AI ships bundled, infrastructure gets unbundled
In one window, nine corpus companies folded a new AI capability into existing plans at no extra charge while nine others put a price on something that had been included. The split runs by cost type, not by novelty: new AI capability is priced as retention, while storage, telephony minutes, trace egress and human hours get a meter. Anthropic added three products to a $20/mo plan in the same window Braintrust began charging $0.50/GB/mo to keep a log file.
What's happening — and why
What's happening: two opposite packaging moves ran at equal volume. Nine companies shipped new AI capability into existing tiers with no price change — Anthropic went from foregrounding Claude Code alone to bundling Claude Code, Cowork, Design and Science into unchanged Pro ($20/mo), Max (from $100/mo) and Team subscriptions; Wispr Flow shipped an entire second product, Notetaker, free on its existing ladder; FLORA put Fashion Studio on every plan including Free; Sequence shipped four agent features into a $799/mo tier that did not move; Firecrawl granted the identical +1,000 search credits a month to every plan card, Free through custom Enterprise. Nine others went the other way: Bland AI removed telephony from its all-in per-minute rate, Braintrust turned data retention from a feature gate into a $0.50/GB/mo meter, Stripe moved advanced usage billing out of its 0.7% Billing rate into a separate Metronome product, CrewAI removed 50 monthly development hours from Enterprise contracts, and Zendesk broke a uniform $50 add-on price by moving Contact Center to $83 per agent/month.
Why: the dividing line is marginal cost, and it runs backwards from intuition. These vendors are giving away the input with the highest apparent marginal cost — GPU inference — and starting to charge for storage, telephony minutes, trace egress and human hours. New AI capability defends something (a seat, a plan, a renewal), so it is priced as retention. Storage, carrier minutes and forward-deployed engineers defend nothing — they are pass-through costs that scale with usage and carry no competitive story — so they get a meter.
Why it is easy to miss: the unbundling arm can run without moving a published number. Bland AI's Start stayed free at $0.14/min, Build at $299 + $0.12/min and Scale at $499 + $0.11/min; what changed is that the per-minute rate now covers LLM, STT and TTS only, with telephony billed separately at pass-through cost. A price diff detects nothing. The two arms are also not symmetrical in visibility — bundling ships as an announced launch, unbundling frequently does not.
How it works
Evidence over time
12 supporting · 0 counter — hover or tap a point for detail, click to jump to the row.
Evidence
| Company | Date | What happened |
|---|---|---|
| Anthropic | Jul 2026 | Claude Pro ($20/mo), Max (from $100/mo) and Team ($20-$100/seat/mo) went from foregrounding Claude Code alone to bundling FOUR sibling products — Claude Code, Cowork, Design and Science — at unchanged subscription prices. |
| Wispr Flow | Aug 2026 | Shipped an entire second product, Wispr Notetaker (meeting transcription, speaker ID, cross-meeting Q&A, MCP export to Claude/ChatGPT), as a new nav item — included free on the same Free/Pro/Enterprise ladder with no new SKU and no price change. Enterprise marked "coming soon." |
| FLORA | Aug 2026 | Launched Fashion Studio and bundled it into EVERY plan including Free (partial access on Free, full on Starter/Pro/Max) with no incremental seat price; Studio generations simply draw from each plan's existing usage pool. |
| Sequence | Jul 2026 | Shipped an AR agent, agent guardrails, a Watchtower review console and Sequence MCP for Claude/Cursor/ChatGPT — none separately priced. Growth held at $799/month; Core and Scale stayed bespoke; the three paid add-ons (quote builder, reporting dashboard, rev-rec) were unchanged. |
| Diffbot | Aug 2026 | Added Web Search as a fifth product at 1 credit per query — the same rate as a page extraction — bundled into the existing "all APIs included" entitlement on every tier including Free, with $0/$299/$899/custom prices untouched. Crawl remains the only gated product. |
