Stripe Wants the Network

If You Only Read One Thing

Stripe is bidding for a network it cannot code, while IBM's customers are reserving parts they cannot postpone. Stripe Wants the Network and Hardware Eats the Quarter look like separate M&A and earnings stories; IBM's own explanation shows the common balance-sheet logic. Scarcity prices the waiting time, so whoever can buy years of distribution or months of component supply gets to move first.

Stripe Wants the Network

Stripe may be about to spend more than $53 billion admitting that payments infrastructure is easier to build than payments distribution.

Stripe and private-equity firm Advent International have offered $60.50 a share for PayPal, a 28% premium to its prior close, Reuters reported. Neither company initially confirmed the approach. If completed, the transaction would join a private payments company valued at $159 billion in its latest employee tender with an incumbent whose shares have lost roughly 80% over five years.

That headline makes the bid look like a generational changing of the guard. The more useful reading is that Stripe wants to buy three distribution layers it cannot cheaply reproduce: PayPal's branded checkout identity, Venmo's consumer graph and Braintree's relationships with large merchants. Stripe already has scale. It said businesses on its platform processed $1.9 trillion in 2025, up 34%, or about 1.6% of global GDP. What it lacks is a consumer destination that merchants already recognize and a second enterprise gateway already embedded across large accounts.

Why it matters: Payments is shifting from a contest over transaction processing to a contest over the entire commerce relationship. Processors have been compressed by competition and merchant bargaining power; checkout identity, fraud data, lending and consumer demand carry better strategic value. Combining Stripe's developer distribution with PayPal's consumer and enterprise surfaces would create a company able to see a purchase from software integration through shopper authentication and settlement.

Advent's role is the second tell. A financial sponsor can supply debt, absorb assets Stripe does not want and make a politically awkward whole-company acquisition look more like a portfolio allocation exercise. PayPal could be acquired as a bundle and later separated into consumer, Venmo and Braintree pieces. The offer may therefore be less about preserving PayPal than assigning its mature networks to new owners.

Room for disagreement: PayPal's assets are not interchangeable with growth. Branded checkout grew just 2% in the first quarter, according to the Associated Press, and Apple has overtaken it at checkout. Stripe could pay a premium for integration complexity while inheriting the decline it is trying to escape.

What to watch: The decisive variable is whether Advent appears as a permanent co-owner or as financing for a pre-announced separation of PayPal's consumer, Venmo and Braintree assets. A separation plan would confirm that distribution, not corporate scale, is the scarce asset.

Hardware Eats the Quarter

IBM just gave the clearest account yet of how the AI hardware shortage reaches a software income statement: the same customer cannot spend the same dollar twice in the same quarter.

IBM disclosed preliminary second-quarter revenue of $17.2 billion, up 1%, and adjusted earnings of $2.93 a share. Analysts had expected $17.86 billion and $3.01, according to the Associated Press. The shares fell 23%. Software grew 5%, consulting was flat and infrastructure declined 7%, a mix that initially looks like weak execution everywhere.

Chief executive Arvind Krishna supplied a more specific mechanism. In late June, customers shifted capital toward servers, storage and memory to secure supply before expected price increases, causing large software transactions to slip beyond the quarter. IBM also faltered on some deals. Yet Red Hat grew 11%, distributed infrastructure rose 37% and IBM ended the period with a $500 million backlog there. The problem was not the disappearance of enterprise demand; it was the order in which constrained budgets could satisfy it.

Why it matters: AI infrastructure inflation is becoming an internal capital-allocation shock for ordinary enterprises. Memory and storage are perishable purchasing opportunities when supply is tight and prices are rising; a software contract can usually wait several weeks. That gives component suppliers and hardware vendors bargaining power over software companies without competing with their products at all.

