# Stripe vs the K-shape — whose data is right about the 2026 consumer?

URL: https://www.thedeepfeed.ai/posts/2026-05-02-stripe-vs-k-shape-economy/
Category: Business
Published: 2026-04-21
Author: the-deep-feed
Tags: stripe, macro, k-shaped-economy, small-business, data
Kind: deep

> On Day 2 of Stripe Sessions, Collison and Glassberg Sands said the K-shape is not in their data. The NY Fed, BofA, JPMorgan, Delta and United say otherwise. Both can be right — depending on the panel.

## TL;DR

- On Day 2 of Stripe Sessions, John Collison said the K-shaped economy "is not in our data" — the high-income to low-income spending ratio in Stripe is **trending down**, not up.
- On the same morning the talk aired, the **New York Fed** published the first piece in a two-part Liberty Street series titled "Tracking the K-Shaped Economy" — concluding the K-shape is real and getting wider, with **high-income households driving most retail spending growth since 2023**.
- BofA Institute's Feb 2026 Consumer Checkpoint, JPMorganChase Institute, Delta's Q4 2025 result (premium revenue overtaking Main Cabin for the first time) and United's **+14%** Q1 2026 premium cabin growth all corroborate a widening split — on the demand side.
- Stripe's panel is a payments panel for **online merchants and SMBs**, not a household consumption panel. It under-samples luxury retail, premium air travel, financial advisory, prestige autos. Its narrowing-gap finding is real *for what it measures* — and that is not the same basket as BofA or the Fed.
- Where Stripe **does** have the better instrument is the supply side: **+130% Y/Y** in Atlas incorporations, the class of 2026 tracking 5x the revenue of 2025, France at a record **1.166M** business births. The K-shape is the wrong frame for that data — and that is the actual signal in the talk.

![The K-shape consensus across NY Fed, BofA, and JPM Chase — the diverging arms Stripe says they don't see.](/post-images/2026-05-02-stripe-vs-k-shape-economy/hero-k-shape-split.jpg)

The most contrarian claim at Stripe Sessions 2026 was not about agents, or stablecoins, or the developer keynote on Day 1. It was a single line buried in the Day 2 morning keynote, delivered by **John Collison** standing in front of a slide showing the high-income-to-low-income spending ratio inside Stripe's payment panel.

> "This stuff gets a lot of retweets, but we don't actually see it in Stripe's data. So if we look at the ratio of high-income to low-income consumer spending in our data, you know, in a K-shaped economy, you'd expect this chart to be going up, but it's the opposite. The line is trending gradually downwards. The gap between high-income and low-income spending has been shrinking."

Twelve hours after the talk went up on YouTube, the **Federal Reserve Bank of New York** published the first piece in a two-part Liberty Street Economics series titled, with no apparent irony, ["Tracking the K-Shaped Economy: Who's Driving Spending?"](https://libertystreeteconomics.newyorkfed.org/2026/05/tracking-the-k-shaped-economy-whos-driving-spending/) The conclusion of the four-author NY Fed paper, [posted by @LibertyStEcon](https://x.com/LibertyStEcon/status/2050234749331943791) the same morning Stripe was indexing the economy: "spending growth has been faster and more consistent for high-income households" since 2023. [Axios summarised it bluntly](https://x.com/axios/status/2050247042660798837): "K-shaped economy is real, per New York Fed research."

So who is right?

This is the rare empirical disagreement worth taking seriously rather than splitting. Both sides have real data. Both sides have respectable institutional backing, Stripe processes [$1.9T](https://stripe.com/annual-updates/2025), the NY Fed has the Consumer Credit Panel. The interesting story is not that one side is lying. It is that **the two sides are measuring different baskets, on different panels, with different definitions of "consumer."** Once you decompose the disagreement, the pieces of the talk that matter, and the pieces that don't, become legible.

