GUY: Good morning, Ava. It is Friday, August twenty-first, twenty twenty-six, and this is Morning Signal. The big idea today is that artificial intelligence demand is still broadening, but the constraint is shifting. It is less about whether people want more compute and more about whether the system can supply the duration, power, and capital required to finance it.

AVA: Good morning, Guy. And the useful tension is that the evidence is not simply bullish or bearish. The written brief found strong earnings breadth, increasingly differentiated AI winners, a bigger sovereign infrastructure build, and consumer products that still fail basic reliability tests. So today we are separating demand from funding, infrastructure from applications, and good investment process from lucky price action.

GUY: Let us start with Goldman Sachs Exchanges. Shawn Tuteja said roughly ninety-two percent of S and P five hundred companies had reported second-quarter results, and sixty-four percent beat estimates by more than one standard deviation. Consensus earnings for twenty twenty-seven rose about two percent to roughly three hundred ninety to four hundred dollars. Nine of eleven sectors produced double-digit year-over-year earnings growth, and the median stock grew earnings fourteen percent.

AVA: Right, and Goldman Sachs Exchanges also supplied the breadth evidence. The equal-weight S and P outperformed the capitalization-weighted index by almost three hundred basis points year to date. That matters because it suggests the move is not just multiple expansion in a few mega-cap technology companies. Earnings revisions are doing real work. The favorable version of broadening is profits and revisions spreading across sectors, not investors merely fleeing one crowded theme.

GUY: Goldman Sachs Exchanges also showed why the calm index can hide a noisy market. Tuteja said the VIX fell from about twenty-one at the end of July to fifteen, while volatility inside artificial intelligence groups was much larger. Technology companies that beat still underperformed the S and P by about one hundred thirty basis points the next day. Memory, optical networking, power, liquid cooling, and semiconductors stopped rebounding as one undifferentiated basket.

AVA: And on Goldman Sachs Exchanges, the positioning signal was unusually concrete. Goldman's prime book saw its strongest net buying over a three-week period since COVID, but the buying spread across sectors after investors reduced concentrated gross exposure. Low implied correlation made index options comparatively inexpensive. That creates a clean distinction: hedge broad macro risk with the index while keeping selected single-name upside, instead of treating every AI-linked company as the same trade.

GUY: Now to Excess Returns, where Andy Constan put hard numbers around the funding problem. He estimated AI-related capital expenditure at roughly six hundred to seven hundred billion dollars this year and around one trillion dollars next year. The sequencing is the issue. Hyperscalers and frontier-model companies have to finance the build now, before investors can observe the final return. That is the hamburger today, payment Tuesday problem.

AVA: Excess Returns also identified a potentially enormous flow reversal. Constan said a long-running equity tailwind of about one trillion dollars in annual net share retirement could approach zero or flip to net issuance next year. Until recently, hyperscalers could use accumulated cash and operating cash flow. As spending exceeds those internal sources, debt and equity issuance becomes part of the operating model. The market must absorb new securities while the Treasury is issuing heavily too.

GUY: Exactly. Excess Returns used Oracle as an early warning. Enthusiasm for a major data-center commitment reversed after investors saw the related debt and equity financing. That does not prove the project's economics are bad. It shows the stock price and the financing plan cannot be separated. The leading indicators are the concession required for each deal, the performance of issued securities after pricing, the effect on buybacks, and whether leverage stays within stated tolerances.

AVA: Excess Returns also described a proposed multi-data-center structure supported by Nvidia credit enhancement and institutional capital. Constan's interpretation was not that the structure itself implies misconduct. It signals that direct customer support is reaching capacity and the ecosystem needs new pools of risk capital. The adverse inflection is not creative financing. It is a broad investor refusal to fund the next deal before committed projects are finished.

GUY: The Indicator from Planet Money connected that corporate supply to the sovereign curve. Its episode cited a five-point-three percent thirty-year Treasury yield, the highest in nearly two decades, and federal debt above forty trillion dollars. Fidelity's Jurrien Timmer, quoted in the episode, argued that hyperscaler bond supply competes with Treasury issuance. Amazon thirty-one-year bonds yielded roughly one percentage point more than comparable Treasuries.

