The week's signal is repeated divergence — between what the infrastructure shows and what the price reflects. The S&P 500 just completed the fastest V-shaped recovery on record from the Iran shock; meanwhile Buffett sits on a $373bn cash pile, the largest in Berkshire history, and Paul Tudor Jones is calling 252% market-cap-to-GDP. AI capex pours into chips and power while software multiples have already mean-reverted. Polygon picks up Meta, Visa and Modern Treasury in the same week Polymarket signals exit. Enterprise merchants run the most informationally rich fraud check in the payments chain — and pay for being wrong twice while only getting credit for being right once. The question for capital allocators is which side of these divergences is the leading indicator.
Chamath Palihapitiya's Q1 deep dive maps the AI stack as six layers — infrastructure, chips, data, models, execution, application — each with a "fulcrum asset" through which all value above it must pass. The strategic point lands hard: the foundation is the most concentrated layer, and the concentration is global, not American. ASML in the Netherlands prints every advanced chip in the world. Four Japanese companies supply the film no chip ships without. A single mine in North Carolina sits beneath every wafer in production. NVIDIA's CUDA runs on top of all of it. At the chip layer the stack forks — software AI, where the price of running a model has dropped 1,500× in six years and intelligence is approaching free; and physical AI, where the binding constraints are energy storage and actuation. The pattern echoes Rockefeller at 90% of refining and Cisco at 85% of routing. The names that will define the next forty years are being claimed at the fulcrum points now.
Goldman Sachs Research now estimates AI has reduced monthly US payroll growth by 16,000 jobs over the past year. The number is small relative to a 160-million-person workforce but directionally significant — AI is a measurable subtraction from the jobs print, not yet a replacement story. The macro consequence: Warsh's view that AI adoption will prove disinflationary now has a payrolls vector behind it.
a16z argues that critical infrastructure abstractions — paper checks, online checkout, energy markets — are not collapsing but leaking, as governments and platforms try to grab the controls underneath. The DOGE directive ending federal paper cheques is the cleanest example. The second-order consequence is rational actors hardening their defences. For incumbents: the cost of operating below the abstraction is rising, and reliable interfaces will command a premium.
Goldman Sachs Research forecasts robotaxis will reach $400bn in revenue by 2035. The number frames the addressable market for the autonomous-vehicle stack and explains why the platform aggregators (Uber positioning itself as the open layer) and the technology owners (Tesla, Waymo, the Chinese players) are now in active jostling for share. The strategic question for investors is how that pool divides between the platform and the underlying technology.
a16z this week reframed an old observation: semis and AI infrastructure trade at a substantial premium while the putative beneficiaries of AI trade at little-to-no premium. Software multiples on a growth-adjusted basis have already mean-reverted to the prior decade's average. The post-GFC mobile cycle ran the same pattern — semis first, software second, with a five-year lag. Reading the same script forward: the AI software re-rate has not yet started.
Dwayne Gefferie's forensic piece on enterprise fraud decisioning lands a structural point. The merchant runs the most informationally rich fraud check in the payments chain — several hundred to a couple of thousand features per transaction, against the 25–60 ISO 8583 fields that reach the issuer. Identity, behavioural cadence, full device fingerprint, network reputation, deep historical velocity. Stripe states roughly 90% of cards on its network have been seen before. Yet the cost asymmetry is brutal: a wrong approval means the chargeback, the cost of goods, the shipping, the interchange, and exposure under Visa's monitoring programme. A wrong decline means a customer who never comes back. Both outcomes cost the merchant; only one appears on the dashboard. The Merchant Risk Council pegs typical false-positive rates at 2–10%, with one in five merchants above 10%. The threshold is a business decision, not a technical one — and the gap between data-science ownership of the model and ops ownership of the rule layer is where most of the value leaks.
The Fintech Wrap Up reports converge on a thesis: a real-time, composable growth layer is forming above legacy cores. Core stays the system of record; the operating system becomes the system of growth — independent of core downtime, capable of activating new payment rails without integration friction. Twenty years of architectural accumulation (middleware, point solutions, BaaS) treated symptoms. This treats the cause. For mid-sized banks fragmented in modernisation, the OS layer is the more capital-efficient route than another core replacement.
Levine's thesis runs against the grain of the agentic-checkout narrative. The substantive shift is not AI agents buying shoes — it is the rewiring of commerce infrastructure underneath: authorisation frameworks, liability rails, reconciliation, identity. Consumer agentic shopping is the surface story. The deeper story is that the rails were built for human-initiated transactions, and re-engineering them for delegated authority at scale is the unglamorous, multi-year build where the value will sit.
Visa added Polygon to its stablecoin settlement programme this week, alongside four other networks. The headline number from Visa: $7bn in annualised volume, growing roughly 50% quarter-on-quarter — implying a near-term run rate above $10bn. Meta also went live with USDC creator payouts via Stripe (Polygon and Solana settlement), live in Colombia and the Philippines with 160 markets planned by year-end. This is Meta's first stablecoin product since Diem was shelved in 2022; using a regulated third-party stablecoin under the GENIUS Act framework rather than building its own is the strategic shift.
