Geode Capital Management 13F Portfolio

The biggest buys table makes it clear: Geode is no longer content to just own the AI winners; it wants to own the picks-and-shovels that feed them.

• KLA: A near-10x increase in shares (up 893.4%) and about $9.88B more at work is the quarter’s headline statement. Geode is buying into weakness here: the position sits roughly -21.4% versus its average cost, suggesting the firm views recent underperformance as mispricing in a critical node of the semicap chain.
• Marvell Technology: Shares are up 67.9%, adding roughly $2.53B. This is a clear call that high-speed connectivity and custom silicon for AI and cloud are still early in their demand cycle.
• Alphabet (GOOG class): A 2.8% add worth about $1.10B extends the bet on hyperscale AI platforms, not just chips. Geode is treating Alphabet as essential AI infrastructure in its own right.
• Nvidia and Broadcom: Adds of roughly $975.6M to Nvidia and $796.6M to Broadcom reinforce the view that GPU and networking demand will remain structurally elevated, even after massive gains since cost.
• Coca-Cola and Netflix: A ~$1.06B increase in Coca-Cola and a ~$482.6M add to Netflix show some capital being channeled into durable brands and subscription cash flows alongside the AI buildout.

The pattern is consistent: Geode is paying up for the full AI stack — fabrication tools, accelerators, interconnect, and platforms — and sprinkling in a few cash-generative consumer names as ballast.

The trims list reads like a roster of successful, now-mature stories used as cash machines for the AI push.

• SanDisk: A 14.7% cut, shrinking the position by about $1.41B, is notable given the enormous gain versus cost (+845.3%). This looks like harvesting a legacy storage winner to fund higher-conviction bets in today’s AI memory and networking names rather than a call on storage demand disappearing.
• Alphabet (GOOGL class): A modest 1.2% trim worth ~$644.9M in the A shares, while adding to the C shares, is classic optimization — not a fundamental change of heart on Alphabet, just fine-tuning share class exposure.
• UnitedHealth: A 4.5% reduction (~$421.1M) signals less enthusiasm for managed care as a growth engine relative to AI and semis. Health care remains important, but it’s clearly no longer the incremental destination for new risk.
• Lam Research and General Electric: Small trims in Lam (down 1.9%, about $289.6M) and GE (down 2.1%, ~$202.6M) suggest Geode is rotating within industrial and semicap winners to names where upside-to-cost still looks more attractive.
• Money-center banks: JPMorgan, Citigroup, Goldman Sachs, and marginal cuts to Bank of America all point the same way. Financials are being treated as ballast: solid, profitable, but not where the next 5 years of alpha will be generated.

In short, Geode is cashing in gains across storage, banks, and defensive health care to pay for a deeper, more concentrated AI infrastructure book.

At the sector level, weights barely budged — technology up from an estimated 62.45% to 62.84%, and modest downticks in consumer discretionary, health care, finance, and industrials. But that understates how aggressively Geode is rewiring the internals of its tech exposure.

Within semiconductors and related equipment, KLA, Marvell, Nvidia, Broadcom, Micron, Intel, Texas Instruments, Applied Materials, and Lam collectively represent a concentrated view that AI-era capex is still in the early innings. Software and platforms — Microsoft, Alphabet (both classes), Meta, Palantir, Oracle, Palo Alto Networks — sit on top of that hardware stack, turning the sector sleeve into a layered AI ecosystem rather than generic “tech beta.”

Outside technology, the rotation is more incremental. Health care’s share slipped (7.54% to 7.41%) as UnitedHealth and Johnson & Johnson were trimmed modestly, while Eli Lilly, AbbVie, Merck, and Philip Morris were gently topped up — a tilt from defensive quality toward higher-growth or cash-rich pharma. Finance eased from 4.89% to 4.74% as banks were chipped away, and industrials dipped slightly with small cuts in RTX. Consumer staples quietly gained ground, mainly via the sizeable Coca-Cola add, giving the portfolio a liquid, defensive counterweight to its AI bets.

The picture: Geode keeps the sector pie looking familiar, but under the hood has reallocated risk toward the parts of tech most leveraged to AI data center buildouts and away from rate- and regulation-sensitive financials.

Put together, this quarter’s moves tell a clear story: Geode believes AI-related capex and data center buildouts will power through whatever the macro throws at them. That conviction shows up in the aggressive scaling of KLA and Marvell, steady adds to Nvidia, Broadcom, and Micron, and continued commitment to hyperscale platforms like Alphabet and Microsoft.

At the same time, the fund is not running a pure momentum book. Trimming ultra-profitable but slower-growing franchises in banks, legacy storage, and managed care, while adding to Coca-Cola and maintaining large positions in Berkshire Hathaway, Walmart, and Procter & Gamble, keeps a shock absorber layer under the AI bet.

Forward-looking, the risk Geode is underwriting is clear: if AI infrastructure demand normalizes faster than expected, this book will feel it in both semis and tools. But if AI workloads, networking requirements, and wafer intensity continue to ratchet higher, the combination of early cost bases and fresh capital in under-loved enablers like KLA and Marvell could drive outsized returns.

For observers, the message is that this is no vanilla index hug. Geode is using its scale and quasi-index footprint as camouflage for a very specific call: the AI buildout has further to run, and the best remaining risk-reward sits not in the headline GPUs alone, but in the complex, capital-hungry machinery that makes the entire stack possible.