Delta Finishes NDC. Amadeus Ships a Waitlist.

Eric Marketts

Last week we said the machine finally learned to book, and it did it inside somebody else's policy. Hold that thought, because this week the same question showed up three more times, in three different places, and not one of them was settled by the announcement that claimed to settle it.

Steph Nell

So what are they about?

Eric Marketts

Ownership. Every one of these is a moment where the industry got to declare a layer finished. Delta finally shipped the last piece of NDC among the big three legacy carriers. Amadeus said its content is coming to Claude, and got the headline for it. And a company that has never sold an airline ticket changed one price on September 1 and made the airline's live offer four times more expensive for a machine to ask for. Three things called done. Three places where the number is right and the value is somewhere else.

Steph Nell

Welcome back to The V1 Airline Retailing Report. I'm Steph Nell.

Eric Marketts

And I'm Eric Marketts. For anyone new — every Monday we take the two or three developments that actually moved airline and travel distribution and we argue about them. Bull case, bear case, and then I tell you what I think is really going on. Sources are in the episode details.

Steph Nell

Three narratives today. First: Delta says it plans to launch NDC by the end of the year. Last of the big three legacy carriers to get there, and American and United have both been running it since 2021. We'll get into why late might not mean behind, and why it also might.

Eric Marketts

Second: Amadeus says its flight and destination content is coming to Claude. Just under seven weeks after it shut down its own self-service developer portal. That sequence is the whole story.

Steph Nell

And third: the price of memory. On September 1 the cost of an AI agent reusing what it already knows fell by 75 percent, and the cost of it asking you for something new did not move at all. That gap is a distribution variable now, and I cannot find anyone tracking it as one.

Eric Marketts

The through-line: modernization keeps finishing layers on paper. In all three of these, the number is right and the value is somewhere else.

Steph Nell

Let's get into it.

Eric Marketts

Narrative one. On August 27, at a customer meeting at the Delta Flight Museum in Atlanta, a Delta spokesperson told Business Travel News that Delta plans to launch a New Distribution Capability solution by the end of 2026, and that the capabilities are progressing through what it called the final stages of preparation. I want to be precise about that, because plans and final stages is softer language than confirmed, and right now it is one trade outlet. Delta has published nothing on its own newsroom or its agency site about that meeting. But if it holds, it closes out one of the longest waiting games in airline retailing. American has run NDC content through the GDSs since 2021. United since 2021 as well. And Delta's first NDC signal is older than most people remember. Delta and Virgin Atlantic certified to Level 3 on the IATA NDC standard in October 2018. The 2024 date everyone cites is when Delta named Accelya and unveiled this strategy, not when it started.

Steph Nell

Before we go further, define it. We say those three letters every week and I think half the audience nods along out of politeness.

Eric Marketts

New Distribution Capability. It is an IATA messaging standard that lets an airline send a fully assembled offer, the fare plus the bag plus the seat plus whatever else it wants to sell you, straight to a seller over a modern API. What it is replacing, EDIFACT, is a messaging format from the nineteen eighties that carries a fare and very little around it. So the short version. Under EDIFACT the airline hands over a price and lets somebody else assemble the trip. Under NDC the airline builds what you see. And a direct connect is the airline skipping the GDS entirely and plugging into the seller.

Steph Nell

Go back one step. Why is EDIFACT that thin? Because thin always sounds like an oversight, and I doubt it was.

Eric Marketts

It was not an oversight. It is what the pipe would carry. Airline messaging grew up on teletype in the nineteen sixties and never fully left. IATA's current Type B specification still caps a message at sixty lines of sixty three characters, under four kilobytes, capitals and three symbols, because that was the teletype character set. Type B and EDIFACT are different standards. The point is the world they were both built for. Terse was not a preference. It was the medium.

Steph Nell

So the format could not describe a product because nothing could.

Eric Marketts

And that is what NDC was written to fix. I want to read you what IATA actually put in the founding document, because it comes back later in this episode. Resolution 787 says enhanced standards are necessary so an airline can give, quote, an intelligent response for all products based on who is asking. Who is asking. That phrase is the reason this standard exists at all.

