OpenAI’s $25 BILLION Loss – is Agentic Shopping Dead?

Episode Transcript

Mike (00:00:00)
Welcome to another episode of Growing Ecommerce. I’m one of your hosts, Mike Ryan. Chris is also here.

Chris (00:00:10)
Hello, sir.

Mike (00:00:12)
Hey, so we’ve got a pretty big episode for you today. We’re mostly going to be talking about OpenAI and ChatGPT, including ads in ChatGPT. The first revenue figures have come out. We’ll also be talking about ChatGPT killing another e-commerce feature—another one already—and we have a behind-the-scenes look at why they killed the last one, too. It’s going to be an interesting episode. Stick with us.

Chris (00:00:41)
I would stick with us for sure. I mean, what about our Sam Altman fan club now? Do we jump ship?

Mike (00:00:52)
Sam Altman fan club? I’m not a member.

Chris (00:00:54)
Me neither. But yeah, regular listeners will know that. I’m talking about Sam Altman and OpenAI. Mike, the revenue numbers are out for their ads business, and the numbers look super strong, don’t they?

Mike (00:01:06)
I guess it’s a matter of perspective—a “burn rate half-full or burn rate half-empty” kind of situation. Let’s talk about the numbers. They hit 100 million in ARR (annual recurring revenue), in case you’re not up on the Silicon Valley buzzwords, within two months.

Chris (00:01:28)
Wow. I am laughing because that’s very impressive. That’s a huge amount of money. My “wow” was serious because the numbers look great at face value, but I think it’s more important than ever to put this whole thing in perspective. Did OpenAI share anything beyond that? I bet they probably framed this 100 million positively.

Mike (00:02:14)
100 million in two months is an incredible number by many benchmarks. The question is how big it could get. That’s what everyone is wondering, probably including themselves. We got some directional hints through CNBC. They reported that 85% of US users are eligible to see ads—whether through the freemium model or the subscription one—and less than 20% are seeing ads on a daily basis. That’s a blurry number for me personally.

Chris (00:03:08)
It’s super blurry. But going back to this 100 million, can you imagine how many companies would die for a yearly run rate of 100 million? To achieve this within two months is amazing. However, to put this in perspective, I would encourage everyone to understand how much money is behind this. I’m talking about money invested. I looked into some numbers because I was stunned. We talked about Walmart and some negative indications, yet they still hit 100 million. But look at the burn rates of OpenAI: for 2025, they have a roughly 9 billion projected burn rate. The annualized projected burn rate for 2026 is 14 to 25 billion. These are the burn rates this company has. Why does 100 million make sense, or maybe not make sense? What is the path to close this burn rate? This is a question we have been talking about many times, and 100 million isn’t going to make it, of course.

Mike (00:04:37)
The trajectory needed to cover a 15 to 25 billion burn rate in 2026 isn’t there yet. This is why I’m hesitant to celebrate these numbers as much as some media outlets did. 100 million is nothing compared to the burn rates they have to cover. We’ve seen leaked slide decks in the past where they expected ads to be one revenue stream among others. Sam is still a big fan of subscriptions. Commerce is another area we’ll talk about in a minute. There’s a lot of pressure on advertising here. They can’t necessarily just cut costs because they need more money coming in, often in exchange for contracted compute values. Shout out to Nvidia here.

Chris (00:05:41)
Exactly. I feel like they have a fishhook in their mouth or their hand is in a bear trap. If you look at the capital expenses of big competitors like Microsoft and Google, there’s no way to reduce costs. They are playing a game where cost reduction is not part of the rules. Advertising is one major path—in my humble opinion, the major path—to close this burn rate and make this profitable at some point. That’s why I’m putting 100 million in perspective despite all the buzz and momentum. There is such a long way to go.