Robinhood fell 12.75% on the week and SoFi 12.12% after Q1 earnings. Both topline beats; both punished. HOOD missed on net revenue ($1.07bn vs $1.14bn) with crypto revenue down 47% YoY to $134m. SOFI delivered record adjusted revenue but Galileo platform revenue fell 27% on Chime exiting, and personal and student loan charge-offs ticked up. The market is repricing crypto-distribution names harder than the underlying assets. Coinbase's 7 May print sets the near-term ceiling.
Two decades of bank transformation have focused on capturing data and improving digital experience. The execution layer has not moved as much. Most work still depends on manual coordination across systems, and even when insights are available, follow-up is inconsistent. Autonomous AI agents close that gap by managing outcomes within defined guardrails, not just delivering recommendations. The architectural shift is from AI as analyst to AI as operator — capable of completing defined tasks, following rules, and moving processes forward across systems. The competitive question for banks is which functions will move first (operations, treasury, compliance, servicing) and where the human-in-the-loop boundary settles. The answer determines unit economics for the next decade.
The S&P 500 sold off roughly 10% during the Iran conflict and recovered in 11 trading sessions — the fastest V-shaped recovery ever on record. The index closed Friday at an all-time high of 7,165, up 30% year-on-year. Two of the most successful investors of the past fifty years are not buying it. Buffett: "this is nothing… we aren't in it to make 5% or 6%" — and Berkshire sits on $373bn of cash, the largest pile in its history. Paul Tudor Jones: the market is at 252% of GDP versus 170% in 2000 and 65% in 1929; a mean-reversion to the 25–30 year P/E implies a 30–35% decline. The structural counter-argument is that prices are higher because there is more cash to chase them — money market fund balances at $8trn, double pre-pandemic; roughly 30% of all US dollars in M2 today were created in the last five years. Profit margins are also double the 1990s average and forecast earnings growth more than double the long-term average. Either the cash dynamics have repriced equity multiples permanently, or the smart money is right about the reversion. The bond and dollar markets will adjudicate first.
Kevin Warsh is expected to be confirmed as Fed Chair ahead of the June FOMC. Per Goldman's Rob Kaplan, three signals matter: a high bar for quantitative easing, scaling back the dot plot, and broader use of alternative inflation gauges (Warsh referenced the Dallas Trimmed Mean in confirmation). His personal lean is dovish — AI and Chinese overcapacity as disinflationary forces — but he needs seven votes, and the committee remains scarred by the "transitory" call. Visible inflation improvement is the gating condition for cuts.
Headline PCE printed 3.5% YoY on 30 April — the hottest reading since May 2023. The FOMC held the policy rate at 3.5–3.75%. The macro backdrop turned hawkish into the move, and risk assets felt it: spot Bitcoin ETFs flipped to three consecutive days of net outflows totalling roughly $491m into FOMC week. The oil-price shock from the Middle East is doing real work in the inflation prints. Warsh's confirmation timeline now coincides with the question of whether the disinflation trajectory has stalled.
Goldman Sachs Research notes some retail investors are withdrawing from private credit funds, but the underlying principles of the asset class are expected to remain intact. The retail leg was the marginal flow of the cycle, and its softening was always more likely to be a liquidity-mismatch story than a credit-quality one. For the institutional thesis — insurance balance sheets, illiquidity premium, direct-origination yields — nothing has changed. The retail outflow is a feature of the product structure, not a verdict on the asset class.
Pattern recognition from a16z: in the post-GFC mobile cycle, semis and infrastructure (Qualcomm, ARM) led from 2010–2012; platform layers (Apple, Samsung) carried 2013–2014; software emerged as the clear market leader only by year five. The current AI cycle is reproducing the early phase — semis and AI infrastructure ahead, software trailing. If the pattern holds, the software re-rate is a 2027–2028 story. The risk is that the analogy breaks because AI commoditises application logic faster than mobile did.
ARK's update this week underlines a structural shift. ETFs and public companies now hold roughly 12% of total bitcoin supply; price drawdowns across multiple time horizons have reached the shallowest levels on record; and BTC's correlation with traditional risk assets is diverging from its historical pattern. The framing has moved from "digital gold" to "potential safe haven in a shifting global landscape" — and the price action through the Iran shock corroborates it: BTC held the $76,000 zone, stalled twice at $80,000, and traded with materially less volatility than the equity selloff. The investment-committee implication is that bitcoin's portfolio role is no longer purely speculative; it is becoming an allocator question rather than a thesis question. Whether the institutional bid sustains through a genuine credit event remains the only test that matters.
In a single week Polygon picked up Meta (USDC creator payouts via Stripe), Visa (stablecoin settlement), and Modern Treasury (Payments API). The chain ran roughly $37bn in stablecoin volume over the trailing 30 days and generated $11m in Q1 network revenue, up from $2.1m the prior quarter. In the same week, Polymarket VP of Engineering Josh Stevens publicly named "chain migration" a roadmap priority, citing block space, gas costs, and block times. Polymarket generates $2.5m–$4m in weekly fees on Polygon — 50–70% of total chain transaction-fee revenue — and occupies three of the top five gas-consuming contracts. POL closed +0.92% against a basket median of -3.70%: the market is pricing both halves of the story. Polygon is becoming an institutional payments rail at the moment it risks losing its largest crypto-native consumer application. May's question is whether it can be both at once.