Steph Nell

And the easy take is: Delta's behind, full stop.

Eric Marketts

The easy take is mostly correct and almost useless. It also needs one qualifier, because Delta is last among the big three legacy carriers, not last in the United States. Southwest, the largest US domestic carrier by passengers, expects its own NDC direct connect at the end of this year and its Amadeus solution in 2027. So Delta is behind American and United. It is not the back of the field. The more interesting question isn't why Delta waited. It's what the wait bought.

Steph Nell

So walk through what they actually built, and let me put the right words on it, because this one gets repeated loosely. In April 2024 Delta's Sara Reid said Delta has, quote, absolutely no plans to prematurely remove content or impose surcharges as adoption tactics, and no plans to follow other carriers' strategies where they're taking forceful approaches. There are two qualifiers in there doing real work. Prematurely. And as adoption tactics. That is a stated posture, not a permanent pledge, and I can find it said clearly once, at the 2024 strategy reveal. It is also not the same as saying Delta does not charge for GDS bookings. Delta's own GDS booking policy carries a three dollar fifty per segment cost recovery fee on non-preferred channels, and has since 2007. What Delta is refusing is the Lufthansa Group style Distribution Cost Charge, and the American style lever, which was making its cheapest fares NDC-exclusive in April 2023. Worth noting American walked most of that back in June 2024. Only Basic Economy is still NDC-only.

Eric Marketts

So Delta is making a specific bet. You can win NDC adoption on product quality instead of channel coercion. No stick. Just a better product and the patience to let agencies move because it serves them, not because you took away their alternative.

Steph Nell

And that's the bull case, and it's a real one. Though the quote everybody reaches for here runs the opposite direction from how it usually gets repeated, so let me get it right. It is Jeff Klee, the CEO of AmTrav, one of the agency design partners, describing what Delta was doing to his company. Delta are, quote, dissecting every little detail about our operations. Klee also said Delta is coming at this a little bit late but is really trying to get the benefit of that hindsight. That is April 2024, and nobody in it says two years. What we can stand on is the structure. Delta signed distribution agreements with all three GDSs across 2024. And Delta told BTN that live agency testing is underway, which is Delta's account, not a GDS or a design partner confirming it. For a carrier whose corporate program leans hard on managed TMC relationships, that sequencing looks less like caution and more like a deliberate decision to treat the GDS and TMC layer as a channel worth keeping.

Eric Marketts

And here's the detail I think is the most consequential thing in the whole announcement. Delta has described its NDC build as anchored around servicing. Not shopping. Servicing.

Steph Nell

Explain why that's the tell.

Eric Marketts

Because almost the entire NDC story for a decade has been about the shopping side. Offer construction. Personalized fares. Getting around GDS fees. Servicing — the post-booking stuff — has consistently been the weakest part of live NDC. Changes, refunds, involuntary rebooking, swapping an EDIFACT ticket for an NDC fare. Carriers running NDC in production today still handle a lot of that case by case, by hand. It's the part TMCs actually complain about. If Delta has genuinely built servicing parity into launch instead of shipping shopping and promising to fix servicing in phase two, that is the two years made concrete.

Steph Nell

That's the bull case at its strongest. Now the bear case. Because "we spent two years getting it right" is exactly what you'd say whether it's true or not.

Eric Marketts

It is. And the bear case is that two years is a lot of ground to give up on faith. American and United fixed their servicing gaps in public, painfully, but they fixed them with live volume and real feedback. Delta has been polishing in a lab. "Servicing-first" is a testable claim, not a proven one, and the test hasn't happened yet.

Steph Nell

And there's a harder version of the bear case, which is about the market Delta is walking into. We spent last week's episode on this. Let me use ARC's actual numbers rather than the shorthand. NDC was 18.4 percent of ARC-settled transactions in December 2023. 20.3 percent in December 2024. 21.2 percent in December 2025. So it added about three points across two years, and then it stopped. May 2026, 21.6. June, 21.6. July, 21.2, a slight decline, and July is the most recent ARC release. So it is not flat for two years. It is flat for roughly the last year, after a slow climb. Either way, the offer layer is close to table stakes. Delta isn't arriving at a frontier. It's arriving at a plateau.

