The number is not a verdict. It is a signal. In Q2 2025, KPMG published a survey that should not be cited in boardrooms as evidence of failure, but as proof of a transfer mechanism — a reallocation of capital away from unproven promises. Only 7% of global business leaders can demonstrate a positive return on their AI investments. Ninety-three percent cannot. This is not a tech failure. This is a liquidity failure.
The quiet of the bear market taught us to count coins, not to count on narratives. The enterprise AI sector is entering its own bear market — not in price, but in proof. The implications for digital assets are not parallel. They are direct. The same capital that funds NVIDIA's GPUs and Microsoft's copilots is the capital that rotates into BTC ETFs and DeFi treasury yields. A crack in the AI procurement armor is a crack in the global risk asset dam.
I have mapped liquidity flows from ICO era through the DeFi summers and the post-ETF institutional rotation. The KPMG data points to a deceleration node. When corporate CFOs admit they cannot measure the output of a $200 billion annual AI capex cycle, they are not confessing ignorance. They are announcing a pending shift in spending preferences. That shift will compress tech stock multiples, raise the cost of capital for unprofitable AI startups, and ultimately increase the relative appeal of scarce, verifiable assets — Bitcoin first, then selective yields.
We do not predict the storm; we build the hull. But to build the hull, we must first map the ice.
Context: The Global Liquidity Map
The current macro backdrop is a fragile expansion. The Fed has paused rate hikes but has not returned to quantitative easing. Global M2 money supply is recovering at a modest 3.5% year-over-year, driven by Japan's continued yield curve control and Europe's reluctant fiscal loosening. In this regime, capital allocators are not chasing maximum returns. They are chasing verifiable returns.
The KPMG survey arrives at a critical intersection: the U.S. corporate fiscal year budget cycle, roughly Q4 2024 to Q1 2025, has just closed. Those budgets included aggressive AI spend. Salesforce, Microsoft, and AWS have all reported strong cloud and AI-related growth. But the CFOs who signed those purchase orders are now facing the uncomfortable task of justifying that spend to audit committees and shareholders in Q2-Q3 2025.
The narrative in the KPMG report is not benign. It states that only 7% of leaders can substantiate ROI. That means that for 93% of enterprise AI spending, the investment is held together by narrative, not by numbers. For a liquidity-driven analyst, this is a short-term bullish signal for AI infrastructure (gpu cloud, data platforms) and a medium-term bearish signal for AI application-layer companies that rely on recurring software revenue. But for the crypto market, the signal is more complicated.
In the quiet of the bear, we count the coins. In the loud of the bull, we count the leaks. The KPMG data is a leak from the traditional finance system — a canary in the liquidity coal mine. When CFOs lose trust in their AI suppliers, they do not simply stop investing in AI. They reduce overall risk appetite. They delay, they shrink, they concentrate purchases into a few proven vendors. This behavior reduces the velocity of money in the tech sector, which historically precedes a rotation into safe-haven assets. Bitcoin, as the most mature crypto asset, stands to be a direct beneficiary of any "capital flight to proof" behavior.
But I am not here to give you a simple "AI crash = crypto pump" thesis. That is immature thinking. The actual transfer mechanism is slower, more structural, and more profitable for those who position early.
Core: The Attribution Gap and the Measure of Value
The first lesson from the KPMG data is that enterprises, not AI scientists, now control the pace of adoption. The bottleneck is no longer model capability; it is the ability to prove that a model's output translates into a financial outcome. I have spent the past decade building scripts to track yield differentials across Aave and Compound, mapping capital flows in ICO pre-sale whale wallets, and stress-testing ETF custody vulnerabilities. The common thread is that all value in a financialized system ultimately rests on observable, auditable, and repeatable evidence of value creation.
In the enterprise AI world, that evidentiary framework is nascent. The 7% statistic reflects a systematic absence of "attribution protocols." When a bank deploys an AI system to improve customer service, the output is a mix of human agent changes, system changes, and seasonal demand shifts. Isolating the AI's incremental contribution to net promoter score or call resolution time requires a controlled experiment. Enterprises are not built for controlled experiments. They are built for quarterly projections.
This is not a judgment on AI technology. Task-level productivity gains are real. GitHub's coding assistant has demonstrated 30-50% faster completion times for routine tasks. Customer support AI tools show 20-30% improvement in first-contact resolution. The gap is not at the task level; it is at the process level. The inability to measure process-level ROI is a governance problem, not a physics problem.
And governance problems are precisely the kind of problems that create liquid markets. Think about what happened when enterprises realized they could not prove the security of their DeFi protocols. The market responded by creating insurance protocols (Nexus Mutual, Sherlock) and audit marketplaces (CertiK, Hacken). A similar response is already emerging in the AI space. Tools like Arize AI and Weights & Biases provide observability, but they do not yet translate technical metrics into CFO-friendly financial terms. The demand for "AI Financial Analysts" is skyrocketing. The firms that build this bridge between model quality and monetary value will capture more value than any single AI application.