AVA: The Indicator also noted that the Treasury's newly announced buyback initially drew a muted market response. That makes the causal chain circular. AI spending raises expected productivity and equity earnings. Financing the spending adds duration supply and can push up real yields. Higher real yields then raise the hurdle rate for the same projects. The system works while deals clear and revisions rise. It gets fragile when financing concessions become large enough to hurt existing holders or make projects fail their return thresholds.

GUY: Excess Returns offered a less apocalyptic interpretation of the long end. Constan argued that strong real growth and productivity expectations naturally raise real yields because investors prefer productive projects to fixed-rate bonds. He does not see an imminent sovereign debt crisis. He wants to know whether private projects earn more than their cost of capital, and whether the funding window stays open long enough for the build to reach cash generation.

AVA: Excess Returns also challenged the usual short-rate framework. Constan said much private-sector debt was locked in at low coupons, reducing sensitivity to short rates. Demand was supported by income, accumulated wealth, rising asset prices, and cheap market finance. In that view, inflation returns to target only if the wealth effect weakens or supply expands enough. He sees no present AI-driven supply disinflation large enough to do that job, making long rates and balance-sheet policy more important.

GUY: Monetary Matters supplied Luke Gromen's more extreme fiscal-dominance case through authoritative show notes, not a full transcript. Gromen claimed entitlements, interest, and veterans benefits equal roughly one hundred five percent of federal receipts, with those obligations growing about seven-point-five percent annually against receipts growing about four percent. He viewed the larger Treasury buyback as evidence that fiscal arithmetic is already constraining policy.

AVA: Monetary Matters also recorded Gromen's hypothetical legal revaluation of U.S. gold from forty-two dollars to twenty thousand dollars an ounce, which he said could place roughly five trillion dollars in the Treasury General Account. He preferred gold to long bonds and cited a sixty-five-trillion-dollar gross, twenty-two-trillion-dollar net offshore dollar position plus a thirteen-to-fourteen-trillion-dollar carry trade. Those are guest claims from primary notes, not our base case.

GUY: So the disagreement between Excess Returns and Monetary Matters is useful, not confusing. Constan sees growth plus supply pressure, while Gromen sees an emerging hard-currency debt spiral. Both direct us to the same observable evidence: auction tails, dealer take-down, term premium, corporate spread concessions, and newly issued security performance. That is how we distinguish ordinary supply indigestion from a structural loss of confidence.

AVA: Let us move into technology with Thoughts on the Market. Morgan Stanley's Ariana Salvatore described the American approach to sovereign AI as selective access: preserve national-security guardrails around sensitive capabilities while keeping the broader American stack available to allies. Stephen Byrd described China's approach as more indigenous across chips, compute, cloud, and models, combined with lower-cost open-weight models, subsidized compute, cloud partnerships, and infrastructure exports.

GUY: Thoughts on the Market says the inefficiency is the point. Countries want compute, data, energy, and technology under greater sovereign control. Sensitive data stays local, cloud and cybersecurity arrangements diverge, and workloads require more geographically distributed data centers, networking, power, cooling, and infrastructure software than an integrated world would. Fragmentation can therefore reinforce nominal AI capital expenditure even as it raises the cost of delivering it.

AVA: Thoughts on the Market also identified power as the constraint that determines who can actually deliver. Local ratepayers may resist data-center-driven electricity costs. That increases the value of low-cost generation, behind-the-meter supply, off-grid solutions, storage, and grid interconnection. The thesis weakens if standards remain interoperable, localization eases, utilization offsets duplication, or compute efficiency outruns workload growth. It strengthens with more export controls, localization rules, national compute subsidies, and power bottlenecks.

GUY: The Indicator from Planet Money gave that power story an unexpected manufacturing link. Cooper Katz McKim reported a twenty-seven percent year-over-year decline in North American electric-vehicle sales since the prior July, while the rest of the world grew more than ten percent. He linked part of the divergence to policy support, noting that Spain offered buyers more than five thousand dollars while the U.S. federal purchase incentive had ended the prior September.

AVA: The Indicator then followed the second-order effect. Battery factories built since twenty twenty-two for electric vehicles are being redirected toward grid-scale storage. Ford, General Motors, and LG were named, and Ford planned to convert a large EV battery facility into a grid-storage hub. Data centers create a new demand pool for storage as power loads rise and intermittent generation needs balancing.

