Light Street Capital
Glen Kacher (founder, CIO)
'Tiger grandcub' tech specialist running a highly concentrated, overweight-semis-and-infrastructure-software book aimed squarely at AI-fueled compute demand.
- Not open — Not an open-ended vehicle — analysis only, not an available allocation.
- Self-reported — Headline returns are manager- or firm-reported and unaudited.
Top holdings
Top position; AI foundry
Core AI compute
Custom AI silicon / networking
AI GPU challenger
Fintech; non-AI diversifier
Recent moves
Stayed overweight semis and infrastructure software; top four positions (TSM, NVDA, AVGO, AMD) are ~40% of the book — a near-pure semiconductor-AI expression — with Kacher publicly reiterating the overweight in early 2026.
Loading disclosed positions…
Our take
Light Street is the highest-octane direct semiconductor play here — a small, concentrated fund willing to put ~40% in four AI chip names; it's a leveraged read on the compute cycle, which is exactly what drove its 2025 outperformance and what makes it the most cyclically exposed.
The number no marketing deck leads with: chaining the reported annual returns 2021-2025 (-26, -54, +46, +59.4, +37) gives roughly +48.7% cumulative since Jan-2021 — about 7.5% annualised once the reported +37.05% net for H1 2026 is chained on (the +8.5%/1.65% figure was the 2021-2025 window), on the net-of-fee basis hedge-fund returns are conventionally published, and before LP entry-timing effects. The comeback is real and large, but it mostly refills a hole the same strategy dug. That is the signature of a high-beta factor bet more than of demonstrated alpha: outsized moves arriving on the same axis in both directions. A second detail rewards reading the filing closely. Alongside the 14.4% TSM long, the Q1-2026 13F discloses a TSM put position at ~6.9% of reported value. The filing carries no strike, no expiry and no stated intent, and 13F option-reporting conventions vary by filer — so whether that put offsets the long, and by how much, is not recoverable from the document. Reading the top-four weights as ~40% net semi exposure asserts more than the filing supports.
Thesis
A ~$1.1B Tiger-grandcub tech long/short (Glen Kacher, founded 2010) that turned one call — AI compute demand accrues to the semiconductor supply chain — into three straight blowout years (+46% 2023, +59.4% 2024, +37% 2025). The Q1-2026 disclosed book is a near-pure expression of it: TSM 14.4%, NVDA 8.9%, AVGO 8.7%, AMD 8.4%. Our critique is not whether the call worked. It is that the identical concentration produced -26% in 2021 and -54% in 2022, and nothing visible has changed the risk shape.
Concentrated tech long/short out of Palo Alto, Tiger Global lineage. Fundamental single-name selection through very few positions — top 10 = 74.4% of the Q1-2026 13F book — skewed to semis and infrastructure software. A parallel venture franchise (a third VC fund raised into the tech rebound) puts public and private tech exposure on the same underlying thesis. Short book, gross and net leverage are not publicly disclosed.
Assessment
- Thesis is specific and falsifiable — foundry, GPU, custom-ASIC and CPU exposure, not a vague 'AI' basket.
- Willingness to hold size. Top 10 at 74.4% means the good calls actually move the fund rather than being diluted.
- Stayed with the thesis through the 2021-22 drawdown rather than de-risking at the bottom and missing the 2023-25 recovery.
- Sizing tracks stated conviction rather than benchmark weights — the top four disclosed longs sum to 40.3% of the reported book.
- Three good years follow two catastrophic ones. Five-year chained return is roughly +8.5% — the record reads far better in a three-year window than in a five-year one.
- Returns and the AI-semi factor move together, which makes skill and beta hard to separate. No public attribution splits stock selection from factor exposure.
- 13F book fell from ~$650M to ~$534M in one quarter. Marks, redemptions and hedging are indistinguishable from outside.
- Single-decision-maker firm. Kacher is the thesis; no publicly disclosed succession or co-CIO structure.
- The venture funds and the public book express the same tech-cycle bet, so the two sleeves likely correlate exactly when diversification would matter most.
Record
Reported annual sequence, net as conventionally published: -26% (2021), -54% (2022), +46% (2023), +59.4% (2024), +37% (2025), with 2026 YTD not public. Losses and redemptions removed roughly 70% of firm assets across 2021-22. The 2023-25 run is the top of the stock-picking-hedge-fund league table and clearly beat Tiger peers (Maverick ~29%, Whale Rock ~27%, Lone Pine ~23% in 2025). But the drawdown means a 2021-vintage LP has compounded at roughly the pace of cash, while a 2023-vintage LP chained roughly 3.2x. Same manager, same strategy, radically different outcomes purely on entry date — attribution here belongs as much to timing as to selection.
Risks & fit
- A hyperscaler capex digestion pause hits all four top holdings at once — they are not independent bets, they are one bet expressed four ways.
- Small AUM plus large single positions means gates or forced deleveraging under redemption pressure.
- Key-person concentration in Kacher with no disclosed succession plan.
- Illiquid venture marks are stale and correlated to the same cycle as the public book.
- The disclosed 13F is longs-only, ~45 days lagged and excludes the short book — the real portfolio is materially unknowable.
The durability case rests on the edge being stock selection within the AI complex, not levered exposure to it. Two observations break that. First, a drawdown in the next semi correction materially worse than an equal-weight basket of its own top holdings would show the value-add is sizing and leverage, not picking. Second, if a full cycle passes and the chained multi-year return still trails a passive semiconductor index net of fees, the concentration is return-sequencing rather than alpha. Conversely, a positive year while semis fall would be real evidence the short book does work.
This analysis serves readers studying how concentrated factor bets look inside a hedge-fund wrapper — the gap between a headline three-year record and a five-year chained one, and why 13F top-weights cannot be read as net exposure when the same filing carries puts of undisclosed terms. Note the fund is a private vehicle closed to general subscription; nothing here is a view on participating in it, and the disclosed book is a lagged, longs-only fragment of the real portfolio.
Fee terms are not publicly disclosed. Tiger-lineage funds typically run 1.5-2% plus 20%. High-water marks are per-investor: 2021-vintage capital sat under water through the recovery, while 2023-vintage LPs paid full incentive from a fresh mark. Unverified.