# What Stripe showed on stage

Collison's K-shape rebuttal sat inside a wider macro segment. The setup was a discussion of consensus narratives: a viral United 787-9 seat map, [Delta's January announcement](https://www.enginecowl.com/delta-q4-2025-premium-revenue/) that premium ticket revenue had overtaken Main Cabin for the first time in the airline's history, the general "K-shape" framing that has become the [house view at the NY Fed AMEC symposium in April 2026](http://www.newyorkfed.org/research/conference/2026/amec-symposium-on-the-k-shaped-economy) and the [March 2026 Minneapolis Fed review piece](https://www.minneapolisfed.org/article/2026/have-us-consumers-gone-k-shaped-a-review-of-the-data) by Jeff Horwich.

Collison's claim: that whole story does not show up in Stripe's panel. The high-income to low-income consumer spending ratio inside Stripe's data is trending *down*, not up. **Emily Glassberg Sands**, Stripe's Head of Data and AI and a Harvard PhD economist, then took the stage to extend the framing: AI is showing up in the macro data, you just have to know where to look. Their joint conclusion was that Stripe, which processes [almost 2% of global GDP](https://stripe.com/annual-updates/2025), sees a different economy than BLS, BEA and the Fed see, because its panel is closer to the live edge of commerce than government statistics releases that lag by 30 to 90 days.

That part is defensible. The K-shape rebuttal is harder.

![Stripe's payment panel and the BofA / NY Fed consumption panel aren't the same basket; the dots tell different stories.](/post-images/2026-05-02-stripe-vs-k-shape-economy/panels-compared.jpg)

# What the rest of the data says

The K-shape consensus is not a media confection. It has converged across at least six independent panels in the last twelve months.

| Source | Panel | Headline finding |
|---|---|---|
| [NY Fed Liberty Street](https://libertystreeteconomics.newyorkfed.org/2026/05/tracking-the-k-shaped-economy-whos-driving-spending/) (May 2026) | Equifax + CCP merged spending data | High-income households drove most retail spending growth since 2023; low- and middle-income flatter. |
| [BofA Institute](https://institute.bankofamerica.com/economic-insights/consumer-checkpoint-february-2026.html) (Feb 2026) | ~50M BofA cardholders | Income-based divergence in spending widened in early 2026; lower-income spending growth roughly a third of long-run average. |
| [BofA Institute Year in Review](https://institute.bankofamerica.com/content/dam/economic-insights/2025-year-in-review.pdf) (Dec 2025) | Same | "A K-shaped recovery emerged in the second half of 2025." |
| [JPMorganChase Institute](https://www.jpmorganchase.com/institute/all-topics/financial-health-wealth-creation/real-income-sustains-weak-trend-cash-liquidity-remains-stable) (2026) | ~9M Chase households | Real income gains concentrated in top quartile; weak labor market for Gen Z and lower-income. |
| [Philadelphia Fed](https://www.philadelphiafed.org/consumer-finance/evidence-of-diverging-spending-behavior-by-income) (Nov 2025) | Consumer Finance Institute survey + card data | Divergence in spending growth by income is real and durable. |
| [Delta Q4 2025](https://www.enginecowl.com/delta-q4-2025-premium-revenue/) + [United Q1 2026](https://ir.united.com/static-files/4465ec94-3c73-45ff-841a-eac498655855) | Airline ticket revenue mix | Delta premium revenue > Main Cabin for first time ever; United Q1 2026 premium cabin **+14% YoY**. |

This is not a thinly-sourced narrative. The NY Fed's release was the result of a [two-day symposium on April 3 2026](http://www.newyorkfed.org/research/conference/2026/amec-symposium-on-the-k-shaped-economy) attended by Federal Reserve economists, academics and Treasury staff. Bank of America's [October 2025 Checkpoint](https://institute.bankofamerica.com/content/dam/economic-insights/consumer-checkpoint-october-2025.pdf) was titled, in the dry house style of the Institute, "The tale of two wallets." JPMorgan's pulse work has been hammering the same point since [the May 2025 release](https://www.jpmorganchase.com/institute/all-topics/financial-health-wealth-creation/household-finances-pulse-through-may-2025). The airlines are not running a sociology project, they are reconfiguring fleet plans around it. United's [787-9 reconfiguration](https://www.flightglobal.com/strategy/2026/03/will-uniteds-big-premium-investments-pay-off/) goes from 79 to 99 premium seats per aircraft, 46% of total seat count.