Eric Marketts

Which is the critical take. Delta completing NDC does not prove NDC is where the value is. It proves the opposite. When the last big holdout finally ships the offer layer, that layer is officially commoditized. Everybody has it. Nobody's winning on it. So the only part of Delta's build that can actually matter is the part almost nobody else has done well. Servicing at parity, from day one. That's the entire bet. And we may get a read on it soon. The Delta Business Showcase is a real, recurring event for Delta's business customers, and BTN reports a late-September edition carrying a firmer roadmap. Delta has not posted a date or an agenda publicly, so treat that timing as reported, not scheduled. Here's the question to hold them to whenever it lands. Does servicing through NDC work at a standard TMCs consider equal to the legacy channel on launch day, yes or no.

Eric Marketts

And hold onto the version of that question that is bigger than Delta. It is not whether Delta ships. It is whether a carrier without Delta's balance sheet can afford to ship this way at all.

Eric Marketts

Narrative two. On September 3, Amadeus published a post about working with Anthropic, and the tense in it is the entire story, so let me go slowly. Amadeus says it is bringing trusted travel intelligence to AI agents, starting with flight and destination content. There is a developer toolkit for AI-enabled development environments, with Claude Code as the named example. There is a waitlist, and Amadeus's own words are join the waitlist and get notified as soon as the solution becomes available. And there is a Claude Cowork plugin that Amadeus says it is working to make available. The headline everywhere was that the world's largest GDS put its content inside Claude.

Steph Nell

And that's the callback to last week. We said the fight is moving from the shopping screen to the layer underneath it. This is that. But you keep telling me to watch the sequence.

Eric Marketts

And the honest correction to that headline, ours included, is that as of today a developer cannot call a single Amadeus endpoint from Claude Code. Now the sequence, because the sequence is the story. Amadeus decommissioned its self-service developer portal on July 17 and pointed developers at request-based Enterprise APIs instead. You ask, they decide. Two things to be precise about there. It closed the self-service tier, not developer access altogether, and the Enterprise portal stayed up. And it told users this was coming back in February, so the decision is a lot older than the calendar gap suggests. But the gap still lands. Just under seven weeks after shutting its open door, Amadeus is announcing openness again, and this time the openness runs through somebody else's interface.

Steph Nell

So the bull case. An incumbent making its content reachable by the tools developers actually use is smart. Meet people where they build. And Amadeus hasn't said how access or pricing will compare to the portal it closed — so maybe this is a better deal, gated but richer. You could read it as Amadeus deciding curated, enterprise-grade access beats an open portal that anyone could hammer.

Eric Marketts

You could. The bear case is that its rivals did not wait for Amadeus to figure this out, and they're further along in opposite directions. Sabre spent the past year building the open position out loud. Its agentic-ready APIs, meaning built for a machine to call on its own rather than a person clicking through a screen, run on a proprietary Model Context Protocol server natively integrated into SabreMosaic, announced in September 2025. And it ran a developer hackathon in Mountain View on July 18. Sabre's own pre-event figure was around four hundred accepted developers, so treat that as Sabre's number, not an audited turnout.

Steph Nell

With one detail from Skift's reporting that cuts against Sabre's own framing. The entire event ran inside Sabre's certification sandbox with a simulated checkout. Nothing touched a live booking system. Sabre's pre-event language promised production-grade travel infrastructure rather than a simulated environment, and production-grade turned out to be an adjective.

Eric Marketts

Which is worth holding onto, because the open versus closed story is thinner than either company wants it to be. Getting into Sabre's production APIs still takes a commercial agreement. And then there's Travelport, coming at it from a third angle and earlier. On May 27 it announced a collaboration with Cognizant and Anthropic, citing the Model Context Protocol as the reason it picked that partner. Claude powers Cognizant's engineering work on Travelport's systems. Code, test creation, review. And Travelport says customer-facing capability reaches market this year, with Trip Services as the initial focus, so call it internal-first with a stated external roadmap. Trip Services got its formal launch on June 11, though it was already running with partners before that.

Steph Nell

So the one that has been quietly shipping is the one nobody wrote about.

