Overview of AI Fact or Fiction: The Fable Ban, Tokenmaxxing, Saaspocolypse — With Ara Kharazian
Alex Kantrowitz speaks with Ramp lead economist Ara Kharazian about the biggest AI narratives in the market and tests them against Ramp’s business-spend data. The conversation focuses on whether the White House’s export restrictions on Anthropic will hurt the company, how much companies are actually spending on AI per employee, whether “tokenmaxxing” is real, why DeepSeek and other open-source models are gaining attention, and whether the “SaaS apocalypse” is actually happening. The recurring theme: AI adoption is real and growing, but the business reality is more nuanced than the hype.
Key Takeaways
-
Anthropic’s government trouble may not hurt it as much as expected
- Kharazian argues the White House export-controls controversy could even backfire in Anthropic’s favor by reinforcing its safety-first brand.
- He points to a similar earlier episode, when the Department of Defense labeled Anthropic a supply-chain risk, yet business adoption still accelerated.
- The “forbidden fruit” effect may increase demand once access returns.
-
AI spend is growing fast, but it’s still not dominant in most firms
- Ramp data shows AI spend has increased dramatically, but for most companies it remains a relatively small share of overall spending.
- The median firm spends about $11 per employee per month on AI.
- The top 10% spend about $611 per employee per month, and the top 1% spend about $7,500 per employee per month.
-
“Tokenmaxxing” is not really showing up in the data
- Despite fears of runaway AI spend, Kharazian says firms are not wildly overspending on tokens in a way that suggests panic or waste.
- AI spend is rising, but firms are increasingly trying to impose controls and understand where costs are going.
-
OpenAI and Anthropic are still growing, but usage is more fragmented than a simple winner-take-all story
- Anthropic has recently overtaken OpenAI in Ramp’s business-adoption data.
- But many advanced companies use multiple vendors, not just one, to route tasks across different models.
- The most sophisticated AI users are not locked into a single provider; they experiment and switch based on task and performance.
-
DeepSeek and open-source models are gaining share, but from a small base
- DeepSeek is one of the fastest-growing vendors in the data, but its actual usage remains small.
- Kharazian sees open-source and routing layers as a real trend, but not yet a dominant one.
- He believes companies like Google may be underrated winners because they can offer AI as part of a broader ecosystem and aren’t as dependent on token spend.
-
The “SaaS apocalypse” is overstated
- Traditional SaaS is not being wiped out by AI as quickly as some headlines suggest.
- Seat-based pricing remains the dominant model.
- Even AI-native companies are still seeing strong subscription demand, not just token-based usage.
- Legacy software vendors face pressure, but much of it comes from AI-native competitors, not just OpenAI or Anthropic.
Anthropic, Safety Branding, and the Export-Control Controversy
Why the White House move may not be fatal
Kharazian says the government’s actions could actually strengthen Anthropic’s market position. If the model is portrayed as so powerful it must be restricted, that can create a halo effect and signal quality.
Lessons from the Department of Defense episode
- The DOD’s “supply chain risk” label did not lead to a broad customer exodus.
- Businesses largely continued using Anthropic because:
- the label wasn’t taken as seriously in practice,
- the model remained highly competitive,
- the government later softened its position with exceptions.
The brand benefit
Anthropic’s safety-first identity may become more credible if the government itself is acting as though the model is unusually powerful or risky.
AI Spend: What the Data Says
The headline numbers
- Median spend: about $11 per employee per month
- Top 10% of firms: about $611 per employee per month
- Top 1% of firms: about $7,500 per employee per month
What that means
- AI spend is growing at an extraordinary rate, but for most firms it’s still a small slice of total spend.
- The top spenders may be using AI not just for employee-facing tools, but also for internal systems and infrastructure.
- The data suggests firms are experimenting and scaling gradually, not all-at-once.
Why the trend matters
Kharazian argues the shape of the spend curves suggests firms are still learning what “good AI usage” looks like. There is no clear plateau yet, and no evidence that companies have found their final benchmark level of AI spend.
Why Companies Keep Buying Multiple AI Vendors
The “best model” is not the whole story
Kharazian emphasizes that AI adoption is not just about raw model quality. It’s also about:
- workflow integration,
- product experience,
- reliability,
- and how well the tool fits a specific task.
Stickiness comes from workflow, not just performance
- Users often stay with tools like Claude Code or Codex because they fit into their workflow.
- Better models alone do not automatically trigger switching behavior.
- AI products are becoming embedded in work habits, which makes them more durable than people expected.
Multiple vendors are normal in advanced usage
The most advanced firms on Ramp use an average of eight AI vendors, compared with around two for median firms.
DeepSeek, Open Source, and Cost Control
What’s driving open-source adoption
The biggest reason businesses hesitate to adopt AI more broadly is cost uncertainty. They want:
- predictable pricing,
- better admin controls,
- and a way to route simple tasks to cheaper models.
Why DeepSeek spiked, then faded
- DeepSeek had a burst of attention in early 2025.
- Its adoption quickly fell after OpenAI and Anthropic responded with cheaper alternatives.
- Ramp sees a new rise in DeepSeek usage, but from a very small base.
Bottom line on open-source
Kharazian thinks open-source tools are useful and growing, but still overrated in the broader market narrative.
Why Google Might Be the Quiet Winner
Kharazian repeatedly points to Google as an underrated beneficiary of the AI shift because:
- it has multiple revenue streams beyond token usage,
- it can afford to make AI relatively cheap,
- it already has distribution through Google Workspace,
- and it can make AI a loss leader if it grows cloud usage.
In his view, Google is better positioned than OpenAI or Anthropic to give businesses the cost controls they’re asking for.
Why the SaaS Apocalypse Hasn’t Arrived
The two versions of the thesis
Kharazian separates the “SaaS apocalypse” into two claims:
- AI will wipe out traditional SaaS vendors.
- SaaS vendors will shift almost entirely to token-based pricing.
He says both are overstated.
What the data shows
- Seat-based pricing still dominates.
- Metered usage is still a small share of spend.
- Even AI companies are seeing strong subscription growth.
- Traditional SaaS vendors are facing pressure, but not collapse.
Examples
- CRM: Salesforce is still dominant, though AI-native competitors like Attio are growing.
- Design: Figma remains strong despite fears that design tools would be displaced by AI.
Main Insight
The episode’s central message is that AI is changing how businesses buy software, but not in a simplistic winner-take-all way. Adoption is broadening, spend is climbing, and companies are experimenting across multiple tools. But the data does not support the most extreme narratives: Anthropic is not obviously doomed, DeepSeek is not yet a massive threat, and SaaS is not dead.
What to Watch Next
- Ramp’s upcoming research on AI’s impact on jobs
- Whether Anthropic’s government controversy creates more demand or real friction
- Whether businesses shift more spending toward cost-controlled routing layers
- Whether Google keeps gaining share as an AI platform with broader ecosystem advantages