GUY: But The Indicator does not say every stranded battery plant becomes a good data-center asset. The underwriting variables are cell chemistry, cycle life, warranty, system integration, interconnection, and customer concentration. The causal chain is attractive: weaker regional EV demand, manufacturing redeployment, grid-storage supply, then data-center resilience. Yet the bottleneck migrates from factory capacity to whether the converted product can earn an adequate stationary-storage return.

AVA: Now the application layer. The Vergecast's authoritative notes described an identity crisis for Alexa Plus and Gemini for Home. Jennifer Pattison Tuohy and Nilay Patel tested natural-language control of thermostats, locks, pet monitoring, and household memory. The notes describe forgotten cats and hit-or-miss task completion. The important contrast is that sovereign infrastructure spending can rise even while consumer assistants remain unreliable.

GUY: The Vergecast therefore points us toward utility metrics rather than demo excitement. The written brief identifies repeated-command completion, median and tail latency, error and retry rates for locks or thermostats, accuracy of persistent household memory, the privacy burden of cameras and ambient collection, and monetization that does not turn the assistant into an advertising channel. Infrastructure demand is not proof of application-layer product-market fit.

AVA: Hard Fork's authoritative show notes add the frontier-model security layer. The episode says OpenAI stopped training new models while reviewing security measures and asks whether other laboratories will also slow. It pairs that issue with historian Jill Lepore's warning about an artificial state, meaning rule by machines manufactured by corporations, and with Google's planned ten-million-dollar purchase of defunct Spirit Airlines business data.

GUY: Hard Fork's available notes do not establish the hosts' full verdict, so we should not invent one. What the episode does support is a shift in the industrial process. Frontier training is no longer only about scaling. It requires isolation, security review, monitoring, and proof that safeguards keep pace. The Spirit data proposal also shows that failed companies can leave economically valuable operational histories, customer interactions, pricing records, and processes as model inputs.

AVA: Hard Fork makes the regulatory questions concrete: who can train on operating datasets, what security standards apply during training, what disclosure follows a pause, and when a privately owned recommendation system starts functioning like public infrastructure. The near-term test is whether OpenAI resumes large training efforts under disclosed controls and whether peers adopt comparable pauses or monitoring. The official notes do not support a completed policy prescription.

GUY: The Meb Faber Show gives us the process lens. Its authoritative notes say David Booth traced Dimensional Fund Advisors from early index work at Wells Fargo and study under Eugene Fama to more than one trillion dollars in assets. He discussed the Fama-French three-factor model, education-led adviser relationships, small-cap value, AI as a California gold rush, and judging decisions by process rather than outcome.

AVA: The Meb Faber Show's process principle belongs in today's market. A rising AI security can reflect a correct thesis, abundant liquidity, or both. A falling new issue can reflect bad economics, an excessive offer price, or temporary supply. The disciplined approach is to record cash-flow expectations, financing needs, issue price, catalyst, and failure condition before the market outcome. Post-issue performance updates the thesis; it does not retroactively define the quality of the decision.

GUY: On geopolitics, Thoughts on the Market frames AI as a national-resilience stack. Export controls, tariffs, domestic manufacturing incentives, localized data, and subsidized compute increasingly define the market itself. Countries between the United States and China want resilience and flexibility. Vendors therefore win not only through model quality but through power access, sovereign compliance, cyber assurance, colocation, networking, cloud distribution, and supply-chain access.

AVA: Goldman Sachs Exchanges broadens that capital competition beyond AI. The episode identified three sources of back-end bond supply: American hyperscaler capital expenditure, European defense, and Japanese fiscal policy. The Indicator adds U.S. Treasury debt, while Excess Returns adds the possible reversal from corporate buybacks to net issuance. Geopolitics becomes a cost-of-capital variable before it changes revenue because defense, energy security, sovereign compute, and fiscal spending all compete for long-duration capital.

GUY: The second cross-current comes from Goldman Sachs Exchanges and Excess Returns. Broadening is both a diversification opportunity and a financing warning. Equal-weight leadership and nine sectors with double-digit earnings growth reduce mega-cap dependence. But if net buybacks swing toward net issuance, a persistent source of equity demand disappears. We need to monitor R S P relative strength, revision breadth, buyback authorization, and net issuance together.