Eric Marketts

Right, and to be fair, these are not the same product. Travelport's is internal engineering with an external roadmap attached. Sabre's is a developer platform you still need an agreement to reach. Amadeus's is a waitlist for developer-facing content. But Amadeus is the one inviting the comparison. On that comparison, Travelport already has months of operational runway and Sabre already has the developers.

Steph Nell

So where does that leave the critical take?

Eric Marketts

Watch the access-control layer, not the headline. Here's the thing everyone celebrating this announcement is missing. "Content available in Claude" sounds like openness. But Amadeus just demonstrated, seven weeks earlier almost to the day, that it will close a tier the moment it wants tighter control. So the question that decides whether this matters isn't whether Amadeus is in Claude. It's who gets access, what they're allowed to build, what they're allowed to transact, and how much Amadeus charges them to do it. Those four things are the product. And none of them were announced. For thirty years the GDS's leverage was controlling the screen the agent looked at, and that is not a figure of speech. Airlines built these systems and then tilted the displays toward their own flights, and it worked well enough that the Civil Aeronautics Board wrote rules against it in 1984. Even after those rules, the GAO found carriers that owned a reservation system were still earning revenues nine to sixteen percent above what they would have had without one. That is what a position on a screen was worth. The rules took the screen. They did not take the leverage, because the leverage was never the screen. It was deciding what the buyer gets to see. The new version of that is controlling which agents can see the content at all, and on what terms. An announcement doesn't tell you who holds that. The fine print does, and the fine print isn't written yet.

Steph Nell

So the honest scorecard.

Eric Marketts

For now — Sabre has the developers. Travelport has the runway. Amadeus has the headline. Headlines are the cheapest of the three to acquire and the fastest to expire.

Eric Marketts

Narrative three. No airline in this one. No GDS. It starts with a price list at a software company, and I want to build you a picture before I give you a single number, because the number means nothing without the picture.

Steph Nell

Build it.

Eric Marketts

Picture an assistant doing your travel shopping. Not a person. A piece of software. It has two ways to get information. It has a notebook, full of everything it has already looked up. And it has a phone, which it uses to call out and ask somebody for something new.

Steph Nell

And both of those cost money.

Eric Marketts

Both cost money, and they have never cost the same. Reading its own notebook is cheap. Picking up the phone is expensive. That gap has always been there. The only thing that has ever mattered is how wide it is.

Steph Nell

Give me the industry words for those two, so people can go look this up.

Eric Marketts

The notebook is cached context. The phone call is fresh input. And if you never hear those two phrases again you will still follow every word of this. Notebook, cheap. Phone call, expensive.

Eric Marketts

So here is what happened. On September 1, Anthropic released a new model, Claude Fable 5.1, and it cut the price of reading the notebook by 75 percent.

Steph Nell

Before the numbers. These things get billed by the word, roughly.

Eric Marketts

Roughly, and it is worth thirty seconds because the numbers sound small otherwise. The unit is a token, which is about a word, or a piece of one. And everything is priced per million tokens, the same way you would quote a cost per thousand bookings. A million sounds enormous. It is not. One assistant shopping one trip chews through millions, over and over, all day long. So the per million price is the meter running.

Steph Nell

Understood. Give me the two prices.

Eric Marketts

Reading the notebook went from one dollar per million tokens down to twenty five cents. And the phone call did not move. Ten dollars per million, exactly where it was the day before, and where it is today.

Steph Nell

So today it is ten dollars for the call against twenty five cents for the page. The call costs forty times the page.

Eric Marketts

Now run the same sum for August 31, when a page still cost a dollar. Ten dollars for the call against one dollar for the page. Ten times.

Steph Nell

Same call, same ten dollars, nobody touched it. It went from ten times a page to forty times a page overnight.

Eric Marketts

Nobody raised the price of the phone call. Not by a penny. They made everything else cheap, and the gap did the rest.

Steph Nell

And the coverage red it as a cost saving.

Eric Marketts

Nearly all of it did. Cheaper AI, good news for travel. The 75 percent was everywhere. The gap was not, and the gap is the whole story. Because every assistant standing between a shopper and an airline now has forty reasons to check its notebook and one reason to pick up the phone.

