Tech News Briefing

Ramp economist Ara Kharazian says corporate AI spend growth is slowing as price competition intensifies.

11 min Listen to the original Episode page
Corporate AI SpendingOpenAIAnthropicEnterprise TechnologyMarket CompetitionAI EconomicsCybersecurityCustomer Support

Summary

Host Julie Chang and guest Ara Kharazian discuss data from Ramp, which tracks over 85,000 businesses, showing that corporate AI spending growth has decelerated from double-digit to single-digit month-over-month increases. Kharazian attributes this to fierce price competition between OpenAI and Anthropic, which has lowered costs despite increased token usage. The conversation highlights that engineering teams drive the majority of AI usage and are best at controlling costs, while the median firm spends significantly less than the top 1% of adopters. Future growth is identified in customer support and cybersecurity, as the era of unchecked corporate AI spending appears to be ending.

Key takeaways

  1. 1
    amongst both OpenAI and Anthropic every month as businesses were ramping up their spend. And now, over the past couple months, that has started to slow down into the single digits month over month.

    Unidentified speaker (speaker_2) · 1:16

Episode sections

  1. 0:03 – 0:45

    Intro and Market Context

    Host Julie Chang introduces the episode, citing Jensen Huang's resignation at NVIDIA, Anthropic lowering its IPO target, and voter pushback against data centers as signs that the AI boom is facing a reality check. She introduces guest Ara Kharazian, lead economist at Ramp, to discuss data on corporate AI spending.

  2. 0:45 – 1:28

    AI Spend Growth Slowing

    Ara Kharazian explains that Ramp tracks over 85,000 businesses with $275 billion in annual spend. While AI was the fastest-growing category with double-digit month-over-month growth, that growth has slowed to single digits over the past couple of months.

  3. 1:28 – 2:18

    Price Competition and Usage vs Spend

    Kharazian notes that while token spending growth has slowed, actual AI usage (tokens) has increased. This is because OpenAI and Anthropic are engaging in fierce competition, lowering prices and offering discounts to gain market share, leading to less spend despite higher usage.

  4. 2:18 – 4:07

    Shift in Corporate Narrative on AI ROI

    The speaker observes a cultural shift in corporate reporting: it is no longer fashionable to claim AI budgets are being blown through, but rather to report cost-effective usage and ROI. He states the era of uncontrollable AI spending and hypergrowth for AI companies is ending.

  5. 4:07 – 7:52

    Concentration of AI Spend

    Over 70% of enterprise AI usage is by engineering teams, who are best at controlling costs by switching to cheaper models. The top 1% of firms spend ~$10,000 per employee per month on AI, driving 80% of revenue to OpenAI and Anthropic, while the median firm spends only ~$16 per employee.

  6. 7:52 – 11:03

    Future Growth Frontiers

    Kharazian identifies three areas for future growth: a three-way race between OpenAI, Anthropic, and SpaceX AI (plus Meta); increased spend in customer support (13% of classified spend); and cybersecurity, where AI will be used for both attacks and defense, driving frontier model usage.

Assets discussed

AssetKindStanceContext
Meta PlatformsMETAequityneutralAI, and possibly even Meta. So we shouldn't underestimate the level of competition in this sector and that it's going to get even

“Stance” is how the speakers talked about the asset, not a recommendation.

Claims worth verifying

Jensen Huang resigned from NVIDIA.

Unidentified speaker (UNKNOWN) ·0:09

How to verify: Check NVIDIA's official press releases or SEC filings (Form 8-K) for executive departures around the time of the podcast release (Nov 2026). For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.

Primary source

Anthropic lowered its IPO target.

Unidentified speaker (UNKNOWN) ·0:09

How to verify: Search for press releases or news reports from Anthropic or financial news outlets (WSJ, Bloomberg) regarding IPO adjustments in late 2026. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.

Primary source

Ramp covers 85,000+ businesses with $275B+ annual spend.

Unidentified speaker (speaker_2) ·0:56

How to verify: Check Ramp's official website, press releases, or white papers published between Aug 2026 and Nov 2026 detailing their dataset scope. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.

