SaaSCreditMispricedasPEPlaybookBreaks,AIAttachSoars

◆ DEEP DIVES

1. 01

   Software Credit at 90¢ While AI-as-Bundle Prints 32% Growth — The Mispricing Is Two-Directional

   The Dislocation

   Software-backed loans are trading at roughly 90 cents on the dollar and now account for about 35% of all distressed loans, which would be a coherent story if default rates had moved. They have not. The gap between how the paper prices and how the underlying cash flows behave is the widest it has been in years. Thoma Bravo walking away from Medallia — the $6.4B take-private from 2021 — is the clean proof that the leveraged SaaS buyout playbook, or rather the version of it that assumed cheap debt forever, has broken. Debt service went from $100M to $300M on floating-rate terms, multiples compressed, and AI disruption made the exit math unworkable. Blackstone declined to extend a lifeline.

   The PE software-buyout model is not returning in the shape it left. Two of its three legs, cheap debt and cooperative exit multiples, have stopped functioning.

   The Counter-Signal Nobody Expected

   In the same week Atlassian reaccelerated revenue from 23% to 32% YoY and the stock ripped 24% after-hours out of a 57% YTD drawdown, which is a lot of movement for a name the market had collectively decided was roadkill. The driver is that Rovo customers generate 2x the ARR of non-Rovo customers. That is the cleanest public proof point on offer that AI bundled into existing SaaS as a premium tier drives expansion rather than displacement. The 'AI kills SaaS' thesis has been compressing enterprise software multiples for six quarters. It just took its most direct hit.

   Why Both Directions Matter

   The mispricing is running in both directions at once. On credit, performing SaaS loans with workflow-of-record positioning and real data moats are being marked next to the Medallia-adjacent paper that genuinely deserves distressed treatment. On equity, public software multiples sit depressed under the AI-displacement narrative that Atlassian's print just partially retracted, and those depressed public comps are anchoring private AI-application valuations below where the mobile-cycle analogy says they should eventually trade.

   The a16z lens is the useful historical frame. The mobile cycle ran about five years from picks-and-shovels (Qualcomm/ARM) to application winners (Uber/Airbnb/Instagram), and we are roughly two years into the AI equivalent. Private AI-application positions taken now sit roughly two to three years ahead of where public markets will eventually mark them, if the analogy holds. It may not. Analogies rarely do at the edges.

   What To Own vs. What To Avoid

   Own	Avoid
   Performing SaaS loans at 85-92¢ with workflow-of-record moats	Survey, feedback, forms, basic analytics, rules-based workflow
   Private AI apps at Series B/C with $5-25M ARR and data moats	AI wrappers without proprietary data or distribution
   Enterprise SaaS with AI-bundle upside (Rovo template)	Floating-rate debt above 3x EBITDA with AI exposure

   ---

   The Risk

   The honest caveat is that the dislocation could deepen before the picking pays, and patient capital tends to discover it was less patient than it advertised. Jones's 252% market-cap-to-GDP ratio, against 170% in 2000, means the macro will not help with timing. The position that survives both a compression scenario and a recovery is long performing-SaaS credit, long private AI apps, short AI-exposed leveraged SaaS, and prioritizing DPI over TVPI markups in every LP conversation.

   Action items

   • Stand up a software-credit opportunistic sleeve targeting performing SaaS loans at 85-92¢; screen out AI-exposed categories (survey, analytics, rules-based workflow)
   • Audit every PE secondary and portco position with >3x debt/EBITDA for AI-disruption exposure using the Medallia template
   • Screen enterprise SaaS portfolio companies against the Rovo template — identify who can drive 2x ARR lift via AI-tier attach rates
   • Bias LP communications toward DPI over TVPI mark-ups through Q3 2026

   Sources:a16z · Martin Peers · The Information AM

2. 02

   Agent Commerce Infrastructure Just Went Live — And the Platform Layer Is Already Claimed