The alpha hides in the variance others ignore. The variance here is the gap between AI capability and AI measurement. If we accept that 7% of leaders can prove ROI, we must ask: what are they doing differently? My experience with cross-protocol yield arbitrage in 2020 tells me that the 7% likely share three traits: they have a strong data engineering foundation, they embed AI into narrow, high-frequency processes rather than broad strategic initiatives, and they have executive leadership that rewards precision over novelty.
Consider the example of code generation. A mid-sized fintech company deploys a code review AI. The tool flags potential bugs and suggests fixes. In a three-month period, the firm sees a 15% reduction in production defects. The CFO can calculate the savings from reduced outage time and manual review hours. That is ROI: measurable, attributable, and directly tied to a P&L line item. In contrast, the same company may deploy a "strategic analysis assistant" for the executive team. The output is a series of slides. The time saved is maybe two hours per week. The value is real but diffuse. No CFO can justify the seat license. This is the fundamental distinction between high-definition AI use cases and "AI theater."
My own foray into AI-agent economic modeling in 2025 was grounded in this principle. I simulated autonomous agents buying and selling compute resources on decentralized networks. The model showed that 15% of smart contract interactions could be machine-to-machine by 2026. But the financial validity of that forecast depends on measurable outcomes — cost per transaction, latency reduction, and error rates. No one will fund a "general AI agent economy" without a unit economics deck that shows clear margins.
In the crypto world, we already have a primitive version of this measurement problem. Proof-of-reserves audits for exchanges like Binance and Coinbase are considered best practice, but they are static snapshots. The AI world needs a "proof-of-inference" standard, one that verifies not only that a model ran, but that its output led to a specific business outcome. This is a new asset class — call it "Evidence Infrastructure." The delta between the 7% and the 93% is precisely the delta between a company that has adopted evidence infrastructure and one that has not.
The Institutional Capital Flow Shift
Now we arrive at the investment thesis. The KPMG report does not exist in a vacuum. It aligns with a pattern of earnings call language from major AI vendors. Snowflake's 2025 guidance implies slower consumption growth. Palantir's U.S. commercial revenue growth, while strong, increasingly relies on "boots-on-ground" consulting rather than pure software leverage. Microsoft's Copilot for M365 has been repeatedly criticized by analysts for low seat penetration rates. The narrative is shifting from "AI revolution" to "AI integration," a period where the market punishes companies that grow revenue without corresponding proof of net dollar retention.
For the crypto market, this narrative shift has two effects. First, it reduces the congestion in the pipeline of "risk-on" capital. Traditional tech funds that previously allocated 5% to speculative AI software names will rotate into cash or into assets with a deterministic value proposition. Bitcoin, increasingly correlated with the NASDAQ but with a capped supply, becomes a candidate. Second, it creates a positive cost side effect: lower AI hardware prices. If enterprise AI budgets tighten, hyperscale cloud providers will have reduced pricing power. This means inference costs for decentralized AI networks like Bittensor or Render could compress, potentially improving their margins and making them more competitive against Web2 AI services.
Do not mistake this for a blanket endorsement of AI crypto projects. Most of them are liabilities. The filter should be the same as for enterprise AI: can they demonstrate measurable unit economics? A decentralized GPU network can show cost per node and utilization rates. A decentralized AI training market can show throughput and validation metrics. Those are the areas where institutional due diligence can align with crypto-native transparency. The projects that fail this filter will sink. The projects that pass will be the first to receive a constructive capital allocation from funds like my own — after we have run their financial models and scrutinized their custody.
The ETF experience taught me that institutional entrance is always preceded by a governance standard. The spot Bitcoin ETF approval in 2024 happened because the SEC finally saw a surveillance-sharing framework that could detect market manipulation. The SEC's regulation-by-enforcement approach was never about ignorance; it was about leveraging a demand for proof. The same dynamic is now playing out in AI. The KPMG report is essentially a regulation-by-market mechanism. It is a demand for proof, enforced not by a government but by the collective conscience of CFOs.
Contrarian: The Bearish Narrative Is the Bullish Alpha
The consensus read on KPMG's "only 7%" statistic is bearish for AI and, by extension, bearish for any assets attached to AI. I reject this read. The contrarian position is that this data point marks the beginning of a sustainable growth phase, not its end. We have seen this transition before in the crypto market. In 2022, when Terra collapsed and FTX failed, the narrative was that DeFi was dead. Those who read the liquidity map instead of the headlines understood that the withdrawals were a purification ritual. By early 2023, yield became harvestable only in the most rigorous protocols. The subsequent recovery was not a revival of the old hyped dApps, but a maturation of the underlying infrastructure.