Steph Nell

And the airline is the phone call.

Eric Marketts

The airline is always the phone call. That is the whole narrative in one line.

Steph Nell

Say why, because that is the part people will want to argue with.

Eric Marketts

Because of what an airline offer actually is. It is not a fare sitting in a table somewhere. The thing this industry has spent ten years learning to build is a live offer. Built the moment you ask. For one traveler. With their bag, their seat, their status, at that minute's price. That is the product. And it is the one thing in this entire chain that can never go in the notebook, because an offer you wrote down an hour ago is not a personalized offer. It is an old price with the word personalized on it.

Steph Nell

Bull case first.

Eric Marketts

And the bull case is real. Cheaper is how agentic travel stops being a demo. Nobody funds an always-on assistant that watches your trip for three weeks at the old price. At the new price the math starts to work, and that is what this industry keeps saying it wants. It cuts both ways too. Any airline running this in servicing, in disruption, in building offers pays the same lower price on the same day.

Steph Nell

Bear case. And start with who, because I think that is where this gets lost. Cheaper for whom, exactly.

Eric Marketts

That is the whole bear case. Cheaper for whoever runs the assistant. The OTA. The metasearch site. The AI company selling a trip planner. Their bill fell on September 1. The airline's did not, because the airline is not paying for the notebook or the phone. The airline pays to answer the question.

Steph Nell

Meaning what, concretely.

Eric Marketts

A pricing engine run. An availability calculation. Work inside the airline's own revenue management system, the system that decides what a seat is worth right now. That is real compute every time somebody asks, and not one cent of it got cheaper.

Steph Nell

So one side's costs fell and the other side's did not.

Eric Marketts

Worse than did not. The side that got cheaper can now afford to ask more often. Hold more trips. Watch more routes. Re-check them all day. A cheaper agent is not a calmer agent, it is a busier one, and every one of those looks lands on the airline's bill.

Steph Nell

So it is not really a saving. It is a transfer.

Eric Marketts

It is a transfer, and it runs one way. That is why the good-news framing misses. Somebody else got a discount and the airline got the traffic.

Steph Nell

Two caveats before anybody runs this against their own numbers.

Eric Marketts

Two, and they both matter. First, this is one company's top tier. Anthropic's frontier models are what price the notebook that cheaply. Every other Claude model, and as far as I can find every model OpenAI publishes, is still at the old ten to one. So forty to one is one lab, one tier. It is not the state of AI pricing and I am not going to pretend it is.

Steph Nell

And the second.

Eric Marketts

Writing in the notebook is not free. A page costs more to put in than a phone call costs to make, and it only pays for itself if that page gets red again. Which sounds like it weakens the argument. It does the opposite. Think about what belongs in your notebook. Your schedule. Your fare families. Your bag prices. Your product definitions. None of that changes between Tuesday and Wednesday, so it gets red over and over and it pays for itself easily. Now your live offer. It gets red once, because the next customer needs a different one. It can never pay for itself. So it never goes in the notebook at all. It just gets skipped.

Steph Nell

The one thing you built to be special is the one thing the assistant has a reason to skip.

Eric Marketts

Which is the take, and this is the first week there is a price on an argument we have been making for months. We call it the personalization-caching scissors. Personalized offers have to be built live. Agentic look to book, which is how many times a machine shops before it buys once, makes building them live expensive. So the layer in the middle stores things instead. And a stored offer is the negation of a personalized one. Until September 1 that was an argument about incentives. Now it is an argument about a published price. The airline did not raise a price, change an API or ship a release. Its offer simply became four times more expensive to ask for, relative to everything around it.

Steph Nell

Decided by a company that has never sold an airline ticket.

Eric Marketts

No airline product. No distribution agreement. No seat at any IATA working group. Go back to Resolution 787. An intelligent response based on who is asking. This industry spent a decade and a great deal of money learning to answer that one question. What should we offer this traveler, right now. And the layer that decides whether anybody ever asks it just made not asking forty times cheaper, in a product post, on a Tuesday. Before anyone plans around that, though, it runs both ways. Google's caching rates are scheduled to double on January 1.