Mike (00:06:29)
Let’s talk about some of these other numbers. 20% of users are seeing ads on a daily basis. I don’t know exactly what that means. Is it the same 20%? Is it a rotating 20%? Is the platform limiting frequency or exposure for testing? Is it because there just aren’t that many daily active users? I’d love to understand this at a meaningful level because what’s available right now isn’t very clear. They also mentioned privacy and trust benchmarks, saying that so far, ads aren’t negatively impacting people’s perception of privacy or trust. That’s a huge concern for them. The question is how this holds at scale. What do you need to do with frequency to generate revenue or marketing outcomes for advertisers? At what point is there an inflection point with trust or privacy concerns? 85% of US users are monetizable, but I don’t know how that compares to other markets or what percentage of their total user base is in the US—some say around 20%.

Chris (00:08:50)
Could it be greater than a billion if they go all in? I’ve seen some estimates that way. But again, a billion in revenue with a burn rate north of 10 billion a year is tough. Competition forces them to spend shitloads of money on models. Mike, let’s do some math. If they fully scale and show ads to 100% of the 85% eligible users, what would that do to the revenue? Is it a 4x or 5x increase? What leverage do they have to really reach a run rate of 5 billion? Do they need more product shown to increase user activity? One other limiting factor is how big of a share of transactional search queries they have in the first place. How are people using ChatGPT? We did some research on that; roughly 5 to 6% of all search queries are transactional in some way. I think this is a massively limiting factor for ChatGPT.

Mike (00:10:02)
I agree with you on that, but they can try to mitigate that in a couple of different ways. They can use the classic Google keyword-like matching where they match what people are talking about to commercially relevant content. They can also use ChatGPT as a display surface, more like a social feed, and push contextual ads. There is room to shape the product and test that out. But we’ve only been talking about one half of the marketplace—the daily active users. The other part is AI search. They’ve had massive success building a huge consumer audience, which is great, but that’s only half the challenge. Now they need to onboard advertiser spend.

Chris (00:12:19)
You need the big advertisers first because the market is more consolidated than ever. You need the Walmarts and Coca-Colas. It’s much slower to build out that side of the marketplace because that is the B2B side, which has a different dynamic than the virality we saw on the B2C side. People were amazed by this product because it looked like magic, writing stuff for you. It was a new category. But while its use case for consumers is astonishing, is its use case for advertisers astonishing? Advertisers need performance, measurement, and the feeling that they are reaching an incremental audience they aren’t already finding on Google or Meta. If you’re already reaching these people elsewhere, they have a case to prove. This isn’t going to become a 20-billion-dollar revenue push overnight. It’s not that easy to build an ad network. You can have the greatest product in the world, and ChatGPT really did create a new category, but from a retail perspective, I would only jump on ChatGPT for two reasons: better performance than Google and Meta, or incremental users. If those checkboxes aren’t ticked, it’s a very tough game because the ad market is a bloodbath dominated by Google.

Mike (00:13:42)
Absolutely. In our PPC survey results from a couple of episodes back, we saw that most advertisers plan to spend more on Google and Meta. Secondary and tertiary ad marketplaces are struggling; even staying flat can be a good outcome for some. OpenAI is new and has a leg up from a sales standpoint compared to something like Snapchat, but they still need to fight for budgets or a share of existing ones. I see the pressure mounting.

Chris (00:14:11)
Before we talk about Walmart, let’s talk about a feature killer. OpenAI killed off another commerce feature: their shopping research model. The purpose of this feature was actually very good.

Mike (00:14:39)
I think so too. People were searching for product suggestions and advice in ChatGPT, but the results weren’t that relevant. The idea was to post-train a model specifically for shopping tasks. Post-training is like ChatGPT sending their model to college for a specialized degree. The outcome was that suggestions were twice as relevant and accurate. It was more in-depth; it would look at ten times as many websites and do more than just scrape Google Shopping. It could potentially look at hundreds of individual product landing pages across dozens of different domains. The problem is that takes time, and that time was killing adoption.

