Berkshire’s Foreign Bond Shift Is a Warning to Stablecoin Treasuries

In the second quarter of 2026, Berkshire Hathaway reported net profit of $25.667 billion, up from $12.37 billion a year earlier. Revenue declined. Cash fell from $397 billion to $364.7 billion. And inside the fixed-income basket, a small but unusual detail: foreign bonds accounted for $12.668 billion, or 74.4%, of a total $17.034 billion fixed-income portfolio, while U.S. Treasuries made up only $3.002 billion, or 17.6%. In a world of ledgers, who holds the memory? This is not a crypto earnings report. It is the largest traditional capital allocator on Earth telling us something about the reference assets we have been treating as risk-free. For years, I have watched stablecoin issuers and tokenized treasury protocols build their foundations on one assumption: the U.S. Treasury is the ultimate settlement layer. Circle’s USDC reserves are dominated by Treasuries. Ondo’s OUSG, BlackRock’s BUIDL, even ether-denominated yield products all benchmark their returns to short-term dollar rates. This is not an accident. The entire architecture of crypto-native cash management is a derivative of Berkshire’s old playbook: hold cash-like instruments, earn a modest yield, remain liquid enough to move when opportunity arrives. The difference is that Berkshire is now editing the playbook, and the edits point away from the dollar-denominated core. Let us parse the numbers with the discipline of a smart-contract audit. A cash decline of $32.3 billion is an 8.1% drawdown in a buffer that had become a symbol of extreme defensiveness. It is not capitulation. It is a softening of posture. Berkshire’s “liquidity total” — cash plus fixed income — still sits at roughly $381.7 billion, so the pool is still enormous. But the marginal dollar, after years of waiting on the sidelines, is leaving the most sterile asset class. The ratio of fixed income to cash remains tiny: $17.034 billion against $364.7 billion. This is not a directional pivot. It is a change at the margin, and in markets, the margin is where the future is encoded. The net profit figure deserves the same scrutiny. Revenue declined, yet net income doubled. That gap is not an operating miracle; it is a mark-to-market tailwind. Berkshire’s equity portfolio is long duration, long equity beta, and, by extension, long the expectation that rates will not climb much further. Every stablecoin treasury manager makes the same implicit bet when they ladder tokenized Treasury products with seven-day or one-day maturities. The duration is shorter, but the direction is identical. We are all designing instruments that become more valuable when the central bank blinks. In that sense, Berkshire is just a slower, brick-and-mortar version of a DeFi vault manager. The true information gain, however, is the fixed-income split. Historically, Berkshire treated its fixed-income sleeve as a short-duration U.S. Treasury reservoir. Today, 74.4% of that sleeve is allocated to foreign bonds. The absolute amount is small, but the structure is abnormal. Foreign sovereign bonds carry currency risk, jurisdictional risk, and liquidity risk. For an organization that sells prudence, this is not a simple coupon hunt. It is an admission that the riskless asset is no longer singular. If the largest institutional allocator in the world is quietly diversifying away from U.S. Treasuries in its marginal fixed-income decisions, then the entire stablecoin design principle — one reserve asset, one jurisdiction, one oracle feed — starts to look fragile. Tokenized treasury products obsess over on-chain redemption speeds and daily attestations, but they generally assume the U.S. Treasury is the only reference asset worth tokenizing. Berkshire’s split suggests that assumption is beginning to soften. If the next wave of real-world-asset protocols is built on a basket of foreign sovereign bonds, the oracle infrastructure gets dramatically more complex. FX data feeds, country-level default probabilities, and cross-border settlement rails suddenly matter as much as block height. In my audits, I look for hidden trust assumptions hidden in the line items. The line item that once said “U.S. Treasury” now says “foreign bond,” and every consumer of that data needs to understand the new risk surface. I spent weeks in a windowless room in 2017 auditing an Ethereum DAO framework and found three reentrancy vulnerabilities that would have drained millions if exploited. The lesson stayed with me: every line item in a balance sheet is a governance commitment. When Berkshire moves $12.7 billion into foreign bonds, it is not just rebalancing coupons. It is encoding a governance preference about which sovereign’s promise can be trusted. On-chain, that preference would be visible to anyone willing to inspect the vault. We code the trust, but we must audit the soul. Here is the counter-intuitive part: this is not a signal that Berkshire is preparing for a crypto supercycle. It is not about bitcoin exposure. It is more subtle and, to be honest, more concerning for crypto’s existing narrative. The $12.7 billion foreign bond allocation is tiny relative to the $364.7 billion cash pile. If Berkshire truly believed the dollar’s dominance was ending, we would see a much larger shift. Instead, we see a marginal diversification of less than 10% of total liquidity. That is a hedge, not a thesis. The alert is not “Buffett is fleeing the dollar.” It is that the marginal institution is no longer comfortable with a single reference asset, even when the core position remains unchanged. And for stablecoins, a single-reference-asset model is exactly the design they are built on. Circle’s compliance-first approach can freeze any address within 24 hours, but it cannot freeze the geopolitical risk embedded in a reserve mix. The same logic applies to every tokenized treasury product with a monotonous allocation to U.S. government debt. The deeper issue is oracle feed latency. DeFi has spent years treating Chainlink-style data delivery as the connective tissue between on-chain contracts and off-chain reality. But the oracle problem multiplies when the reference asset stops being a single Treasury yield and becomes a basket of foreign bonds with varying settlement conventions, tax treatments, and political overhangs. A stablecoin backed by a mix of U.S. Treasuries, Japanese government bonds, and euro-area debt is not more stable; it is more dependent on the quality of the data feed. The protocol is neutral, but the user is human. And human users tend to assume that a stablecoin’s reserve attestation covers the same risks they understood yesterday. The moment the reserve basket changes, the meaning of “stable” changes with it. Proof is binary; meaning is fluid. The on-chain proof may be the reserve balance, but the meaning is the stability of the underlying promise. If the world’s largest allocator is quietly widening the definition of risk-free, then the next generation of stablecoin contracts will need to widen the definition of collateral. The industry cannot simply add more tokens to the basket and call it diversification. It needs governance mechanisms that let users see not just the reserve ratio, but the identity, duration, and jurisdiction of every underlying instrument. It needs oracles that can absorb a foreign bond default without triggering a cascade of redemptions. And it needs a conversation about what “trust” means when the anchor asset itself is no longer static. We are not moving money; we are moving belief. And belief, unlike a reserve ratio, is not audited once a quarter. Berkshire’s cash pile still exists, still massive, still patient. But the margin has moved. The fixed-income corner has been rearranged. The question for every stablecoin issuer and every tokenized treasury protocol is simple: if the world’s most cautious investor is no longer sure that the old reference asset is the only safe port, why is the on-chain economy still building all of its harbors in the same bay? In a world of ledgers, who will hold the memory of what “risk-free” used to mean before we all agreed to redefine it at the same time?