| Mem0 | Aug 2026 | Added a "Dream (Memory Consolidation)" capability to Pro ($249/mo) and Enterprise with no price or quota change, and added SOC 2 Type I / HIPAA Ready / GDPR Ready badges — a capability and a compliance upgrade shipped as retention, not as revenue. |
| Firecrawl | Jul 2026 | Granted a flat +1,000 search credits per month to every plan card — Free, Hobby, Standard, Growth, Scale AND custom Enterprise get the identical allotment — on top of the unchanged 2-credits-per-10-results Search rate. |
| Bland AI | Jul 2026 | Ran the opposite play on infrastructure: removed telephony from the all-in per-minute rate. The rate now covers LLM + STT + TTS only; "Telephony is billed separately, on your own carrier or Bland's at pass-through cost." No plan price moved (Start free at $0.14/min, Build $299 + $0.12/min, Scale $499 + $0.11/min) — the number bought less. |
| Stripe | Jul 2026 | Unbundled advanced usage-based billing OUT of the 0.7%-of-billing-volume Billing rate into Metronome, now a separate product card between Billing and Invoicing. The feature row changed from an "Included" checkmark to a "Contact Metronome" link; only basic metering (Meters API, up to 100M events/mo) stays inside Billing pricing. |
| Braintrust | Jul 2026 | Turned data retention from a plain feature gate (14 days Starter / 30 Pro / custom Enterprise, no rate) into a priced meter: Pro holds data past 30 days at $0.50/GB/mo, shown both in the plan card and in the renamed "Included data retention" comparison row. |
| CrewAI | Jul 2026 | Removed 50 hours of development per month from every custom-quoted Enterprise contract, replacing it with a 45-day onboarding programme; forward-deployed engineering and further training are now a la carte at extra cost. |
| Zendesk AI | Jul 2026 | Broke a uniform US$50 add-on price by moving Contact Center to US$83 per agent/month paid yearly (+66%) onto its own pricing surface, selling Minutes Blocks at US$33 and Quality Assurance at US$35 separately. Copilot and the Workforce Engagement Bundle stayed at US$50. |
Counterexamples
- Surfer SEO · — — The clean break in the boundary: on 2026-07-29 it unbundled a genuinely AI-native capability — AI-visibility/brand-mention tracking — into a standalone, slider-metered product priced by prompts tracked per day (50 = $82/mo, 100 = $158, 200 = $308 annual, unlimited sales-led), separate from its unchanged $49-$299 content ladder.
- Frase · — — Put dollar tiers on a previously free bundled perk: FraseCMS hosting went from "we host it for you, no caps disclosed" to Free (100K page views) / Basic $19 (250K) / Pro $99 (1M), billed ON TOP of any paid Starter/Professional/Scale plan (2026-08-04).
- Vercel · — — Priced net-new AI telemetry rather than bundling it: AI Gateway Trace Drains launched 2026-08-01 at $0.05 per 1,000 traces plus $0.50/GB egress, billed against general Drains usage rather than AI Gateway Credits.
Trivia
-
Anthropic added three products to a $20/month plan in the same window Braintrust started charging $0.50/GB/month to keep a log file. Claude Pro, Max and Team went from foregrounding Claude Code alone to bundling Claude Code, Cowork, Design and Science at unchanged prices on 2026-07-23; Braintrust converted data retention from a plain feature gate into a priced meter on 2026-07-22, one day earlier.
-
Bland AI unbundled without moving a single published number. Its per-minute rate now covers LLM, STT and TTS only — telephony is billed separately at pass-through cost — while Start stayed at $0.14/min, Build at $299 + $0.12/min and Scale at $499 + $0.11/min. The number did not change; what it buys did.
-
Firecrawl granted the identical +1,000 search credits per month to every plan card on 2026-07-30 — Free, Hobby, Standard, Growth, Scale and custom Enterprise all got the same allotment, so the giveaway is worth proportionally most to the free tier and least to the largest contract.
