments from deployed capital. The verifiable near-term anchor under the 600,000-GPU headline is an 18,000-unit GB300 first cluster for the xAI deployment, with 'several hundred thousand more' framed as forthcoming — and gated by US export approvals that are policy, not contract.
- 2026-07Update 30 Jul 2026: EA confirmed all regulatory approvals obtained (EU cleared 23 Jul); the take-private is set to close on or about 4 Aug 2026, PIF at 93.4%. The CFIUS hurdle and 28 Sep outside date fall away.
Thesis
PIF is not a fund in the return-generating sense — it is a state industrial-policy balance sheet that reports an AUM number. Its headline growth came overwhelmingly from the government transferring Aramco equity onto its books (~$140B of the ~$150B 2024 increase to an audited $913B, per AGSI's reading of the annual results), not from investment skill. The AI push (HUMAIN, Nvidia, xAI, Groq) is real and the most aggressive sovereign compute bet anywhere, but it is capex in service of diversifying an oil economy — judged by GDP and jobs at home, not by IRR.
Dual mandate: build the domestic economy (gigaprojects, HUMAIN, Lucid, tourism) while holding an international sleeve for return. Vision 2030 pushed the mix heavily domestic. Post-2025 the AI stack is the flagship: HUMAIN as the vehicle for 'AI factories' — Nvidia GB300 clusters plus announced AMD, Qualcomm, AWS and xAI partnerships, against a stated ambition of up to 600,000 GPUs over three years. Funded from Aramco dividends, transferred equity, and a growing debt program.
- 2026-06Audited FY2025 statements filed to the LSE around 30 Jun 2026 put assets at SAR 4.54tn ($1.21tn) at end-2025, up 5% from SAR 4.32tn. The $913B measures end-2024, a full reporting year behind.
Assessment
- Truly patient capital: no LP redemptions, no gate risk, no quarterly marks — it can hold a 10-year infrastructure bet through a drawdown that would force a levered fund out.
- Counterparty gravity. Nvidia, AMD, Qualcomm, AWS and xAI take the call, so PIF gets allocation in a GPU market where access, not price, is the binding constraint.
- Vertical logic is coherent: cheap domestic power plus land plus capital is a genuine comparative advantage for compute, unlike most sovereign tech-diversification attempts.
- Disclosure has improved — audited annual reports and a capital-markets program force accountability a pure sovereign vehicle could avoid.
- AUM growth is largely asset transfer, not performance: of the ~$150B 2024 increase to an audited $913B, ~$140B was transferred Aramco equity. The headline measures balance-sheet reshuffling, not returns earned.
- Reported return decayed from 8.7% annualized (through 2023) to 7.2% (Sept 2017–end 2024) — arithmetically implying roughly flat 2024, with gigaproject values down ~12%.
- The 13F is unusable as a signal: three material names, ~86% in two, ~45-day lag, US-listed longs only, and it excludes essentially the entire AI thesis it is most often cited for.
- Governance concentration. The chair is the Crown Prince and the governor sits on the boards of the assets; capital allocation and statecraft are the same decision, with no independent check on a bad project.
- Funding is oil-levered. Aramco dividend policy and the Saudi fiscal deficit set the deployment budget — the AI capex schedule is a function of crude, not conviction.
Record
There is no fund-style NAV or total-return series to evaluate, and that absence is the finding. The one reported number — 7.2% annualized total shareholder return Sept 2017 to end-2024, down from 8.7% through 2023, per AGSI's reading rather than a confirmed PIF primary report — implies a roughly flat 2024, below what passive global equity delivered over the same span. Attribution matters more than the level: ~$140B of the ~$150B 2024 AUM increase was transferred Aramco equity, so almost none of the growth is investment return. Gigaproject and NEOM marks are internally determined and fell ~12% in 2024, so the largest, least liquid part of the book has the weakest valuation evidence. Lucid carries a large unrealized loss — PIF has not sold, holding ~58% via Ayar and adding into the ~$1.05B raise. No comparable 2025 figure exists; HUMAIN-era performance is unmeasurable, its assets having no market price.
- 2026-07FY2025 results published Jul 2026: net income SAR 65.1bn (+152%), operating profit SAR 77.9bn (+125%), revenue SAR 449.9bn (+9%), return restated as above 7% since 2017. A 2025 comparable now exists.
- 2026-07PIF's Lucid holding is reported above 54% in late Jul 2026, diluted by the raise and Uber's 11.5% (~37.8M shares); Prince Alwaleed separately disclosed 5.00% on 23 Jul. LCID +~33% on the month, -77% over a year.
Risks & fit
- Deal-close risk is live: the EA take-private still needs CFIUS clearance, and a foreign-sovereign buyer of a US publisher is precisely the profile that review exists to scrutinize.
- Export-control reversal: GB300 allocations rest on US policy approvals that can be narrowed or withdrawn without breach of any contract.
- Oil price. A sustained crude decline compresses Aramco dividends and the fiscal budget simultaneously, forcing capex deferral at the wrong point in a build cycle.
- Depreciation mismatch: GPUs on a 3–5 year useful life funding 20-year infrastructure ambitions; a compute-price collapse strands the asset before payback.
- Key-person and succession risk is absolute — mandate, governance and deal flow all route through two individuals with no institutional redundancy.
Two observations would force a revision. First, if PIF's next annual report shows returns rising back toward or above 8.7% annualized with the gain attributable to investment performance rather than further Aramco or state-asset transfers, the 'balance sheet, not fund' read weakens materially. Second, if HUMAIN reaches audited third-party compute revenue at scale from non-Saudi customers, with installed capacity and utilization disclosed against its stated GW path, the buildout is a commercial business rather than industrial policy and the announced-versus-deployed gap closes on its own.
A useful read for anyone tracking Gulf AI capex, Nvidia's sovereign demand, or the durability of compute buildouts financed by commodity revenue — PIF is the best proxy for whether announced sovereign AI commitments convert into installed capacity. It is not accessible: a Saudi state vehicle with no outside investors and no feeder, so this is analysis of a bellwether's behavior, not of an opportunity. The application is second-order: sizing the real order book behind vendor guidance.
No fee structure exists — PIF is a state-owned sovereign vehicle with a single shareholder (the Saudi government), not a fee-earning manager. The relevant cost is its funding cost: a growing bond and sukuk program layered on oil-linked equity funding.