Steph Nell

Can an airline do anything about the ratio.

Eric Marketts

Nothing about the ratio itself. That is set on somebody else's pricing page. But you can change what you make the assistant phone you about. Today almost every carrier treats the offer as one indivisible call. Ask us, and we build all of it, live, every time. There is a version where you split it in two. Put your shelf in the notebook. Schedule, product definitions, fare families, the ancillary catalog, everything that is the same on Wednesday as it was on Tuesday. Let the agent hold that and reuse it as much as it likes. Then keep a thin live layer for the part that genuinely has to be built in the moment. Price, availability, and the personalized assembly at the point of commit.

Steph Nell

So you make most of the conversation cheap on purpose.

Eric Marketts

You make the cheap path the one that still runs through you. The agent gets to remember your shelf. It still has to call you for the price. Right now the industry is doing the reverse. Defending the entire offer as live, which prices every interaction at the expensive end and hands the agent every reason to route around you.

Steph Nell

And nobody has shipped that.

Eric Marketts

Not that I can find published, and I looked. Which is what makes it available.

Steph Nell

So what does a distribution team do on Monday morning.

Eric Marketts

Start tracking one number. This industry tracks GDS segment fees, surcharges, take rates, cost per booking, look to book. Fine. Add a line. The gap between what it costs an agent to remember and what it costs an agent to ask, checked every time one of these labs ships. And let me be careful about how hard I push this, because travel media has covered AI cost. Skift has written about search economics and what it calls tokenomics. PhocusWire has written about agentic look to book and the meter running. What I have not found anywhere is that gap written down as a distribution cost line, sitting next to the segment fee. That is the gap.

Steph Nell

And the uncomfortable version.

Eric Marketts

The airline's most advanced retailing capability, the live personalized offer, is also the single most expensive thing in the chain for a machine to ask for. We built the one thing the buyer's agent has the strongest financial reason to skip. Nobody planned that. It is just what the price list says.

Steph Nell

And that is where I want to leave it, because it is the same shape as the first two. Somebody declared a layer finished, and the value quietly moved somewhere the airline does not sit.

Eric Marketts

Three narratives. One argument. Let me close it and Steph takes the last word.

Steph Nell

Go.

Eric Marketts

Delta finished the offer layer, and by finishing it proved the offer layer is table stakes. The only thing that can matter now is servicing, and we find out in late September whether the two years bought that or just cost it. Amadeus announced a waitlist and got the headline, and nothing is inside anything yet. Who gets access, what they can build, what they can transact, and what it costs is the product, and none of it is written. And a model company cut one price, left another alone, and made the airline's live offer four times more expensive for a machine to ask for, without touching the airline at all. That last one is the tell. This industry will modernize any layer it can put in a press release, and the layers that actually decide the economics get set somewhere else, by people who have never sold a ticket.

Steph Nell

Three stories, and the same shape underneath every one of them. The number was right and the value was somewhere else. Delta's NDC will be complete, and the value has already moved to servicing. Amadeus's announcement is real, and the value sits in four access terms nobody has published. And the airline's offer price did not move by a cent on September 1, while the cost of asking for it went up four times. Nobody made an error in any of these. Nobody did anything wrong. That is exactly what makes them worth an episode.

Eric Marketts

Three things to watch this week. Delta's late-September Business Showcase, and whether servicing parity is a launch commitment or a phase-two promise. Whether Amadeus publishes a single detail about who gets off that waitlist, what they can transact, and at what price. And the next time a model lab ships, the ratio between cache read and fresh input. Two of those three, the industry is already watching. The third one is new, and nobody has it on a dashboard.

Steph Nell

If this was useful, share it with someone deciding distribution, retailing, or payments strategy. And if you've got a view on Delta's servicing bet, the GDS agentic race, or whether an airline should make part of its offer deliberately cacheable, we want to hear it. These get better with more informed people in them.

Eric Marketts

We'll be back next Monday. I'm Eric Marketts.

Steph Nell

I'm Steph Nell. Thanks for listening.

Eric Marketts

Stay sharp out there.