The same purification is occurring in AI. The 93% who cannot prove ROI are not failures; they are early-stage adopters who have not yet built the measurement machinery. The KPMG data is the wake-up call that will trigger procurement of AI value management tools. This is not a market contraction. It is a market reorganization. The capex dollars will not disappear. They will shift from "category momentum" to "specific application," from "technology experiments" to "process engines." This is the exact pattern I observed in the ICO era. In 2017, I mapped the capital flows of the top 50 ICOs and found that 60% of successful launches were preceded by whale accumulation on Ethereum. The signal was not the whitepaper; it was the wallet activity. Similarly, the signal in enterprise AI today is not the model; it is the influx of demand for "AI ROI" consulting contracts at Big Four firms.
There is a second contrarian angle: the timing. KPMG released this report during a period when the U.S. corporate budget cycle is in its validation phase. The CFOs who cannot prove ROI are now going to be held to a higher standard in the next budget cycle. This will cause a flight to quality. In practice, this means that companies with high-quality, quantifiable AI use cases (code generation, customer service automation, document processing) will double down on their AI spend. They will not retreat. They will concentrate. The same logic applies to crypto. The projects that will survive are those that directly reduce an enterprise's cost base or increase its revenue with measurable confidence. This is why I have been allocating a significant portion of my fund into a particular type of token: not the flashy AI protocols, but the infrastructure tokens that enable the computation itself. The "picks and shovels" of the proof economy.
The alpha hides in the variance others ignore. The variance here is the difference between the way a CFO in the financial sector measures ROI (strict payback periods) and the way a CFO in the entertainment sector measures ROI (strategic optionality). KPMG's data lumps these together, obscuring the fact that some industries have already figured out how to prove value, while others are still in the discovery phase. My 2017 ICO liquidity mapping taught me to look past the aggregate number and into the sector-specific behaviors. The financial sector, driven by regulatory scrutiny, is likely to be among the first to develop AI ROI frameworks. The healthcare sector, with its long approval cycles, will lag. This creates a roadmap for capital allocation: bet on AI tickers in finance-first, healthcare-later.
The most overlooked consequence of the KPMG report is in the labor market. If corporations cannot prove AI ROI, they will be reluctant to fire the human workers who perform the tasks AI was supposed to automate. This means that AI adoption will not immediately translate into labor cost savings, further delaying AI ROI. But this is a temporary state. As measurement frameworks improve (and they will, because a million consultants will be paid to create them), the hiring freezes will begin. When that happens, the productivity edge of AI will be undeniable, and the ROI will be instantly demonstrable in headcount reductions. This second wave of AI integration is the wave that will truly eat the world.
Takeaway: Cycle Positioning and the Next Window
We have seen this cycle before. First comes exploration, then comes evaluation, then comes regulation, and only then does true industrialization arrive. The KPMG report marks the midpoint of the evaluation phase. In the crypto market, this is analogous to the post-DeFi-Summer period of 2020-2021, when protocols had to show actual usage and fees, not just liquidity mining rewards. Those that survived the "Real Yield" revolution are the ones that trade at high multiples today. The same principle applies to AI.
My recommendation is not to fade AI or to fade crypto. It is to reposition. Over the next 18 months, I expect to see a significant expansion in "AI Value Management" as a distinct software category, with a corresponding set of tokens in the Web3 space that attempt to decentralize this measurement process. Enterprises will demand on-chain provenance for their AI training data and inference logs to satisfy their CFOs. The ability to prove to a third party that a model was trained on a particular dataset, that a particular inference run occurred, and that the output led to a specific action, is a cryptographic challenge. It is a challenge that blockchain is uniquely suited to solve.
The current bull market in crypto is not a reflexive AI pump. It is a rational response to a macro environment where central banks are injecting liquidity but the private sector is deleveraging its questionable investments. The KPMG data indicates that the private sector is about to begin a divestment cycle from unproven AI assets, freeing up capital that will eventually seek returns in assets that are provably scarce. Bitcoin, with its fixed issuance and decentralized settlement, is the ultimate proof-of-scarcity asset. Ethereum, with its robust fee market, is the ultimate proof-of-usage asset. These are the proven engines of the digital asset space.
We do not predict the storm; we build the hull. The KPMG report is a storm warning for AI application-layer valuations. But it is also a blueprint for the hull that will carry the next wave of value: infrastructure that measures, proves, and audits. In the quiet of the bear, we counted our coins. In the loud of this bull, we will cost the AI accounts. Prepare for the second half of the cycle, where the 7% who can prove ROI will dictate the terms. They will be the ones who own the metrics. And in the crypto market, the ones who own the metrics are the ones who own the keys. The cycle marches on, disciplined as ever.