   The Week Agents Got Wallets

   Four announcements landed in the same cycle and collectively did something more interesting than any of them did alone, which is stand up a new category of infrastructure for machines to transact. Cloudflare and Stripe shipped a protocol that lets AI coding agents provision accounts, buy domains, and spin up paid subscriptions on their own, with Stripe doing identity attestation and a $100/month default spend cap standing in as the trust primitive (the phrase 'default' is doing real work there). Meta launched USDC creator payouts on Polygon and Solana, routed through Circle and Stripe, live in more than 160 markets. Link CLI and Meta Ads CLI handed agents working payment and ad-buying primitives in the same week. And OpenAI quietly pulled back on Instant Checkout, citing fraud safeguards, which is the tell: the identity and trust layer does not yet exist.

   Agents now have money-in and money-out. The infrastructure sitting around those primitives — spend governance, agent identity, attribution, eval, abuse detection — is where the next cohort of AI-infra winners will come from.

   Stablecoins Are the Settlement Layer

   The parallel story, and arguably the more important one, is that stablecoins finished their migration from DeFi thesis to mainstream payment plumbing. Visa's stablecoin settlement hit $7B annualized, up fifty percent quarter-on-quarter, across nine chains and more than 130 card products. Circle has put together the full vertical stack while no one was watching: USDC issuance, the Arc L1, nanopayments across eleven chains, and the default routing for Meta via Stripe. Five US regional banks — Huntington, First Horizon, M&T, KeyCorp, Old National — are moving deposits on-chain through ZKsync Prividium. That is the first tier-two bank migration that counts.

   Update the sector thesis. Replace 'crypto disrupts card networks' with 'crypto is the settlement substrate beneath incumbent rails.' Visa is co-opting this, not being disrupted by it, and the authorization-versus-settlement distinction is the one that will carry deal evaluation for the next twenty-four months.

   The Standards War Nobody Has Won

   Agent payments is a pre-revenue, pre-consolidation standards fight. Three protocols are racing for capture and none of them has won:

   Protocol	Approach	Partners
   Stripe Link	OAuth-based agent spend delegation	Meta, Cloudflare
   OKX APP	Open protocol, multi-chain	AWS, Alibaba, Ethereum, Solana, Base, Uniswap
   AgentCash / x402	HTTP 402-based API payment rails	Early-stage

   Adjacent to this, the KYA (Know Your Agent) framework from AstraSync and Persona extends KYC one layer up into agent identity and accountability, which is either the obvious next primitive or a solution hunting for a mandate. OpenAI's retreat from Instant Checkout is the de-risking signal the tier-one generalists have been waiting for.

   ---

   Where the Alpha Sits

   This is probably wrong in one direction or another, but the asymmetric wedge is the vertical governance layer between Stripe's generic hundred-dollar cap and what regulated industries actually require. Healthcare agent procurement, financial services spend controls, SOX and HIPAA-compliant audit trails — none of that gets served by a platform default. The cost of chasing that wedge is not chasing the horizontal protocol fight, which is crowded and already well-funded. The seed and Series A window stays open for perhaps two quarters before the standards consolidate and pricing moves to consensus.

   Action items

   • Open a focused seed/Series A thesis on AI agent commerce infra — target 3 investments across agent wallets, agent identity, and agent-to-agent escrow before standards consolidate
   • Source institutional DeFi/ZK deals with live bank pilots — prioritize compliance-native rails with signed tier-2/3 bank LOIs
   • Update sector thesis doc: move 'crypto disrupts card networks' to contrarian short; replace with 'crypto is settlement substrate under incumbent rails'
   • Map the agent commerce stack: identify 5 companies building vertical agent spend governance for regulated industries before a16z publishes the obvious memo

   Sources:TLDR Crypto · TLDR DevOps · TLDR Founders · Risky.Biz · TLDR Product

3. 03

   Apple Changed From Dividend Machine to Dry-Powder Machine — And the Memory Supercycle Is the Catalyst

   The Balance Sheet That Moved

   On the April 30 call Apple did something it had not done since 2018, which is abandon the net-cash-neutral doctrine it had been quietly defending for years. CFO Kevan Parekh confirmed the target is dead. Buybacks were halved while free cash flow grew twenty-eight percent, R&D jumped thirty-four percent, and the net cash pile has fallen from $163B to $62B since 2018. Tim Cook, who does not mention costs for sport, flagged 'skyrocketing memory chip prices.'