Against that wall of data, "trending gradually downwards" inside one company's payment data needs a cleaner explanation than Collison gave on stage.

![The spending-category stack: Stripe sees SaaS and DTC, the Fed sees premium air and luxury where the K-shape lives.](/post-images/2026-05-02-stripe-vs-k-shape-economy/category-stack.jpg)

# Why the Stripe panel diverges

The Minneapolis Fed's [March 2026 review](https://www.minneapolisfed.org/article/2026/have-us-consumers-gone-k-shaped-a-review-of-the-data) of K-shape evidence, written by Jeff Horwich and titled, with the unusual restraint of a regional Fed bank, "the story is not so simple", is the right starting point. Horwich's central argument is that K-shape findings depend heavily on what panel you use, what year you index to, and what spending categories you can actually see. Different panels produce different K's because they sit on different parts of the economy.

This is exactly where Stripe's data diverges from BofA's and the NY Fed's. The basket Stripe sees is not the basket the household consumption surveys see.

| Spending category | In Stripe panel? | In BofA / NY Fed / Chase panel? | K-shape signal? |
|---|---|---|---|
| SaaS subscriptions | Heavy | Light | Flat across income |
| DTC e-commerce (apparel, beauty, food) | Heavy | Heavy | Mild |
| Marketplace fees / creator economy | Heavy | Light | Skews younger / lower-income |
| Software-bought-by-software, agentic | Stripe-only | None | Not income-correlated |
| Premium air travel | Almost none | Heavy via card auth data | Strong K (Delta +>50%, United +14%) |
| Luxury retail and prestige autos | Light | Heavy | Strong K |
| Hotels, fine dining, country clubs | Partial | Heavy | Strong K |
| Wealth management and advisory fees | None | None directly | K via deposits |
| Rent, mortgages, healthcare | Almost none | Partial | Mixed |

Stripe's payment panel is dominated by *online merchant volume*: SaaS, DTC commerce, marketplaces, software-as-a-service for businesses. That basket has a structurally different income elasticity from the basket BofA and Chase see, which is dominated by total card spending across all merchant categories, including the categories where the K-shape is most visible. Premium air travel does not run on Stripe; it runs on airline IT stacks built on Sabre and Amadeus settlement rails. Luxury retail at the LVMH or Richemont end runs through bespoke merchant acquirers. Country club dues, advisory fees, prestige auto leases, these are not Stripe categories.

Meanwhile, the Stripe panel over-indexes on categories where price points are flatter across income. A $20-a-month SaaS subscription is paid in the same dollars by a graphic designer in Queens and a partner at Bain. A $79 DTC skincare order is similar across income deciles. The marketplace and creator-economy categories that Stripe is heavy in, Substacks, Gumroad, Shopify, actually skew toward middle-income consumers spending on middle-income makers, by construction.

This is not a flaw in Stripe's data. It is what Stripe's data is. But it is a serious problem for the claim "the K-shape is not in our data." The K-shape is, by definition, most visible in the categories Stripe processes least.

A second issue: **Stripe's "high-income vs low-income consumer" buckets are inferred, not reported.** Stripe does not see W-2 income. It sees BIN data, billing addresses, average ticket size, merchant mix. Whatever proxy is used to bucket "high-income" inside Stripe's data, it is a noisier instrument than the Chase Institute's approach (which has actual deposit and payroll data on ~9M households) or the Fed's Equifax-linked Consumer Credit Panel. None of this was unpacked on stage. ["Stripe econ needs to get it together,"](https://x.com/hypersoren/status/2050001430782616001) wrote Soren Larson, the technologist behind one of the sharper public counter-threads, less politely.

# The Coasean side-argument and its strongest critic

The K-shape rebuttal was tucked inside a larger thesis Collison and Sands developed on stage and Collison reposted on X the same afternoon: that AI is reducing transaction costs *both inside and between* firms, in a way Ronald Coase would have recognised, and that Stripe is the rails for the unbundling.

