Ramp's Economist Found Cracks In The AI Thesis | The Brainstorm 156

Anthropic

Across the shows: 101 speakers in other episodes, 126 bullish and 85 bearish calls, past 30 days

• Ara Kharazian · RampBullish

  Anthropic is now the most-used business AI model, with consumer strength driving enterprise adoption.

  Per Ramp data, Anthropic is now the most popular AI model among businesses and has the largest business spend. Its rise came over roughly a year. It lines up with the launch of Claude Code and the coding-agent wave, and with its pivot to consumer (Super Bowl ads, and a celebrity posting about using it), which was the season of its run-up. Ara argues that consumer product strength can drive enterprise adoption and spend, which suggests the consumer brand is feeding its enterprise share gains.

  “I wouldn't underrate the extent to which consumer strength as a product can actually drive enterprise adoption and spend in this market.”at 15:16

• Brett Winton · ARK InvestBullish

  Anthropic's enterprise traction comes from startup B2B customers, not its consumer brand.

  Brett argues that Claude's consumer adoption is not a leading indicator of its enterprise penetration. Anthropic goes after customers who would not typically adopt, namely startup and enterprise B2B buyers, and delivers a product that feels safe and is heavily customizable. Ramp data shows Opus 5.5 taking a large share. Anthropic's aggressive pricing ahead of its IPO is designed to drive token volume and lock in revenue, which helps it cover its infrastructure commitments.

  “Call it startup enterprise b to b, and they're able to deliver them this product that is magic, but that they also feel safe with and that is heavily customizable to the enterprise.”at 16:01

Meta Platforms

Across the shows: 69 speakers in other episodes, 75 bullish and 36 bearish calls, past 30 days

• Ara Kharazian · RampUncertain

  Meta's enterprise AI entry should not be underrated, though it has no meaningful share yet.

  Ramp data shows no significant Meta share in business AI today, because Meta only recently entered enterprise. Ara still argues that competition is underrated: new entrants have taken more than half the market within months, and Anthropic's rise took about a year. Business AI spend also lags consumer adoption, so Meta's consumer reach could help it win enterprise spend later.

  “And I wouldn't underrate Meta either because part of what's so interesting about this market being a market for what is essentially b to b spend is that it's actually developed in many ways as a sort of lagging indicator of how the consumer market has developed.”at 14:17

Enterprise AI adoption & AI software

• Ara Kharazian · Ramp

  Top 1% AI spenders are slowing spend even as their token usage keeps rising.

  Ramp data shows the top 1% of AI-spending businesses, spending roughly $7,000 to $10,000 per employee per month, slowing from the hyper-growth of the first half of the year (September versus July, with early October data showing the same). Token volumes over the same period kept rising, and seasonality does not explain it. Ara attributes the slowdown to falling prices (AI cost almost halved over the window, so usage can double while token spend flatlines), and to top firms adding spend controls such as limits on frontier models for automated tasks, better defaults and routers. Ramp itself is a top 1% business and is starting to implement these controls. She says the slowdown is not a crash, but that it is observable in the top 1%.

  “And it's not that they're using it less. Right? They actually are using it more based on all the indicators we have. They're using more tokens.”at 2:35

• Brett Winton · ARK Invest

  A three-month slowdown is not a trend; top 1% AI spend is still growing several-fold year over year.

  Brett notes that the top 1% spend grew roughly four to five times year over year, and that the 'slowdown' rests on a short window (September versus July). Three months is not a trend, and he doubts that the top 1% will stop growing spend. Frontier labs are among the fastest-growing companies ever. Over time, he expects spend to be driven by the median firm and diffusion, not by the top 1% alone, because the top 1% will not hold a steady level of spend.

  “And I'd just be careful about, like, a month to month change or call it a three month change is... Does not mean that kind of like even amongst those top 1%, they're gonna stop growing spend.”at 11:43

• Ara Kharazian · Ramp

  Coding agents are the most commercially advanced AI, but median firms won't match engineers' value from them.

  Coding agents are clearly productivity-enhancing for the tech sector and for the job category that uses them actively. Ara finds it hard to see a world in which the median business gets as much value from a coding agent as a typical software engineer at a tech company, so she does not expect the median firm to reach those levels. She acknowledges that new products may bring another coding-agent-style revolution, but argues that current expectations should be grounded in the data.

  “At the same time, today, the most commercially advanced form of AI is coding agents, which are clearly productivity enhancing, at least for the tech sector and for that job category that uses them actively.”at 23:13

• Brett Winton · ARK Invest

  Models still need heavy hand-holding for core business outputs; spend on frontier models is necessary.

