Bond crisis looming? GOP abandons DOGE, Google disrupts Search with AI, OpenAI buys Jony Ive's IO

Friedberg notes the CBO’s long-run models assume a 3.6% interest rate, but 30-year Treasuries are already around 5.1%. Every 1% above 3.6% adds roughly $350B/year in interest—so the current spread implies ~$5T in extra interest over a decade. Because higher rates mean more interest, which forces more borrowing at higher rates, the relationship between deficits and debt becomes recursive and non-linear. Markets are starting to demand higher yields, signaling growing discomfort with U.S. fiscal trajectory.

The House bill makes Trump’s 2017 tax cuts permanent, averting what Sacks calls “the largest tax increase in decades” for both the wealthy and middle class, while adding an estimated $3–5T to the national debt over 10 years. It extends standard deduction and child tax credits, cuts taxes on tips and overtime, and funds border and energy priorities, but also includes eleventh-hour changes the hosts say reflect poor financial literacy—especially on energy tax rules, Medicaid cuts, and assumptions about future interest rates. Chamath and Friedberg see it as anti-DOGE and out of step with the austerity voters thought they were electing.

Chamath and Friedberg argue that all the promise of AI, automation, and biotech hinges on cheap, abundant energy. China is adding roughly a “new United States” of power generation every ~18 months, moving from ~3 to ~8 terawatts, while the U.S. hovers near 1 terawatt going to 2 over 15 years. At the same time, last-minute tax changes in the bill weaken U.S. tax-equity incentives for near-term solar, storage, and other private power projects. They warn this will constrain electrons for both households and AI data centers, raising prices, creating allocation conflicts (homes vs. AI), and undercutting America’s ability to “grow out of” its debt.

Google’s I/O announcements introduce an AI Mode in Search that offers synthesized answers alongside the legacy 10-blue-links interface and ads—essentially a Perplexity-like experience integrated into Google. Friedberg highlights a new $250/month ‘AI Ultra’ bundle (YouTube Premium, 30 TB storage, and access to high-end Gemini + Flow video tools) as evidence of a deliberate pivot toward high-ARPU subscriptions, not just ads. The revival of Google Labs as a launchpad for experimental AI products gives Google a structured way to iterate on new interfaces and business models before rolling them out broadly.

OpenAI is acquiring Jony Ive’s hardware startup IO for $6.5B in stock (Altman/ OpenAI already owned ~23%), with the goal of creating a new AI-native device. The hosts mock the romanticized promo video but concede that if Ive can deliver an ‘iPhone moment’ for AI hardware—a phone alternative, pendant, or glasses that tightly integrates conversational AI—the 2% equity price tag could be cheap. Sacks and JCal question whether Ive’s legend is media-inflated and note much of Apple’s industrial design lineage traces back to Braun’s Dieter Rams, but agree the payoff is binary: either a dud or a device that could double OpenAI’s value.

Sacks describes a new ‘AI acceleration partnership’ framework with Gulf states: for every dollar they invest in local data centers, they must invest a dollar in AI infrastructure in the U.S., and at least 80% of chips in regional centers must be owned/operated by U.S. hyperscalers. This replaces Biden-era export controls that required case-by-case licenses for GPUs in the region and were pushing Gulf states toward a Huawei+DeepSeek Chinese stack. The hosts argue this approach makes NVIDIA/AMD the de facto global standard, improves the U.S. trade balance, and ensures energy-rich allies fund American AI rather than Chinese AI—directly tying foreign policy to platform and ecosystem strategy.

In Science Corner, Friedberg explains a case where doctors used custom CRISPR base editing to change a single A→G mutation in a baby’s CPS1 gene, allowing liver cells to produce a crucial urea-cycle enzyme and preventing lethal ammonia buildup. The team rapidly designed and tested base editors and guides in vitro, in mice, and in monkeys, then delivered an LNP-coated editor intravenously to target liver cells in vivo. With tens of thousands of diseases traceable to similar point mutations—and many undiagnosed due to lack of affordable sequencing—this illustrates how gene editing, paired with AI and automation, could drive massive gains in healthspan and productivity, provided the U.S. solves its energy and regulatory bottlenecks.