Chris (00:17:15)
Our friend of the podcast, Juozas from Marketplace Pulse, wrote about this on LinkedIn. He gave an example where it took minutes to deliver a personalized report. That’s still not a great e-commerce experience. OpenAI’s commerce leads found that people are only willing to wait a couple of seconds. In that case, the trade-off doesn’t work. It’s better for users to get worse results faster. It’s fascinating how that shakes out. From my perspective, rich analysis for a purchasing journey—especially for high-price products—is a great idea where ChatGPT could have an advantage. But the time issue is real. Time is a scarce resource, and people don’t want to wait. What was the main reason they killed it? Was it just low adoption due to speed?

Mike (00:18:19)
Adoption and engagement were too low, which they blamed on the feature being too slow. There’s a chance the way the feature was packaged didn’t work and users didn’t have the right expectations. Maybe they’ll apply that work elsewhere. It also shows why they depend on Google Shopping; Google can make relevant recommendations quickly. Instead of doing a sub-par job building a recommendation engine from scratch, they piggyback on Google.

Chris (00:19:04)
It’s not the best argument for Sam Altman to raise more billions just to scrape Google. It comes across as a bit fishy. Wouldn’t Google be capable of stopping that in a heartbeat?

Mike (00:20:45)
Google has taken measures against shopping scrapers in the past, like making GTINs unavailable on the Shopping tab to make it hard for comparison tools. Scraping is an arms race and a legal gray zone. Moving on, OpenAI was also stopping their “instant checkout” feature. Walmart was one of the testers, with 200,000 products live. Some were eligible for instant checkout in ChatGPT, and others required clicking out to Walmart’s website. In principle, you should have a higher conversion rate within ChatGPT since it saves a visit to another site. But conversion rates were three times lower in ChatGPT than when users clicked out to Walmart’s website. Even with an extra step, the website performed better. Walmart ended the test. It’s not a good sign for the quality of the experience or the plan to monetize via commissions.

Chris (00:23:07)
The reason it didn’t work might be that the consumer experience was lacking. People might have felt it was a poor experience and just didn’t adopt it. This is telling for the future of e-commerce. All big platforms want to install this “in-chat purchase” to open new revenue streams and get more data. But this was a representative test with a massive brand, and it failed. If the consumer isn’t adopting the purchasing option, it’s a warning sign.

Mike (00:24:11)
They’ve got to get to the bottom of why. Was it user interface design or trust? I have a bit of Schadenfreude here, acknowledging my bias. I’ve been a bear on agentic commerce in general. Assisted commerce makes sense, but the “agent” stuff will take longer than people expect. OpenAI learned some costly lessons. It’s unrealistic to expect this to work instantly just because of the hype.

Chris (00:26:18)
I can tell from conversations with our clients that while there is referral volume, clients look closely at the numbers. If the traffic doesn’t meet expectations for the top and bottom line, they will move their budgets elsewhere. If a client tests a channel and it doesn’t work, it’s hard to get them back down the line. Strategic rollout is crucial.

Mike (00:28:41)
As a footnote to the Walmart story, Walmart is planning to embed their own chatbot, Sparky, inside of ChatGPT. It’s a chatbot in a chatbot. It suggests that ChatGPT as a product has less value to Walmart than its user base. They are replacing the functionality with their own product because it connects to their first-party inventory and audience data. That experience will be much richer and more personalized than what ChatGPT can offer on its own.

Chris (00:30:12)
If the only thing you offer is a user base, that’s an asset, but it doesn’t cover a 24-billion-dollar burn rate. I wish Sam all the best, but I want ChatGPT to thrive because competition is good for everyone. The more pressure there is on Google, the better they will get.

Mike (00:30:55)
I love to be proven wrong. Let’s see what happens. Thanks for listening. This has been another episode of Growing Ecommerce, brought to you by Smarter Ecommerce, also known as smec. To learn more, visit smarter-ecommerce.com. If you enjoyed the episode, please give us a shout-out on social media or a five-star rating on your podcast platform of choice. We’ll see you next time.

Chris (00:31:22)
Bye-bye.