The pressure extends beyond IBM. IDC says worldwide server spending grew 30.7% in the first quarter and expects constrained memory and NAND pricing through at least the first half of 2027. If enterprise budgets remain fixed, every urgent hardware order creates a queue elsewhere: applications, consulting projects and cloud migrations. The AI buildout is not merely raising data-center costs. It is changing the timing and reported growth of the rest of the technology sector.

Room for disagreement: IBM may be turning an execution miss into a macro story. Its infrastructure decline, missed large transactions and reliance on mainframe cycles are company-specific. A delayed software contract is also not revenue until it closes, and customers can cancel rather than merely postpone.

What to watch: The clean external test is whether enterprise-software peers describe the same hardware-first budget sequencing in their next results. If they do not, this was IBM's quarter, not the industry's constraint.

The Contrarian Take

Everyone says: IBM's miss shows that the AI spending boom is finally cannibalizing enterprise software.

Here's why that's wrong (or at least incomplete): Cannibalization implies customers have chosen hardware instead of software. IBM's evidence points to sequencing: Red Hat still grew 11%, infrastructure demand produced a backlog and late-quarter software deals moved out rather than vanished. That distinction matters. A permanent substitution would compress software valuations and product demand; a working-capital queue creates ugly quarters but preserves the contracts. The bearish case becomes convincing only if those deals fail to close after constrained hardware has been secured.

Under the Radar

• Dependabot is adding a quarantine period — GitHub will now wait three days before opening routine version-update pull requests, while security updates remain immediate. The default cooldown gives maintainers time to discover a poisoned release before automation spreads it downstream. Software supply-chain defense is moving from checking code to controlling how quickly trust propagates.

• A username became a root shell — Tailscale patched an SSH flaw in which a username beginning with a hyphen could be interpreted as a command option and produce a root session despite access rules intended to allow only non-root users. The narrow security bulletin carries a broad lesson: identity-aware access systems remain exposed where structured policy eventually meets old command-line parsing.

Quick Takes

• OpenAI wants the computer after the screen — Its first hardware device is reportedly a mobile, screenless speaker with cameras, sensors and mechanical movement that can use email and other digital context. Calling it a companion is marketing; the strategic bid is to own an ambient interface before phones make assistants into interchangeable apps. The hard problem will be earning permission to observe enough context to be useful. (Source)

• Meta's management system is now the defendant — Twenty-six employees allege the company used internal AI, activity tracking and performance dashboards to select them for layoffs, disadvantaging workers with disabilities or protected leave. Meta says humans made the decisions. The suit matters because algorithmic management makes discrimination easier to scale and harder to locate: the contested act can sit in a ranking pipeline long before a manager signs the termination. (Source)

• Washington built a front door for software flaws — The White House's Gold Eagle initiative will accept, prioritize, verify and coordinate remediation of vulnerabilities across agencies and industry partners. Central intake is useful, but coordination without patch deadlines or replacement budgets can become a better catalog of unfixed risk. Its value will be measured by how quickly verified findings change deployed systems, not how many organizations join. (Source)

The Thread

These stories separate reversible decisions from perishable ones. Stripe can keep building payments infrastructure, but PayPal's checkout graph may change owners before Stripe can reproduce it. IBM customers can sign software later; they cannot recover memory allocations that another buyer takes now. Dependabot deliberately makes upgrades safer by waiting, while Tailscale shows a modern policy undone at the final parsing step. The advantage belongs to whoever decides when waiting is safe and when delay forfeits the asset.

Predictions

New predictions:

• I predict: By August 15, PayPal's board will either formally reject the $60.50 offer or invite a higher or revised proposal rather than accept the reported terms unchanged. (Confidence: medium; Check by: 2026-08-15)

• I predict: IBM's July 22 results and call will attribute most of the shortfall to transaction-processing, mainframe and deal-timing effects without reporting a broad deterioration in Red Hat demand. (Confidence: medium; Check by: 2026-07-23)

Issue date: July 15, 2026 · Generated: 3:28 AM ET