> We've been thinking a lot at Stripe about the Coasean lens on AI.
>
> — [@collision](https://x.com/collision/status/2049974499278528952), May 1, 2026

732 likes, and a quote-retweet from Marc Andreessen.

The sharpest pushback came from Soren Larson:

> Put directly, this is incorrect. AI modularizes execution, which sends scarcity to context and trusted operations. Transacting externally is now — apparently counter intuitively — more expensive. Stripe benefits from a Coasean Singularity, but market incentives do not support it.
>
> — [@hypersoren](https://x.com/hypersoren/status/2050001428173746359), May 1, 2026

Larson's framing matters because it reframes Sands' on-stage argument as inverted. Her claim was that AI lowers *both* within-firm and between-firm coordination costs, with the medium-term effect that more activity moves to markets, fewer people per firm, more firms total. Larson's claim is that AI *modularizes execution*, the doing of work, and as a result, **scarcity flees to the things AI cannot do**: trusted relationships, high-entropy proprietary context, brand reputation. In that world, the price of buying execution drops, but the price of *trusting* a counterparty rises. The transaction costs Coase wrote about do not disappear; they relocate.

This is not pedantry. It is the actual debate inside the platform-vs-foundation-model wars of 2026. If Larson is right, the median enterprise sale becomes *harder* not easier, because it requires more trust to extend autonomy to an AI counterparty. That is bad news for the agent-buys-from-agent demos Sands ran on stage. Larson noted, in [a follow-up reply](https://x.com/hypersoren/status/2050035582026318163) into Andreessen's thread, that major retailers have been actively *blocking* third-party agents since late 2025, a fact Sands' demo skipped past.

The Coasean argument and the K-shape claim share a structural weakness: both extrapolate from Stripe's window onto a much wider economy where most of the surface area is invisible to Stripe.

![Stripe's supply-side instrument: Atlas +130% Y/Y, France at 1.166M business births — the buried headline of the talk.](/post-images/2026-05-02-stripe-vs-k-shape-economy/supply-side-bloom.jpg)

# What Stripe is genuinely right about

Now the steelman, because the harder parts of the talk are also the more interesting ones.

Where Stripe is unambiguously the better instrument is the **supply side of the economy**, firm formation, born-global startups, solopreneur scaling, the velocity of new business cohorts. None of these are what BofA, JPMorgan or the NY Fed are panel-built to see. And the supply-side numbers from the talk hold up to independent corroboration.

| Stat from the talk | Independent verification |
|---|---|
| US solopreneurs >$100K revenue ≈ **5M** | Census 2023 Nonemployer Statistics: 29.8M nonemployer firms; Stripe's $100K cohort is consistent. |
| France new business registrations record in 2025 | [INSEE Feb 2026](https://www.insee.fr/en/statistiques/8735786): **1,165,800** business births in 2025, +5%, after +6% in 2024 — a record. |
| Stripe Atlas hits **100,000** incorporations | [Patrick Collison, May 1 2026](https://x.com/patrickc/status/2050206647793144098): 100K all-time, **Q1 2026 +130% Y/Y**. |
| Atlas class of 2026 tracking 5x revenue of class of 2025 | Internal, but consistent with [Stripe Atlas 2025 review](https://stripe.com/blog/stripe-atlas-startups-in-2025-year-in-review) — Delaware C corp formations +41% YoY over six months. |
| Top 100 AI startups on Stripe sell into 55 countries in year one | Consistent with Stripe's [Indexing the AI Economy](https://stripe.com/guides/indexing-the-ai-economy) report. |
| US high-propensity business applications | [BFS / FRED](https://fred.stlouisfed.org/series/BAHBATOTALSAUS): 145,918 in Feb 2026, structurally higher than the pre-pandemic baseline of ~95K-100K. |

This is the part of the talk that lands. Stripe sees the *fact* of new business formation faster than the BLS Business Formation Statistics, in more granular form (revenue trajectories, not just applications), and across 50+ countries simultaneously. France's INSEE confirms the record. The Census high-propensity application series confirms the US elevation. Stripe's class-of-2026 cohort number is internal and unverifiable, but the direction of travel, that AI-native cohorts are scaling revenue faster than prior cohorts, is consistent with everything coming out of the Y Combinator demo days, the Stripe Atlas annual letter, and the new firm formation discussion that economists like [Walker Wright have been pulling out](https://x.com/WalkerWright86/status/2049523670716985847) of recent NBER papers.