For buyers
Read a free capability addition as a renewal argument, not a discount — it is worth something, but it does not lower your bill and it does not offset an increase elsewhere. The money moved in the other direction this window, so audit scope at every renewal: is telephony still inside the per-minute rate, is retention past 30 days still free, is egress still bundled, are onboarding or development hours still in the contract? Ask what the per-unit rate covers rather than what it costs. Bland AI's published numbers were all unchanged — Start free at $0.14/min, Build $299 + $0.12/min, Scale $499 + $0.11/min — while the rate quietly stopped covering telephony. A price diff will not catch that; only an included-features comparison will.
For vendors
The bundling arm needs an entitlement model where a new product can draw on an existing pool without minting a SKU — FLORA's Fashion Studio generations come out of each plan's existing usage pool, and Diffbot's fifth product bills at the same 1 credit per query as its first. Mem0 shows the purest version of the play: a new capability plus SOC 2 / HIPAA / GDPR badges shipped together at an unchanged $249/mo, which is a retention package, not a product launch. The unbundling arm needs per-resource metering (GB-months, carrier minutes, trace volume, engineering hours) and a real communication plan, because a scope reduction is invisible on the pricing page and lands as a surprise on the invoice.
Outlook — what to watch
Expect the split to hold while inference keeps deflating: as long as model capability is the cheapest thing a vendor can add and the strongest reason to renew, it will keep arriving free. The boundary is a tendency, not a rule, and one company breaks it cleanly — Surfer SEO unbundled AI-visibility tracking, a genuinely AI-native, net-new capability, into a standalone slider-metered product at $82/mo for 50 prompts tracked per day, $158 for 100 and $308 for 200, separate from its unchanged $49–$299 content ladder. Vercel did something adjacent, pricing net-new AI telemetry (Trace Drains at $0.05 per 1,000 traces plus $0.50/GB egress) rather than bundling it. If standalone pricing on net-new agent capability becomes the majority behaviour, this trend is falsified rather than sharpened. And because the unbundling side is rarely announced while the bundling side ships as a launch, treat nine as a floor on that arm, not a count.
Bottom line
Nine corpus companies bundled new AI capability into existing plans at no extra charge in the same window nine others put a price on infrastructure that used to be included. The sort is by marginal cost and it runs backwards — inference is given away, storage and carrier minutes get a meter. Watch the scope of what you already pay for, not the headline price.
FAQ
Is a free AI feature actually a discount?
No. A capability added at no extra charge is a retention argument — it defends a seat, a plan or a renewal — and it does not lower your bill. In the same window nine corpus vendors shipped new AI capability free, nine others put a price on something previously included, so the net direction of spend was flat to up.
Why do AI vendors give away inference but charge for storage?
Because the sort is by what defends the account, not by what costs most to run. New AI capability is the strongest reason to renew, so it is priced as retention; storage, carrier minutes, trace egress and forward-deployed engineering hours defend nothing and scale with usage, so they get a meter. Anthropic added three products to a $20/mo plan in the same window Braintrust began charging $0.50/GB/mo to hold data past 30 days.
What should I re-check at renewal?
Scope, not price. Confirm in writing what your per-unit rate covers (Bland AI's per-minute rate now covers LLM, STT and TTS only — telephony is billed separately at pass-through cost), how long data is retained before a per-GB charge starts, whether advanced usage-billing features are still inside your platform rate (Stripe moved them out of its 0.7% Billing rate into Metronome), and whether onboarding or development hours are still included (CrewAI removed 50 hours a month from Enterprise contracts).
Will a pricing-page price diff catch this?
Not on the unbundling side. Bland AI's published numbers did not move — Start free at $0.14/min, Build $299 + $0.12/min, Scale $499 + $0.11/min — while what the rate buys shrank. The bundling side ships as an announced launch; the unbundling side often does not, so scope has to be tracked by comparing what is included, not what it costs.