SemiAnalysis拆解企业AI预算：Meta曾单月消耗70万亿tokens，但真正的风险不是客户不用AI了
Related
SemiAnalysis breaks down enterprise AI budgets: Meta once consumed 70 trillion tokens in a single month, but the real risk isn't that customers are no longer using AI.
According to BlockBeats, on July 1st, enterprise AI usage is shifting from "use as much as possible" to "use within limits." SemiAnalysis, in its Token Budgeting report released on July 1st, stated that the tokenmaxing trend popular at the beginning of the year—encouraging employees to consume as many AI tokens as possible to boost productivity—is being replaced by more realistic budgeting systems. However, the firm believes that media narratives about companies cutting AI spending are exaggerated, and OpenAI and Anthropic's API businesses did not face substantial budget risks in the second half of the year. The SemiAnalysis team stated that after communicating with over 50 enterprise clients via Slack, phone calls, and the Databricks AI Summit, they found that while most companies are indeed starting to set caps on AI usage, a unified standard has not yet been established. Low-end budgets may be only $250 to $500 per person per month, while high-end budgets can reach $2,000 or even tens of thousands of dollars per month. A major US aerospace and defense manufacturer set a monthly budget of $250 for some employees, while a major pharmaceutical company set it at $500; more cutting-edge technology companies like Workday and Stripe have budgets of around $2,000 per month for some employees. This contrasts with the "token maximization" trend at the beginning of the year. The report mentions that companies like Meta and Salesforce encouraged employees to use AI tools extensively. Meta even had an internal dashboard called "Claudeconomics" that ranked the company's top 250 heavy users. Data showed that Meta employees consumed over 60 trillion tokens in 30 days, with the highest single user consuming approximately 280 billion tokens. This dashboard was shut down two days after the related reports. Uber was also reported to have exhausted the annual budgets of Claude Code and Codex within four months, subsequently setting a limit of $1,500 per person per month, with any excess requests requiring case-by-case approval. However, SemiAnalysis believes these extreme cases reflect loose incentive mechanisms and management rather than a general peak in enterprise AI spending. The report states that the top 10% of high-spending clients contribute the majority of AI lab revenue, and these clients are at low risk of cutting API spending for the remainder of the year. Even though Meta consumed approximately 70 trillion tokens per month in February, and at a price of nearly $50,000 per employee per year, SemiAnalysis estimates this still only accounts for 3% to 5% of Anthropic's revenue. Enterprise spending is also highly unevenly distributed. SemiAnalysis cites Ramp data stating that the top 1% of clients spend an average of [missing data - likely referring to a specific amount of money] per employee per year...
SemiAnalysis: Meta will accelerate, rather than slow down, its computing power procurement; it is in talks with Anthropic to build its own AI model service platform.
According to BlockBeats, on July 3rd, SemiAnalysis stated in its latest report that after news broke that Meta might become a new Neocloud, the market's initial reaction was to sell off computing power cloud companies like CoreWeave and Nebius, and to renewed concerns about "AI computing power oversupply." However, the firm's assessment is the opposite: this concern may be wrong. Meta's data center and computing power procurement will not slow down, but rather continue to accelerate. The article mentions that in the first half of this year alone, Meta has already signed contracts for over 5GW of capacity in the cloud services and managed data center sectors, and this does not include its accelerating self-built projects. SemiAnalysis states that Meta is in final negotiations with Anthropic, hoping to obtain access to Claude private instances. If this comes to fruition, the significance goes beyond "Meta buying more computing power"; it suggests that Meta may be building its own AI model service platform. This model is somewhat similar to AWS's Bedrock, Microsoft's Foundry, and Google's Vertex. Meta can initially use Claude internally, or it can package the model capabilities as a token-as-a-service to provide services externally in the future. In the short term, it might use its own models externally and Anthropic models internally; in the long term, Meta might even incorporate Anthropic and OpenAI models into its external service system. SemiAnalysis states that the underlying logic is that Meta has computing power, advertisers, social network distribution capabilities, and consumer-end entry points. If it can combine cutting-edge models, intelligent agents, and sales and marketing SaaS, it will not just be a company that buys GPUs, but will be moving towards the upper layers of AI application and model distribution.
FalconX withdrew 73,900 HYPE tokens from Gate in the past 7 minutes, equivalent to approximately $5.03 million.
According to BlockBeats, on July 8th, Onchain Lens monitoring showed that FalconX withdrew 73,900 HYPE tokens from Gate in the past 7 minutes, worth approximately $5.03 million.
AKE's major shareholders transferred another $2.24 million worth of tokens to Binance Alpha and then sold them off, causing AKE to drop another 33%.
According to BlockBeats, on July 7th, on-chain analyst Yu Jin monitored that three days ago, approximately $1.22 million worth of AKE (3.944 billion tokens) was transferred to Binance Alpha by AKE market makers and then sold off. AKE fell 40% that day, from $0.0005 to $0.0003. In the past few hours today, approximately $2.24 million worth of AKE (9.825 billion tokens) continued to be transferred to Binance Alpha by AKE market makers, while a sell-off is also occurring on-chain. Affected by this, AKE was again dumped by 33%, with the price falling from $0.0003 to $0.0002.
Jupiter Strategic Reserve Trust increased its holdings by approximately 186,500 JUP tokens, bringing the total value of its holdings to approximately $34.8 million.
According to BlockBeats, on July 4th, the Jupiter Litterbox Trust, Jupiter's strategic reserve trust, increased its holdings by 186,546 JUP, worth approximately $45,000. This brings its total purchases this month to 1,226,119 JUP, worth approximately $300,000. Its total purchases to date amount to 145,028,229 JUP, worth approximately $34.8 million. The Jupiter Litterbox Trust is Jupiter's official on-chain treasury fund, automatically transferring 50% of Jupiter protocol revenue to it. Through smart contracts, it continuously buys and holds JUP tokens on the open market, often jokingly referred to by the community as the "litterbox trust."
Palantir CEO: Enterprises are dissatisfied with "cutting-edge labs" like OpenAI and Anthropic, which only pursue token maximization.
According to BlockBeats, on July 2nd, Palantir CEO Alex Karp, in an interview with CNBC's "Squawk Box," strongly criticized leading AI model companies, calling the way AI is sold "completely wrong." Karp emphasized that companies are already dissatisfied with "cutting-edge labs" like OpenAI and Anthropic, believing they only pursue token maximization, wasting companies' time and money while handing over proprietary value and IP. Karp stated that companies are "angry" and will commit to owning their own AI production resources rather than relying on third parties. On June 29th, Palantir partnered with Nvidia to deploy Nvidia Nemotron open AI models in sovereign environments, primarily serving the US government and critical infrastructure customers. The collaborative system reportedly integrates Nvidia AI technology with Palantir's AIP, Foundry, Ontology, and Apollo platforms, helping organizations train, customize, and deploy AI locally while maintaining complete control over data, intellectual property, and models.