AI spend growth slowed from double-digit to single-digit month-over-month growth over the past couple of months.

Unidentified speaker (speaker_2) ·1:16

How to verify: Review Ramp's published data reports or blog posts from Sept/Oct 2026 showing month-over-month AI spend trends. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.

Over 70% of enterprise AI usage is by engineering teams.

Unidentified speaker (speaker_2) ·4:27

How to verify: Check Ramp's specific report on AI usage by department, likely published in late 2026. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.

Top 1% of firms spend ~$10,000 per employee per month on AI.

Unidentified speaker (speaker_2) ·5:17

How to verify: Verify against Ramp's data report on AI spend distribution, specifically the section on top spenders. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.

Top 10% of firms spend ~$800 per employee per month on AI.

Unidentified speaker (speaker_2) ·5:37

How to verify: Verify against Ramp's data report on AI spend distribution. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.

Median firm spends ~$16 per employee per month on AI.

Unidentified speaker (speaker_2) ·5:42

How to verify: Verify against Ramp's data report on AI spend distribution. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.

Top 1% of firms drive 80% of revenues to OpenAI and Anthropic.

Unidentified speaker (speaker_1) ·5:59

How to verify: Check Ramp's analysis of revenue concentration for AI model providers. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.

Customer support accounts for ~13% of classified AI spend.

Unidentified speaker (speaker_2) ·8:47

How to verify: Check Ramp's report on AI use cases by department/function. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.