This is the actual story in the talk. Not "the K-shape is over." Not even "Coase wins." It is: **the rate at which new firms are starting, scaling and going global has changed in 2024-2026, and the change is large enough that government statistics are still catching up.** That is a real macro signal. It is also one that Stripe's panel is uniquely well positioned to see, and that BofA, JPMorgan and the NY Fed, by panel design, cannot see at all.

![Both panels are right: Stripe's narrowing gap is real for SaaS/DTC; the K-shape is real for premium air and luxury.](/post-images/2026-05-02-stripe-vs-k-shape-economy/both-true.jpg)

# So whose data is right

The honest answer is that both panels are right about the parts of the economy each can see, and both are wrong about the parts they cannot.

The K-shape is real on the consumer demand side. The NY Fed's Liberty Street piece, BofA's monthly checkpoints, JPMorgan's pulse, and the airline cabin data are not all simultaneously misreading the data. The split is most visible exactly where Stripe is least visible: premium air travel, luxury retail, prestige auto leases, financial advisory, deposits-as-savings. Stripe's narrowing-gap finding is genuine *for the basket Stripe processes*, SaaS, DTC, marketplaces, software-bought-by-software, and that basket has structurally different income elasticity than total household consumption.

Stripe is right that something has shifted. But the shift is not the disappearance of the K-shape. The shift is on the supply side of the economy, where Stripe has the better panel and the rest of the data establishment is two years behind. The class of 2026 scaling 5x revenue of class of 2025; France at 1.166M business births; Atlas at +130% Y/Y in Q1; born-global startups selling into 55 countries in year one, these are the parts of the talk that should have been the headline. Burying them inside a contestable rebuttal of the K-shape consensus was a rhetorical choice that obscured the real story.

The frustrating thing about the talk is that the strongest data point in the room, Stripe Atlas at 100,000 incorporations, [up 2.3x in Q1 2026 alone with aggregate revenue up 5x](https://x.com/aleximm/status/2050320475310678280), was not in tension with the NY Fed at all. It is a different story about a different layer of the economy. Collision and Sands had a clean shot at "Stripe is the supply-side macro instrument the Fed doesn't have." Instead they took a shot at the demand-side consensus, with a panel that wasn't built for it, and lost a small amount of credibility on the way to a much bigger argument they could have won outright.

For readers of [Part 1](https://thedeepfeed.com/posts/2026-04-30-stripe-sessions-2026-developer-guide/) of this series, this matters because it is the second time in two days at Sessions that the messaging skipped past the real story. Day 1 was an agent-payments shipping firehose dressed as a stablecoins bet. Day 2 was a supply-side dynamism story dressed as a contrarian macro take. The product is much better than the framing in both cases. If the next Stripe annual letter leads with "Stripe is the supply-side index of the global economy", the [Sequoia services-as-software thesis](https://thedeepfeed.com/posts/2026-04-30-sequoia-services-as-software-thesis/) recast through 5 million SMBs and 100,000 Atlas cohorts, that letter will age better than the one that argued the K-shape is closing.

The K-shape is not closing. The economy under it is being rewired. Those are different sentences. The Day 2 keynote ran them together, and the data does not.

