Berkshire Hathaway (value/ballast counterpoint)
Greg Abel (CEO, succeeding Warren Buffett); investment decisions increasingly Abel-led as of 2026.
Included as the deliberate value/ballast counterpoint — Berkshire is NOT AI-thesis-led. Its relevance to this page is exactly the opposite stance: a ~$400B cash hoard held as dry powder against AI-era valuations, with the AI exposure being indirect (Apple, still ~22% of the book) plus a notable new Alphabet position that is its first real lean toward the AI infrastructure trade. It is the page's reminder that the smartest capital allocator is sitting largely on the sidelines.
- Not open — Not an open-ended vehicle — analysis only, not an available allocation.
Top holdings
Top holding but cut ~74% over two years; AI exposure is incidental (on-device AI, not a frontier bet).
+225% in Q1 2026 — New scaled position — ~$10B private placement (Abel) plus ~$10-11B open-market buying. Berkshire's clearest tilt toward
~$400B — Record cash pile — the ballast. Dry powder, implicitly a statement that AI-era equity valuations are unattractive.
added — Insurance — core Berkshire, no AI angle.
Recent moves
Q1 2026: +225% Alphabet (Abel's ~$10B private placement + ~$10-11B open-market) — its first meaningful AI-infra position; continued Apple trim; full exits from Amazon, Domino's, UnitedHealth; large Chevron cut; added Chubb, NYT. Cash reached a record ~$400B. Sources: Seeking Alpha, CNBC, TheStreet.
Loading disclosed positions…
Our take
The counterpoint the page needs: while the other five corporates pour tens of billions into private AI labs, Berkshire holds ~$400B in cash and only just tiptoed into the trade via Alphabet — and even that is a value/cash-flow case (cheap relative to peers, owns TPUs) more than an AI moonshot. Edge: optionality and discipline if AI valuations compress. Caveat for readers: a 13F captures only long US equity, so it understates Berkshire's wholly-owned operating businesses (BNSF, energy, insurance) — but on the AI trade specifically, the signal is genuine restraint, not hidden exposure.
Tripling Alphabet in Q1 (~17.9M to ~57.8M shares, ~$16.6B) is a conventional value trade: cheapest megacap on earnings at initiation, owns its TPU silicon, throws off cash.
The June $10B follow-on is the more informative move — but not for the reason usually given. It was not a 2008-style Goldman/GE rescue preferred. The disclosed terms are plain common stock: $5B Class A at $351.81 and $5B Class C at $348.20, one tranche inside Alphabet's broader $84.75B raise. No warrants, no preferred, no rescue economics. What it demonstrates is anchor participation at size and speed — a balance sheet that can absorb $10B of a primary issue in days, which few allocators can offer and which an issuer will take a call about. That is a distribution edge, not a terms edge, and materially weaker than the counterparty-of-last-resort analogy implies. Read that way, the $397.4B is not a verdict on AI valuations; it is inventory the model requires. Berkshire has held elevated cash through rallies it later conceded it should have joined.
- 2026-07July 2026 gave the cash posture its best evidence yet: BRK.B +2.24% while SOX fell 20.6%, SMH 17.6% and the S&P 500 0.1%. A cash-heavy, AI-hardware-free book won the first month the AI trade broke.
- 2026-07The Alphabet trade has not yet proved itself: GOOGL closed 31 Jul at $356.13, about 1.2% above the $351.81 tranche price and ~11% below its 13 May high of $402.38, after -8.1% in June and -0.4% in July.
Thesis
Berkshire is the anti-thesis on this page: while corporate peers push tens of billions into private AI labs, it ended Q1 2026 with a record $397.4B in cash and T-bills and a 13F equity book of ~$263B. But the popular reading — that the smartest allocator is bearish on AI — is the weakest one. The cash is at least as much a size problem as a market call, and Greg Abel's first year reads better through what he actually bought, and how, than through what he didn't.
A holding company whose engine is insurance float funding wholly-owned operating businesses (BNSF, Berkshire Hathaway Energy, GEICO), with marketable equities as the residual. Under Greg Abel (CEO since 1 Jan 2026; Buffett remains chairman), Q1 2026 cut the equity book from 40 names to 26 — 16 full exits including Amazon, UnitedHealth, Visa and Mastercard — concentrating into fewer, larger positions while net-selling equities (~$24.1B sold vs ~$16B bought).