Do your own research

  • Verify Ramp's methodology for tracking $275B in annual spend across 85,000 businesses via their public reports or whitepapers.
  • Check OpenAI and Anthropic's latest pricing pages and discount structures to confirm the claim of aggressive price competition.
  • Review engineering department budgets in target companies to assess if they represent the majority of AI spend as claimed.
  • Analyze customer support AI adoption rates in specific industries to validate the 13% spend claim.
  • Investigate cybersecurity AI vendor pipelines to assess the growth potential of AI-driven defense and attack tools.
  • Compare median corporate AI spend figures against industry benchmarks to verify the $16/employee claim.
  • Assess the concentration risk of OpenAI and Anthropic revenues by analyzing their top customer lists if available.
  • Look for counter-arguments regarding AI ROI in non-engineering departments to challenge the assertion that only engineers optimize costs.
Full transcriptexpand
0:03
Julie Chang: Welcome to Tech News Briefing. It's Monday, November 16th. I'm Julie Chang for the Wall Street Journal.
0:09
Unidentified speaker (UNKNOWN): From
0:10
Julie Chang: Jensen Huang's resignation at NVIDIA, Anthropic lowering its IPO target, and a string of voter pushback against data centers in the midterms, the AI boom is facing a hard reality check. And across Wall Street, long-held discussions of an AI bubble. are getting louder. What's
0:29
Julie Chang: news host, Luke Vargas, spoke with Ara Carazian, lead economist at Ramp, to break down what the numbers actually tell us about the state of the AI market.
0:45
Unidentified speaker (speaker_1): For businesses. So you've got insight into how companies are actually spending their money, especially on AI, and you've got some brand new data to share with us. What does it show?
0:56
Unidentified speaker (speaker_2): Over the last two years, we've been tracking AI spend in our data set. So we cover 85,000 plus businesses, 275 billion plus a year in annual spend, and the fastest growing spend category over that time has been AI. There were points at which we were observing double-digit double digit growth. month over month
1:16
Unidentified speaker (speaker_2): amongst both OpenAI and Anthropic every month as businesses were ramping up their spend. And now, over the past couple months, that has started to slow down into the single digits month over month.
1:28
Unidentified speaker (UNKNOWN): Now,
1:28
Unidentified speaker (speaker_2): so there are a couple questions that the market has to face today is how do OpenAI and Anthropic now weather this environment in which businesses are increasingly looking to cut costs on AI? And then, second, separate from the question of OpenAI and Anthropic's valuations, how do we get AI into a place where it's actually economically productive for most of these
1:48
Unidentified speaker (speaker_2): businesses? Because that's a very different question from whether or not we're in an AI bubble, is whether or not we'll actually see economic growth from all of this.
1:56
Unidentified speaker (speaker_1): Just on the topic of token spending growth, as you alluded to after a very big month for both OpenAI and Anthropic in July. your data shows that growth has been quite slow for OpenAI and even slower for Anthropic since then, which is somewhat remarkable, is it not, given that Anthropic is planning to IPO. And these are not really the numbers you'd
2:16
Unidentified speaker (speaker_1): necessarily want to see in the run-up to that.
2:18
Unidentified speaker (speaker_2): Well, the challenge is that it's not even a demand issue. You know, over the same period, we see businesses using more and more tokens, essentially using more and more. AI. The problem is that AI itself has gotten really, really cheap, in large part because of this increasing amount of competition between OpenI and Ananthropic. Both of them
2:38
Unidentified speaker (speaker_2): jogging for the top spot in terms of business spend. In order to do that, both of them training very expensive models, using very expensive compute to try to get the best one, offering that model. at a cheaper cost every month to customers in order to gain market share, simultaneously cutting
2:57
Unidentified speaker (speaker_2): prices for their standard and light models, and then also offering discounts on contracts so that businesses can keep more of their spend on OpenIn Anthropic. So you have a couple factors simultaneously moving in the same direction, leading to less spend on AI models, despite rising usage.
3:17
Unidentified speaker (speaker_1): And and it's pretty wild to see this turn of events because only this summer CEOs were saying that they were burning through their yearly token budgets with no obvious relief in sight.
3:28
Unidentified speaker (speaker_2): Well, to be clear, we actually do observe that in our data set. There were several months in our data set where we saw this rapid acceleration in AI spend. And that was kind of that token maxing area for a while. it was quite fashionable if you were a business to report to your board that we are experimenting with AI so much that we're blowing through our budgets, so that we're gonna raise those budgets.
3:48
Unidentified speaker (speaker_2): Now the tide has turned. It's actually quite fashionable to report the opposite that we are using AI in a cost effective way and putting it under greater control and ideally getting some ROI out of it. Now, whether or not businesses know how to measure ROI, that's a completely different question. We're actually not quite there yet. But the era in which it is fashionable to suggest that we are
4:07
Unidentified speaker (speaker_2): spending uncontrollable amounts on AI is over. And subsequently, the era of hypergrowth in terms of revenue for the AI companies on that business spend also seems to be coming to an end. Now that's not to say that there won't be new use cases unlocked. One of the results of our work is finding that more than 70%
4:27
Unidentified speaker (speaker_2): of enterprise AI usage is being used by engineering teams. Now that's all good and makes sense. Coding agents are the most commercially advanced productivity-enhancing form of AI. But those are also precisely the teams that have figured out how to control their AI spend most effectively by switching to cheaper models, oftentimes. Times.
4:45
Unidentified speaker (UNKNOWN): So
4:46
Unidentified speaker (speaker_2): whether or not the AI companies will be able to continue to grow their revenues and whether or not AI will continue. to drive increasing ROI for firms, it's going to depend on whether or not the model companies are able to make their inroads outside of the engineering departments. Yeah.
5:00
Unidentified speaker (speaker_1): Alright, give us a bit more detail on spending per employee because this is quite helpful in terms of understanding just how many businesses out there are really in a position to keep the AI trade as frothy as it's been with their enterprise licenses and things like that. I mean, hit hint, it's not a ton of businesses.
5:17
