The number landed like a bomb: $7,400 per employee per month. US businesses, according to a Crypto Briefing report, are now spending that much on AI. Multiply it by 1.3 billion American workers—assuming a 1:1 ratio, which is generous—and you get $11.5 trillion annually. That is four times the total US corporate IT spend, and 30% of GDP. The number is not just improbable. It is structurally impossible. And yet, the crypto media ecosystem swallowed it whole, repackaging it as a bullish signal for AI-related tokens, GPU cloud providers, and the entire AI+Web3 narrative. I have spent the last three weeks reconciling the data behind this claim. The result is a forensic dissection of how a single dubious statistic can propagate through a hype cycle, and what it means for anyone allocating capital in this space.
Volatility is just liquidity leaving the room. But when the liquidity is based on fiction, the eventual exit is violent.
Context: The Crypto Briefing Report and Its Hype Cycle
The original article, published by Crypto Briefing—a vertical media outlet heavily embedded in the crypto ecosystem—claimed that US business AI spending had surged to $7,400 per employee per month. It did not cite a specific source. No survey methodology, no raw data, no breakdown by industry. Just a number. The timing coincided with a wave of AI-themed token launches, NVIDIA's earnings report, and a general push by crypto projects to rebrand as "AI infrastructure." The narrative was clear: AI spending is exploding, so buy the narrative, buy the tokens, buy the hype.
But the numbers do not hold. IDC’s worldwide AI spending forecast for 2025 (including government, consumers, and all industries) stands at $300–350 billion total. The US share is roughly 40%, or $120–140 billion annually. That is $10–12 billion per month. For the $7,400 per employee figure to be accurate, the US would need to be spending over $900 billion per month on AI alone—a factor of 90x higher than the most optimistic independent forecasts. The discrepancy is not a rounding error. It is a category error.
Trust is a variable I refuse to define. But when the data itself is a variable that cannot be verified, the entire analysis becomes a house of cards.
Core: Systematic Teardown of the Data
Let me be clear: I am not arguing that AI spending is insignificant. It is growing, and the divergence between high- and low-spending firms is real. But the specific number presented—$7,400 per employee per month—is a composite of at least four distinct errors: sample bias, misclassification of capital expenditures, conflation of total AI spend with per-employee averages, and omission of the denominator (the number of employees actually using AI).
First, sample bias. The likely source of the data is a survey of Fortune 500 firms or AI-native companies—not a representative sample of all US businesses. When you survey only the top 1% of spenders, the average is meaningless. The median is a better measure, and it is likely below $100 per employee per month. As of 2025, Microsoft Copilot costs $30 per user per month. ChatGPT Enterprise is $50 per user per month. Even the most aggressive AI deployment—full workflow automation for a software engineer—might cost $1,000 per employee per month in GPU rental and API fees. $7,400 is a fantasy unless you are buying entire clusters of H100s as a monthly subscription.
Second, misclassification of capital expenditures. The figure likely includes the amortized cost of data center construction, GPU procurement, and infrastructure upgrades. Those are one-time capital investments, not recurring monthly operational expenses. When you spread a $1 billion data center build over 10,000 employees for 12 months, you get $8,333 per employee per month—close to the reported number. But that is not a monthly spend; it is a capital project with a 5-year depreciation schedule. Conflating CapEx with OpEx is a classic error in crypto reporting, where speed of narrative often trumps accounting precision.
Third, conflation of total AI spend with per-employee averages. The report does not specify whether the $7,400 is for all employees or only those who directly use AI tools. If it is the latter, the denominator shrinks dramatically. A firm with 10,000 employees might have 500 AI-intensive power users. The average per power user could be $7,400, but the per-employee average for the entire firm is $370. That is a 20x difference. The headline obscures this nuance.
Fourth, omission of the denominator. The report never states the number of employees surveyed or the total spend. Without that, the figure is an orphan statistic. In my experience auditing smart contracts, I have seen similar data manipulation: a protocol claims $1 billion in total value locked, but when you dig into the wallets, 80% is a single whale's deposit that is soon withdrawn. The same principle applies here. The $7,400 figure is a synthetic average that masks a heavily skewed distribution.
Code doesn't lie. People do. The data is the code. And this code is full of reentrancy vulnerabilities.
Contrarian: What the Bulls Got Right
Despite the flawed data, the underlying trend is real. AI spending is diverging between firms. The top 10% of US companies are likely spending 10–100x more per employee on AI than the bottom 50%. This divergence is accelerating. A 2024 McKinsey survey found that firms with advanced AI adoption reported 20% higher EBITDA growth than laggards. The gap is widening. The structural advantage of early adopters is real, and it will compound.
But the bulls miss a critical point: the divergence is not a function of spending alone. It is a function of integration efficiency. A firm that spends $7,000 per employee on AI but deploys it poorly will be outperformed by a competitor that spends $700 per employee but integrates AI into core workflows. The latter is more common than the former. In my work auditing DeFi protocols, I have seen teams with $10 million budgets build insecure, bloated systems, while lean teams with $500,000 produce audited, efficient contracts. Spending is not a proxy for competence.
Furthermore, the open-source AI ecosystem (Llama 3, Qwen, Mistral) is democratizing access. Small firms can now run models locally at a fraction of the cost of API calls. The $7,400 per employee figure assumes a closed, API-heavy model. The reality is that many firms are moving toward on-premise or hybrid deployments, which drastically reduce per-employee costs. The bulls are extrapolating a linear future from a sample that is already at the top of the S-curve. The next phase will see cost compression, not expansion.
Takeaway: Accountability in Crypto Media
The $7,400 per employee AI spending myth is not an isolated incident. It is a symptom of a media ecosystem that prioritizes narrative over verification. Crypto Briefing, like many outlets, operates on a scarcity of attention. A shocking number generates clicks. Those clicks fuel token speculation. The cycle repeats.
Audit reports are hope dressed as documentation. The same applies to these spending surveys. They are hope dressed as data. As an auditor, I have a responsibility to flag the discrepancy. As a reader, you have a responsibility to demand transparency. Next time you see a headline with a single, round, and suspiciously large number, ask: where is the raw data? Who was surveyed? What is the denominator?
If you cannot explain the exploit, you caused it. And if you cannot explain the data, you are funding the next collapse.