AVA: The third cross-current combines Thoughts on the Market with The Indicator. Sovereign duplication creates local data centers, local data centers create local electricity load, and storage becomes one tool for congestion and intermittency. Underused EV battery capacity may help, but chemistry suitability, interconnection, dispatch economics, and customer credit determine the return. The physical bottleneck can move even while headline demand keeps rising.

GUY: The fourth cross-current connects The Vergecast, Thoughts on the Market, and Hard Fork. Infrastructure demand, frontier-model safety, and consumer utility are three separate adoption curves. Governments can force compute duplication. Labs can pause training for security review. Household assistants can still forget the cat. A booming data-center market therefore does not validate every consumer agent, and unreliable consumer software does not disprove policy-driven infrastructure demand.

AVA: One more evidence boundary from the written brief. Four of today's eight episodes had full transcripts, while Monetary Matters, Hard Fork, The Vergecast, and The Meb Faber Show were supported only by authoritative RSS notes. That means no quote-level tone analysis and no conclusions beyond those summaries. A false Morgan Stanley match and an unrelated StarTalk item returned by the configured No Priors feed were quarantined and did not enter the analysis.

GUY: The actual No Priors episode featuring Sarah Guo and Max Hodak was published sixty-four minutes before the twenty-four-hour cutoff, according to the written brief, so it was excluded as stale. That is a useful reminder: a complete daily product is not one that stuffs in every interesting item. It is one that respects identity, time, and evidence gates, then states what remains unavailable.

AVA: So what are we watching? Goldman Sachs Exchanges points first to Jackson Hole in the week of August twenty-four and Chairman Warsh's treatment of balance-sheet policy, long-end yields, and forward guidance. A preference for market-led tightening would reinforce the duration-supply thesis. Explicit support for the long end would weaken it. The remaining second-quarter reports also test whether the sixty-four percent large-beat rate and revision breadth survive the final eight percent.

GUY: Excess Returns and The Indicator put Treasury auctions and large AI-linked financings next on the list. Watch bid-to-cover, auction tails, dealer allocation, duration mix, book coverage, pricing concession, and aftermarket performance. The funding-risk view weakens if new deals remain well absorbed, issued securities stay above offer prices, spreads stabilize, buybacks remain material, and capital expenditure converts into wider earnings revisions.

AVA: Thoughts on the Market says to monitor new export controls, data-localization mandates, national compute programs, data-center permitting, grid pricing, and behind-the-meter power contracts. The Indicator says to watch North American EV units, factory-conversion milestones, stationary-storage orders, interconnection, warranty terms, and utilization at Ford, General Motors, and LG-linked facilities. Those indicators tell us whether sovereign duplication and battery redeployment are becoming cash flows rather than narratives.

GUY: Hard Fork says the frontier-model catalyst is whether OpenAI resumes its largest training efforts and discloses stronger controls, and whether competitors respond similarly. The Vergecast says consumer assistants must prove reliability through repeated-task completion, latency, memory accuracy, privacy disclosure, and monetization. Nvidia's next earnings update and hyperscaler capex guidance are also key tests, although the valid source set did not establish exact reporting dates.

AVA: The portfolio conclusion is selective rather than dramatic. Stay constructive on earnings breadth while treating financing capacity, not end-market AI demand, as the nearer tail risk. Decompose the AI basket. Separate the duration hedge from the equity thesis. Use primary-market plumbing as a leading indicator. And treat gold versus long Treasuries as a policy-confidence diagnostic, not proof of the Monetary Matters gold-revaluation scenario.

GUY: Exactly. The favorable state is broader upward earnings revisions, differentiated leadership, stable credit spreads, successful issuance, and inexpensive index volatility. The adverse state is worsening post-issue performance, wider debt and equity concessions, higher long yields despite cooler data, negative technology reactions to beats, and capex plans becoming conditional on external funding. That is a falsifiable framework, not a slogan.

AVA: That is the Morning Signal for Friday, August twenty-first. The build remains powerful, but power, duration, and funding now decide who gets through it. We will be back after the next evidence arrives.

GUY: Have a good Friday. Watch the revisions, watch the auctions, and watch what happens after the deal prices. See you next time.