Unidentified speaker (speaker_2): Well, this is one of the main features of AI spend by businesses is that it is highly concentrated in specific sectors and amongst highly intensive businesses within those sectors. So the top 1% of firms spend about $10,000 per employee per month on AI. For the top 10% of firms, it's $800
5:37
Unidentified speaker (speaker_2): per employee per month. But the median firm that spends on AI spends only about $16.
5:42
Unidentified speaker (speaker_2): So that's about a chat subscription or two per employee per month. But if you're thinking about where the money's gonna come from for open and anthropic, you know, most of it is coming from that top one percent group, which spends so much that they drive 80% of the revenues toward open eye I and anthropic.
5:59
Unidentified speaker (speaker_1): I wonder how much that leaves for other companies, because just walking around at a Web Summit this week. there are a ton of businesses pitching this and that different AI solution, trying to ride on the coattails of the biggest AI labs. But also, is there even enough here for anthropic and OpenAI to reach their own super lofty hope-for valuations if it really is just America's biggest businesses opening
6:19
Unidentified speaker (speaker_1): up their checkbooks for them?
6:20
Unidentified speaker (speaker_2): Well, much of the AI trade over the past year or so has focused on that top one percent. 1% figure because it's kind of this sign of, hey, who knows where this is gonna land?
6:31
Unidentified speaker (UNKNOWN): You know,
6:32
Unidentified speaker (speaker_2): as the top 1% continues to spend increasing amounts on AI,
6:36
Unidentified speaker (UNKNOWN): you
6:36
Unidentified speaker (speaker_2): know, imagine where the median is gonna end up when it gets somewhere close to that. And most companies are spending thousands of dollars a month.
6:42
Unidentified speaker (UNKNOWN): I
6:43
Unidentified speaker (speaker_2): think the problem is that those assumptions are now breaking because the top 1% of firms are starting to slow their spend. Growth has slowed for them. If anything, over the past couple months, it's kind of stuttered up and down.
6:54
Unidentified speaker (UNKNOWN): And
6:54
Unidentified speaker (speaker_2): now there are new questions about whether or not the median or top 10% firm can ever end up spending thousands of dollars per employee per month on AI, given that they seem to be using it more and more every month, but using it more cheaply every month. And that's great for them. You know, that's great for the median business because it means they get a better product every
7:15
Unidentified speaker (speaker_2): month. The models are getting better, they're spending less on it. But it does create new questions for the model companies as far as whether or not they will have the margins to invest in their models and make them better over time.
7:27
Unidentified speaker (UNKNOWN): Have
7:27
Julie Chang: you been using AI more or less these days? If you're a listener on Spotify, let us know in the comments. Coming up, what AI companies will need to do in order to stay relevant. if the hype continues to cool. That's after the break.
7:52
Unidentified speaker (speaker_1): Are there any green shoots that you'd highlight that maybe speak to potential areas of AI spending growth in the months to come next year?
8:00
Unidentified speaker (speaker_2): There are still a couple areas of growth in and what I would call new. fronts in the AI trade, things that we should be focusing on that we haven't yet so far. One is that this two-way race in AI models is likely to become a three-way race pretty soon. It's long been a game of open I versus anthropic. I think soon it's going to become OpenAI, Anthropic, SpaceX
8:20
Unidentified speaker (speaker_2): AI, and possibly even Meta. So we shouldn't underestimate the level of competition in this sector and that it's going to get even
8:26
Unidentified speaker (UNKNOWN): more.
8:27
Unidentified speaker (speaker_2): greater. Second, there are areas of spend that are starting to increase outside of engineering, and that is going to require models that are targeted to those use cases. Engineering use cases dominate 70 plus percent of AI spend, but customer support as it applies in sort of enterprise use is also starting to rise. Now about 13%
8:47
Unidentified speaker (speaker_2): of classified spend appears to be going toward customer support. That's going to rely on cheaper and fast models in addition to highly performant models. So model companies will need to prioritize whether or not they can deliver fast models, not just really smart ones.
9:01
Unidentified speaker (UNKNOWN): And
9:01
Unidentified speaker (speaker_2): then there's another category that I'm watching, and it's AIs, it's used for cybersecurity. There have already been a number of cybersecurity hacks that have been accelerated by the availability of AI models that are used to deploy those attacks. You're gonna see AI deployed in defense of cyber attacks as well. That's actually gonna drive more frontier model usage.
9:21
Unidentified speaker (speaker_2): And it's gonna mean frontier models that are designed to address those cybersecurity issues that will actually drive some more revenue toward the AI companies, ironically enough.
9:29
Unidentified speaker (UNKNOWN): I mean
9:29
Unidentified speaker (speaker_1): just putting all this together then, would it be fair to say that sure we are seeing some bubble-like bubble like characteristics here? But that kind of broadly speaking, ending the year, we are not in some sort of larger all hands on deck panic mode moment for the AI industry.
9:45
Unidentified speaker (speaker_2): It's kind of always been all hands on deck if you really think about it. You know, people just should not get too comfortable with their assumptions in this market. I think we've learned that again and again, because it's so easy for a new entrant to completely change how we think about the trajectory of AI spend by enterprise. And so that will continue. What I can say today is that the data
10:05
Unidentified speaker (speaker_2): shows that AI is getting cheaper and cheaper. That's generally going to be pretty good for businesses. That will unlock new use cases, but that also requires the AI model companies to resolve what is essentially a product issue, which is that today, AI models are generally designed for and particularly advanced for engineering use cases. But in order to drive greater growth, particularly
10:24
Unidentified speaker (speaker_2): in enterprise and outside of startups, they are going to need to address other enterprise workflows. Customer support's going to be a big one, cybersecurity is going to be a really important one as well.
10:36
Julie Chang: That was the WSJ's Luke Vargas speaking with Ramp's lead economist, Ara Carazian. And that's it for Tech News Briefing. If you're a listener on Spotify, be sure to leave us a comment. Today's show was produced by me, Julie Chang. with supervising producer Katie Ferguson. We'll be back tomorrow with a new episode.
10:56
Julie Chang: Thanks for listening.
This page was produced automatically: the audio was transcribed with nvidia/parakeet-tdt-0.6b-v3+sortformer-v2.1 and interpreted by Qwen3.6-35B-A3B-UD-IQ3_S.gguf. Transcription and interpretation can contain errors; always confirm against the original episode. Nothing here is investment advice.