## Sources

- [Stripe Sessions 2026 — Indexing the economy (YouTube, 30 min)](https://youtu.be/-vRY2dtD7iQ)
- [Stripe — 2025 annual letter ($1.9T TPV, 1.6% of global GDP)](https://stripe.com/annual-updates/2025)
- [Stripe Atlas startups in 2025: Year in review](https://stripe.com/blog/stripe-atlas-startups-in-2025-year-in-review)
- [NY Fed — Tracking the K-Shaped Economy: Who's Driving Spending? (Liberty Street, May 1 2026)](https://libertystreeteconomics.newyorkfed.org/2026/05/tracking-the-k-shaped-economy-whos-driving-spending/)
- [NY Fed — Explaining the K-Shaped Economy: What's Behind the Divide? (Liberty Street, May 2026)](https://libertystreeteconomics.newyorkfed.org/)
- [NY Fed — Media advisory on K-shape research series (Apr 23 2026)](https://www.newyorkfed.org/newsevents/mediaadvisory/2026/0423-2026)
- [NY Fed AMEC Symposium on the K-Shaped Economy (Apr 3 2026)](http://www.newyorkfed.org/research/conference/2026/amec-symposium-on-the-k-shaped-economy)
- [Minneapolis Fed — Have U.S. consumers gone "K-shaped"? A review of the data (Mar 20 2026)](https://www.minneapolisfed.org/article/2026/have-us-consumers-gone-k-shaped-a-review-of-the-data)
- [Philadelphia Fed — Evidence of Diverging Spending Behavior by Income (Nov 2025)](https://www.philadelphiafed.org/consumer-finance/evidence-of-diverging-spending-behavior-by-income)
- [Bank of America Institute — Consumer Checkpoint: Weathering the storm (Feb 2026)](https://institute.bankofamerica.com/economic-insights/consumer-checkpoint-february-2026.html)
- [Bank of America Institute — Consumer Checkpoint: The tale of two wallets (Oct 2025)](https://institute.bankofamerica.com/content/dam/economic-insights/consumer-checkpoint-october-2025.pdf)
- [Bank of America Institute — 2025 Year in Review (Dec 22 2025)](https://institute.bankofamerica.com/content/dam/economic-insights/2025-year-in-review.pdf)
- [JPMorganChase Institute — Real income sustains weak trend (2026)](https://www.jpmorganchase.com/institute/all-topics/financial-health-wealth-creation/real-income-sustains-weak-trend-cash-liquidity-remains-stable)
- [Delta Air Lines — premium ticket revenue overtakes Main Cabin for first time (Q4 2025 results, Jan 2026)](https://www.enginecowl.com/delta-q4-2025-premium-revenue/)
- [United Airlines — Q1 2026 earnings: premium cabin revenue +14% YoY](https://ir.united.com/static-files/4465ec94-3c73-45ff-841a-eac498655855)
- [INSEE — Business births in 2025: France 1,165,800 new businesses (+5%)](https://www.insee.fr/en/statistiques/8735786)
- [US Census BFS — High-Propensity Business Applications (FRED series, Mar 2026)](https://fred.stlouisfed.org/series/BAHBATOTALSAUS)
- [@collision — "Coasean lens on AI" thread (732 likes)](https://x.com/collision/status/2049974499278528952)
- [@hypersoren — Coasean Singularity counter-thread (43 likes)](https://x.com/hypersoren/status/2050001428173746359)
- [@hypersoren — "Stripe econ needs to get it together"](https://x.com/hypersoren/status/2050001430782616001)
- [@patrickc — Stripe Atlas hits 100,000 incorporations (933 likes)](https://x.com/patrickc/status/2050206647793144098)
- [@axios — "K-shaped economy is real, per New York Fed research" (May 1 2026)](https://x.com/axios/status/2050247042660798837)
- [@RecessionAlert — Q1 2026 GDP, AI capex doing the heavy lifting](https://x.com/RecessionAlert/status/2049904324692595019)
- [@LibertyStEcon — Tracking the K-Shaped Economy announcement](https://x.com/LibertyStEcon/status/2050234749331943791)
- [TDF — Stripe Sessions 2026 developer guide (Part 1)](https://thedeepfeed.com/posts/2026-04-30-stripe-sessions-2026-developer-guide/)
- [TDF — Sequoia services-as-software thesis](https://thedeepfeed.com/posts/2026-04-30-sequoia-services-as-software-thesis/)

---

Canonical: https://www.thedeepfeed.ai/posts/2026-05-02-stripe-vs-k-shape-economy/
Site: https://www.thedeepfeed.ai
Full corpus: https://www.thedeepfeed.ai/llms-full.txt