- 2026-07The anti-AI framing understates the book: Berkshire holds ~$31.2B of Alphabet — 68.5M Class A plus 17.9M Class C, over 9% of a ~$342B equity book — and $10B of it was bought into a primary AI-capex financing.
Assessment
- Structural rather than analytical edge: a balance sheet that can absorb $10B of a primary equity issue in days. Distribution capacity does not decay as information diffuses.
- Cash is not idle — $339.3B sat in short-term Treasuries at Q1 2026 per the 10-Q, so waiting earns a real carry rather than nothing.
- No management or performance fee, no lockup, no gate — a structural multi-point annual head start versus the 2-and-20 vehicles on this page.
- Buyback discipline is observable, not asserted: repurchases resumed March 2026 at ~1.4x book after a 21-month pause — a stated price test, not a promise.
- Scale has broken the compounding math. Only elephant-sized deals move the needle, so 'discipline' and 'no adequate opportunity set' look identical from outside — and management prefers the former label.
- The resumed buyback was ~$234M against a $397.4B cash pile. If the stock was cheap enough to buy, that size is a rounding error; it reads as ambivalence, not conviction.
- Abel's first 13F cut the book from 40 names to 26, several long-held Buffett-era. Either the old book was over-diversified (an admission) or a new CEO is establishing authorship. Both trouble anyone extrapolating Buffett forward.
- Succession is not clean. Buffett remains chairman and the buyback program requires the CEO to consult him. Attribution for 2026 decisions is genuinely ambiguous — exactly when you would want it not to be.
- The 'negotiated capital on special terms' framing applied to the Alphabet placement does not survive the actual terms: common stock, no preferred, no warrants. The edge there is size and speed, not economics.
- 2026-07The June Alphabet tranche did carry a price concession: $351.81 against a ~$359 market print on 3 Jun, described as a 6.5% discount in one account. Magnitude is disputed, direction is not — terms were not flat.
Record
Berkshire publishes no fund-style return — book-value growth and BRK.B price are the only proxies, and both blend the operating businesses with the equity book, so neither isolates allocation skill. The Q1 2026 13F shows the equity book falling from ~$274B to ~$263B, with ~$24.1B sold against ~$16B bought (net ~$8.1B of selling) and the name count cut 40 to 26. Apple was ~22% of that $263B Q1 base (~228M shares); measured against the ~$351B portfolio press trackers cited on 14 July 2026, the same stake is ~20.6% — the weight quoted depends entirely on which base is used. Worth flagging: Abel did not continue the Apple trim in Q1. The share count held steady, ending nearly two years of selling; any 2026 account saying otherwise repeats a stale narrative. Attribution across leverage, float cost, operating earnings and security selection is not separable from public disclosure.
Risks & fit
- Key-person risk is relabelled, not retired. Buffett as chairman still gates buyback decisions; his full departure is an untested transition.
- Insurance is the real leverage. A severe catastrophe year or reserve deficiency at GEICO or the reinsurance book hits harder than any equity mark.
- Concentration: Apple was ~22% of the Q1 13F equity book. Multiple compression in one megacap swamps any stock-picking contribution.
- Rate risk on the $339.3B T-bill stack — a sharp fall in short rates removes a large interest-income line and turns patience from cheap to expensive.
- BNSF and Berkshire Hathaway Energy carry rate-regulated and wildfire-liability exposure uncorrelated with the AI question framing this page.
The 'disciplined dry powder' reading fails if cash stays above ~$300B through a genuine AI-valuation drawdown with no large deployment — waiting is only skill if the wait ends in a purchase. Our own read — that the edge is anchor capacity in primary issuance rather than stock-picking — fails if the Alphabet placement proves a one-off and the next two years bring only open-market 13F buys at prevailing prices, which would be replicable by any index-tracking vehicle at lower cost. Watch deployment structure, not the cash number.
Understanding what a permanent-capital, float-funded holdco does differently from a fund, and as a calibration point against the AI-forward corporates elsewhere on this page. Read it as an argument about the cost and value of optionality at scale. Note that the 13F lens materially misreads Berkshire: it captures only long US-listed equity and omits the wholly-owned operating businesses and insurance float that generate most of the economics.
No management or performance fee — a structural advantage versus the 2-and-20 vehicles on this page. Offset by corporate-level taxation at the holdco, so the after-tax comparison is narrower than the headline zero-fee framing suggests.