  Brett, who has used these models aggressively, finds them powerful but not very good at core business outputs. For a half-hour business task, the output is poor and requires hand-holding. For bureaucracy-type work they are reasonable. How far you can loosen the reins depends on the model's power and the amount you spend on it. He sees strategic necessity in spending on frontier models and increasing that spend over time for exploration and head-to-head competition.

  “give it a task that would take you half an hour and really try to have it do it and deliver the business product outcome, and it's just not good.”at 26:26

• Ara Kharazian · Ramp

  Median firm AI spend per employee will fall under $1,000 per month within five years.

  Ara estimates the median firm's AI spend per employee per month will be under $1,000 in five years, and says she is giving herself room. She benchmarks against the spend on a computer and argues that predictions of tens of thousands per employee per month are unrealistic. She expects AI to become table stakes, like Google Ads for a restaurant, so firms that adopt it gain an edge and then need it to compete. Median firms can get significant ROI without heavy spend, and AI use will extend beyond knowledge workers.

  “I'll say for the median firm, it's probably gonna be easier for me. The median firm per employee under 1,000 per employee per month in five years.”at 29:05

AI pricing & unit economics

• Ara Kharazian · Ramp

  AI prices are falling fast from OpenAI-Anthropic competition and cheaper models, pushing down enterprise AI spend.

  AI is getting cheaper because of intense competition between OpenAI and Anthropic, which are announcing price cuts and releasing more efficient models. Enterprises are also getting better deals, as both labs now offer large-enterprise contracts with new discounts. New high-performing models at standard and light tiers add further downward pressure. Over the roughly three-month window in question, the price of AI fell by almost half. Ara argues that competition is no longer just between OpenAI and Anthropic, so these price declines will continue and get stronger over time.

  “AI is getting cheaper because there's so much competition between OpenAI and Anthropic and [stating] price cuts and also introducing more efficient models.”at 2:42

• Brett Winton · ARK Invest

  Falling token prices let firms keep their AI activity while halving costs every couple of months.

  OpenAI cut Luna's prices by 80% to win marginal share, and Luna is a big spend driver in the dataset. Efficient models such as Opus 5.5 and the omitted 6.1 also lower costs. If performance per dollar improves about every two months, firms can keep the same set of activities and halve their costs every two months. The question is whether firms will switch to cheaper providers, which will take time because of the friction of status-quo bias.

  “If you have performance per cost having, like, every two months or whatever one, that means you can keep up your same set of activities and half your costs every two months.”at 22:20

Venture capital & startups

• Ara Kharazian · Ramp

  AI demand is concentrated in VC-backed firms, so a slowdown there would cut AI spend in unison.

  Ara agrees with Chamath that AI revenue is not circular but is a revenue quality problem. The top 1% of AI spenders on Ramp's platform are mostly VC-backed tech companies. That is fine while AI spend keeps growing and those companies keep growing their own customer bases. The concentration is also a correlation risk: if the AI trade slows, all of these firms would cut AI spend simultaneously, which could cascade across the sector.

  “I do think he's actually right about it, that it's not circular revenue problem, but it is a revenue quality problem.”at 18:19

• Sam Korus · ARK Invest

  AI spend stops once AI startups can't raise their next round, so revenue quality depends on funding.

  Sam cites Chamath's point that perhaps 70 to 80% of AI company revenue comes from other AI companies funded by venture capital. He frames it as a revenue quality problem, not a circular one. He concludes that if AI companies cannot raise the next round, their AI spending stops. He also questions how many two-year-old companies valued at $100 billion can exist.

  “Companies, if you can't raise the next up round, then your AI spend stops.”at 19:02

Cybersecurity

• Brett Winton · ARK Invest

  Cybersecurity will become a spend category every firm must fund as AI-driven attacks rise.

  Brett imagines non-tech firms without developers being crippled by cyberattacks from open-weight models. Even non-tech firms have substantial tech exposure that they must protect. Each firm's infrastructure is a unique nested stack of acquired or purchased systems that needs its own security function, and any open channel is a recipe for infiltration. Ara agrees that cybersecurity will be a rising spend category.

  “I think that cybersecurity is going to be a [spend] category that everybody's gonna have to spend on.”at 25:09

AI: overall outlook

• Sam Korus · ARK Invest

  Model capabilities are advancing faster than even top AI spenders can absorb; very bullish on the technology.

  Sam's optimistic read of the spend slowdown is that the underlying models are improving faster than the top 1% of companies can figure out how to spend to keep up. More capability is being released than even leading-edge companies can deploy. He takes this as a reason to be extremely bullish on the technology, not as evidence that the AI thesis is breaking.

  “The capability of the models are moving faster than even the 1% of companies... there's more power being released by the technology than even the leading edge companies can figure out how to spend enough to keep up with it. And so all of this makes me extremely bullish on the technology.”at 28:05