Ramp data shows AI token spend stalled at 1.5% growth in August and September, days before Anthropic's IPO push

Spending on AI tokens by American businesses nearly stopped growing in late summer, according to Ramp, the corporate card and spend management company. The figures, reported by TechCrunch on November 6, 2026, arrive as Anthropic prepares to market its IPO and as lab valuations are being repriced. The data is one company's customer base, but it is the cleanest public series on what businesses actually pay for AI.

The numbers

Ramp tracks token spend from its customers' card and bill payments. TechCrunch reports that token spend grew 42% from June to July 2026, then 1.5% across August and September. It re-accelerated in October: 8.9% from September 23 to October 23.

Ramp's own AI Index post, dated October 14, 2026, agrees on the flat stretch: "virtually no growth in August or September" for token spend, while total AI spend reached a new peak in early October, 1.5% above the August peak. Per-employee spending among the top 1% of firms fell 3.1%, from $9,873 in July to $9,568 in September. TechCrunch rounds the peak to roughly $10,000.

Adoption is not falling. In September, 43.8% of Ramp businesses paid Anthropic and 40.6% paid OpenAI, while overall AI adoption was 55.9%, up 0.62 points, per the index. More firms are buying. They are not spending proportionally more.

Who is spending

The distribution is steep. Median firms spend $12 to $15 per employee per month on tokens, while the top 10% spend $600 to $1,000 and the top 1% about $10,000 at their peak, per TechCrunch. Ramp's earlier post, cited in its index, found that 80% of OpenAI and Anthropic spend comes from 1% of firms.

Ara Kharazian, Ramp's lead economist, told TechCrunch that "Almost the entire spend on AI is from venture-backed, high-growth tech companies." That should temper any extrapolation to the economy at large. A handful of AI-native firms can set the trend while median companies barely register.

Price cuts muddy the signal

Prices fell in the window. TechCrunch reports that OpenAI cut API prices in half for GPT-6 Sol and GPT-6 Luna in September, and that Anthropic cut prices for Fable 5.1 and Opus 5.5, which Anthropic said in a recent filing reduced its revenue growth by about $600 million. Our own frontier API price tracker records the list prices.

When prices fall, flat dollar spend means more tokens bought, not less usage. Kharazian put it this way: "OpenAI and Anthropic price cuts have…just made it cheaper to use AI," and "You can use more AI, but use it for less." The Ramp page shows a volume view alongside spend, but TechCrunch does not describe sample size or weighting, so outsiders cannot tell how much of the flattening is price.

The valuation backdrop

TechCrunch ties the data to a repricing. It cites Bloomberg's report that Anthropic's potential valuation fell from $2 trillion in early October to $1.2 to $1.3 trillion, and the Financial Times report that OpenAI lowered its proposed valuation to $1.25 trillion from $1.4 trillion. We have not verified either valuation against a filing, and both are reported figures.

What is on the record is Anthropic's amended S-1, which showed revenue growth of about 47% quarter over quarter in Q3, down from about 144% in Q2, as we covered in our report on the amended filing. Ramp's series and the filing point the same direction: still large, but no longer vertical.

What the data cannot say

Ramp customers skew toward US small and mid-size companies that pay by card, and large enterprises with negotiated contracts are underrepresented. Ramp warns that top-1% estimates are volatile because the group is small, and revised them up across all periods after adding vendors. The author also wrote, in a rare moment of candor, "I sometimes question the value of this metric."

October's 8.9% rebound is a single month. Ramp's Kharazian also wrote that "It feels like we're at the top of the S-curve" in the October 14 post, which is an opinion, not a measurement. Treat it as a hypothesis to test against next quarter's data.

How to use the index

Two habits help. First, read volume and dollars together. The Ramp page offers a token volume view, and Kharazian's price-cut point means dollars understate usage when prices fall. Second, watch the cohort, not the headline. The top 1% moved most, and it is also the least stable group because it is small.

The October 14 post contains other lines worth a note. Kharazian wrote that "It's clear this effect is not seasonal," and, about negotiating dynamics, that "the customer is in the favorable negotiating position." We read those as the author's judgments in a short analysis post, not as findings from a published method, and the quoted fragments are all we have of that context.

For labs, the implication is the old one: growth that depends on a few hundred heavy buyers is exposed to those buyers' budgets. For everyone else, it is a reminder that a median company is paying about what a streaming subscription costs per employee.

Our take

The story is not that AI demand collapsed. It did not. It is that the easy compounding of mid-2026 paused, that price cuts blur the dollar signal, and that the spend sits in a thin layer of firms. Anyone pricing the labs off the summer run rate should look at the October data, the volume series and the labs' own filings before trusting a straight line.