More from Tech News Briefing

Tech News Briefing

3273: Are Always-Listening Wearables Cool or Creepy?

Episode discussion: Guest Nicole Nguyen tests Apple Watch Siri Recap and Amazon B Frontier, citing privacy and consent challenges

Speakers’ claims: Nicole Nguyen discusses her experience with always-listening wearables, specifically the Apple Watch Series 12 and Amazon B Frontier. She notes that Apple's Siri Recap generates emotionless bullet points from ambient audio, while Amazon's B device provides word-for-word transcripts and psychological insights. Nguyen highlights significant privacy concerns, noting that these devices resemble health trackers and may record bystanders without their knowledge. She describes social friction caused by consent requirements in all-party consent states like California. The episode concludes that a lack of federal law creates legal limbo regarding ambient recording.

WearablesPrivacyAI
10 min0 assets10 claims to verify
Tech News Briefing

3268: What the Jensen Huang Investigation Means for Nvidia

Episode discussion: Nvidia Board Investigates CEO Jensen Huang Amid Espionage Allegations and Stock Drop

Speakers’ claims: Host Julie Chang and analyst Amrith Ramkumar discuss Nvidia's board investigation into CEO Jensen Huang following his admission of an affair with Alice Huang, a Chinese executive suspected of foreign espionage. The episode details how Huang's relationship with Alice Huang, who worked at Megaspeed International, intersected with U.S. export control policies during the Trump administration, potentially influencing decisions regarding chip sales to China. Ramkumar notes that while no charges were filed against Megaspeed, the scandal has triggered an 8% drop in Nvidia's stock and raised concerns about succession planning and corporate governance. Political risks are highlighted, with Democrats gaining congressional control and pushing for stricter chip export laws, while President Trump publicly defended Huang. The segment concludes by examining the broader impact on the AI industry and Nvidia's deep integration with major tech firms.

NvidiaJensen HuangExport Controls
11 min1 assets10 claims to verify
All insights