   The honest read, or rather the more interesting version, is that Apple is clearing the deck for an AI-device acquisition cycle under incoming CEO John Ternus on September 1. This is probably wrong in its specifics, but it redraws the strategic-buyer map for anyone holding AI silicon, health, or hardware-adjacent assets. What Apple is no longer doing is returning every spare dollar to shareholders.

   Apple as a plausible strategic acquirer changes the exit math on anything adjacent to devices, silicon, or health — and the portfolio should be positioned to be bought, not just built.

   The Memory Tax on Everything

   AI training and inference have quadrupled RAM prices, and Samsung's memory division printed $30B+ in a single quarter, a forty-eight-fold year-on-year profit jump. Cook's warning formalizes what memory buyers figured out months ago: hyperscalers pulling forward HBM orders are crowding out everyone else, and 'everyone else' is the consumer hardware cohort investors were treating as a cyclical recovery.

   Apple absorbs this through stockpiled inventory, which is a competitive advantage smaller OEMs don't have and cannot assemble in two quarters. Margin compression across consumer hardware starts Q3. Multiple expansion for Micron, SK Hynix, and Samsung HBM lines persists, with the supply crunch forecast through 2027.

   The Bifurcation in Context

   Apple hoarding cash while Meta issues $55B in bonds over six months is the kind of divergence worth staring at. Meta fell eight percent on raised capex guidance, which is the first real crack in the market's willingness to fund AI infrastructure without revenue attach. AI infrastructure is now a credit story at Meta, and bond markets do not forgive narrative the way equity markets do.

   Company	Capital Posture	Market Reaction
   Apple	Buybacks halved, FCF +28%, cash hoarding	Stable (+0.4%)
   Meta	$145B capex, $25B bonds, capex raised	-8% on earnings
   Alphabet	$35.7B capex, cloud +63%	+10% record

   The market has decided that capex that demonstrably converts to cloud revenue, which is Alphabet, gets rewarded, and capex that requires a narrative bridge to monetization, which is Meta, gets punished. That distinction flows into private AI infrastructure multiples on roughly a two-quarter lag.

   ---

   Portfolio Implications

   The memory supercycle is the under-modeled cost shock sitting in every AI hardware company's COGS line, and portfolio companies with more than fifteen percent BOM memory exposure need stress-testing for Q3-Q4 2026 margin impact. The opportunity, assuming Apple does behave like a plausible acquirer and memory scarcity holds, splits into two trades: position portcos as Apple M&A candidates across AI silicon, health wearables, and AI-native consumer hardware, and build memory supply chain exposure through HBM specialists, thermal management, and advanced packaging. Either the acquirer shows up or it does not. The memory bill arrives regardless.

   Action items

   • Map portfolio companies fitting Apple's likely M&A targeting — AI silicon, health/wearables, AI-native consumer hardware — and draft strategic narratives for Q3/Q4 outreach before Ternus takes the chair
   • Stress-test portfolio companies with >15% BOM memory exposure for Q3-Q4 2026 margin impact; require hedge/stockpile disclosure at next board meeting
   • Build a shortlist of memory supply-chain longs (HBM specialists, DRAM pure-plays, thermal management, advanced packaging) for LP co-invest sidecar
   • Reduce exposure to leveraged AI infra names where cash flow coverage depends on hyperscaler capex continuing at current clip

   Sources:Martin Peers · Bloomberg Technology · Techpresso · Morning Brew