Business IEA Says Data Center Power Demand Could Double to 1,000 TWh by 2030, With AI Supplying Three Quarters of the Growth Chatbots ChatGPT Now Transcribes as You Talk: OpenAI Removes the 10-Minute Cap and Streams Words Live on Every Plan Business Ex-Alibaba Executive's Everful Raises $48.5 Million to Put AI Agents in Charge of Sourcing From 30,000 Factories AI Agents Google Built a Mirror Dimension of the Internet to Train Its Chrome AI Agents Without Breaking the Real Web Productivity Canva Finally Adds an Off Switch for Its Generative AI Features, and a Filter to Hide AI-Made Templates Business Anthropic Begins Trading Today: 980 Million Shares at $83, a $1.5 Trillion Valuation and the Largest IPO on Record News Google Books Blue Origin New Glenn to Launch TPU-Carrying AI Satellites in Early 2029 Security Hackers Used the AI Penetration Tool ARTEX to Disrupt European Commission Systems for Over Ten Days Business IEA Says Data Center Power Demand Could Double to 1,000 TWh by 2030, With AI Supplying Three Quarters of the Growth Chatbots ChatGPT Now Transcribes as You Talk: OpenAI Removes the 10-Minute Cap and Streams Words Live on Every Plan Business Ex-Alibaba Executive's Everful Raises $48.5 Million to Put AI Agents in Charge of Sourcing From 30,000 Factories AI Agents Google Built a Mirror Dimension of the Internet to Train Its Chrome AI Agents Without Breaking the Real Web Productivity Canva Finally Adds an Off Switch for Its Generative AI Features, and a Filter to Hide AI-Made Templates Business Anthropic Begins Trading Today: 980 Million Shares at $83, a $1.5 Trillion Valuation and the Largest IPO on Record News Google Books Blue Origin New Glenn to Launch TPU-Carrying AI Satellites in Early 2029 Security Hackers Used the AI Penetration Tool ARTEX to Disrupt European Commission Systems for Over Ten Days

Databricks AI Index: Anthropic's Share of Enterprise Tokens Climbed to 68.8 Percent as Open-Weight Models Quadrupled to 30.9 Percent

Databricks AI Index: Anthropic's Share of Enterprise Tokens Climbed to 68.8 Percent as Open-Weight Models Quadrupled to 30.9 Percent

Databricks launched an AI Index built on anonymized data from more than 10,000 enterprises using Unity Gateway. It reports Anthropic's token share rising from 57.6 to 68.8 percent in six weeks, open-weight models climbing from 7.3 to 30.9 percent in four months, and per-task cost gaps up to 32x between models with identical 89.1 percent success rates. All figures are company-reported.

Databricks has launched an AI Index that tracks how enterprises actually adopt, spend on and evaluate AI models, and for the first time it has opened up its internal evaluation suite alongside the market data. The Index draws on two sources: anonymized, aggregated model-call and spending data from more than 10,000 organizations using Databricks' Unity Gateway, spanning digitally native companies, Fortune 500 firms and Global 2000 enterprises, and a coding benchmark built from the real workflows of Databricks' own engineers, using private repositories and hidden tests to limit benchmark contamination.

The headline finding is how fast enterprises stopped betting on a single vendor. In July, about 64 percent of enterprises on Unity Gateway relied on just one model provider; by November 2 that had fallen to 28 percent as organizations adopted two or more providers. Over the same period open-weight models' share of enterprise token traffic grew from 7.3 percent in July to 30.9 percent by November, more than quadrupling in under four months.

Anthropic, meanwhile, gained share even as the market diversified. The company's overall token share rose from 57.6 percent to 68.8 percent in the six weeks after Claude Opus 5.5 launched, with Opus 5.5 alone climbing from 7.2 percent to 37.9 percent of proprietary-model tokens. Databricks also observed that a new flagship proprietary or open-weight model has launched roughly every week since May 2026, or every 3.7 days when regional labs are counted.

The Index's most quotable numbers come from its coding benchmark, where identical success rates hide enormous cost differences. GPT-6 Luna and Claude Opus 5.5 both achieved an 89.1 percent task success rate, but Luna averaged $0.05 per task against $1.62 for Opus 5.5, roughly a 32-times gap. The reverse trap exists too: Gemini 3.8 Flash has one of the lowest token prices, yet averaged 140 turns per task and finished among the most expensive models at $11.97 per task. In the quality ranking, DeepSeek V4.1 and GLM 5.3 tied for third at a 93.1 percent pass rate, putting three open-weight models in the top ten.

The data covers September 21 through November 7 and the Index will update weekly. Everything in it is company-reported from Databricks' own customer base, and the company sells the gateway the data flows through, so the sample skews toward sophisticated, multi-cloud enterprises rather than the broader market. But the analytical point survives that caveat: cheaper tokens do not mean cheaper tasks, because reasoning turns, output length and retries drive the real bill.

That reframing is the Index's real contribution to enterprise procurement. Teams that budgeted AI the way they budgeted SaaS seats are discovering that the unit of cost is the completed task, not the token, and Databricks' numbers suggest that companies are already switching providers fast enough to exploit those gaps. Expect model vendors to respond with per-task pricing and more aggressive caching rather than further token price cuts.

Comments (0)

Log in to join the discussion

Log In

No comments yet