The same bank that helped transform Bitcoin from a peer-to-peer cash system into a Wall Street yield product now wants to commoditize artificial intelligence. Bank of America’s new AI tracking tool, which monitors model intelligence and costs, isn’t a neutral transparency initiative. It’s a liquidity map for the next great narrative—and a signal that the institutional capture of AI has begun.
Chaos is just liquidity waiting for a narrative.
In 2023, the AI landscape was a cacophony of benchmarks, API pricing pages, and fragmented leaderboards. Researchers spoke in MMLU scores, developers in tokens per second, and investors in TAM projections. The market was informationally inefficient. Bank of America—a bank that manages over $3 trillion in assets—saw that inefficiency as an opportunity to impose order. Their tool, described in a Crypto Briefing report, aggregates two key metrics: model intelligence and cost. The intelligence metric likely compiles scores from public benchmarks like MMLU, HumanEval, and MATH, while the cost metric tracks API pricing per million tokens. The result is a standardized comparison across models from OpenAI, Anthropic, Google, and others.
But this is not a neutral research product. It is a strategic asset. Bank of America’s global research division serves thousands of institutional clients. By offering a unified framework to evaluate AI models, the bank positions itself as the gatekeeper of AI investment decisions. The tool is not a standalone SaaS product; it’s a hook—a product that pulls clients deeper into the bank’s ecosystem of trading, investment banking, and asset management. The real revenue doesn’t come from subscription fees. It comes from the commissions, underwriting fees, and advisory mandates that flow when institutions act on the insights the tool provides.
Context: The Fragmented State of AI Evaluation
Before this tool, the closest thing to a standardized AI benchmark was the LMSYS Chatbot Arena, where users blind-tested models and voted. Or the Hugging Face Open LLM Leaderboard, which aggregated scores from a limited set of tasks. Neither was designed for institutional investors. They lacked cost data, financial context, and the credibility of a Wall Street brand. Meanwhile, independent analysts like Artificial Analysis and Vellum provided pricing snapshots, but their audience was technical, not financial.
Bank of America filled a vacuum. They combined the two variables that matter most for enterprise adoption: performance and price. But the simplicity of the tool is also its greatest weakness. Intelligence is not a single number. A model that scores 90% on MMLU may fail catastrophically in a legal document analysis task. Cost is not just the API price; it includes latency, compute, retraining, and the opportunity cost of vendor lock-in. The tool condenses these complexities into a single dashboard, feeding the illusion that AI investment can be reduced to a cost-per-point ratio.
Core: The Crypto AI Angle
For the crypto ecosystem, this tool is both a threat and an opportunity. The threat is that it shifts the narrative of AI value toward centralized, API-based models. Decentralized AI networks—like Bittensor, Akash, or Render—do not neatly fit into the tracker’s framework. Bittensor’s subnetworks don’t produce a single model score; they produce a dynamic market of intelligence. Akash’s compute marketplace doesn’t have a fixed API price; it’s a spot market for GPU time. The tracker’s metrics implicitly favor the walled gardens of OpenAI and Google, where centralized pricing and performance are easy to quantify.
The opportunity is that the tracker exposes the vulnerability of those centralized models. If Bank of America’s tool highlights a smaller, cheaper model that performs nearly as well as GPT-4, it will accelerate adoption of alternative models—including open-source ones. In the long run, commoditization of intelligence drives down the unit economics of AI, which benefits decentralized networks that can offer lower-cost, trustless inference. But only if those networks can prove their reliability outside the tracker’s metrics.
Based on my experience auditing cross-chain liquidity flows during the 2020 DeFi summer, I’ve seen how institutional tools create self-fulfilling prophecies. When a bank publishes a benchmark, capital flows toward the assets that rank highest—regardless of underlying fundamentals. The same will happen here. The models that score well on the Bank of America index will attract investment, partnerships, and talent. The ones that don’t will be starved of capital, even if they are technically superior.
This is not a neutral evaluation. It is a vector of narrative control. The tracker’s choice of benchmarks determines which capabilities are valued. If it weights reasoning over creativity, OpenAI’s o1 will score higher than Anthropic’s Claude. If it weights cost, Google’s Gemini will dominate. The bank becomes the arbiter of what “intelligence” means in the context of investment.
Contrarian: The Decoupling Thesis
The conventional wisdom is that this tool will democratize AI evaluation. It will help small companies choose the right model, empower investors to make informed decisions, and drive down costs through transparency. I disagree. The tool is a centralizing force that will entrench the largest players.
Value is the illusion we agree to sustain.
Consider the parallel with Bitcoin. After the 2024 ETF approvals, Bitcoin became a Wall Street product. Its price now correlates more with macro liquidity cycles than with on-chain activity. The narrative of “peer-to-peer electronic cash” died. The same fate awaits AI. The tracker will create a new set of metrics that institutions trust, and the metrics will become the reality. Decentralized AI projects that don’t fit the framework will be dismissed as “too risky” or “unproven.” The tracker’s “cost” metric will ignore the real cost of centralized control: censorship, data extraction, and single points of failure.
But there is a contrarian opportunity. The very act of centralizing evaluation creates a target. If Bank of America’s tool becomes the standard, any flaw in its methodology—a misweighted benchmark, a stale price, a model that hacks the leaderboard—will crash the credibility of the entire evaluation framework. The crypto AI projects that survive will be those that can prove their value through mechanisms that the tracker cannot measure: censorship resistance, verifiable inference, and community ownership.
Takeaway: Positioning for the Cycle
The Bank of America AI tracker is a signal that the market is maturing. But maturity in Wall Street’s language means standardization, commoditization, and control. For crypto investors, the question is not whether your project scores high on the tracker. It’s whether your project can operate entirely outside its purview.
History doesn’t repeat, but it does rhyme.
In the early 2020s, DeFi protocols that chased TVL through liquidity mining died when the subsidies stopped. The protocols that survived built real utility and sustainable revenue. The same logic applies here. AI projects that optimize for the tracker’s metrics will end up as centralized products with a decentralized veneer. The projects that ignore the tracker and focus on untrackable value—like Bittensor’s incentive alignment, or Akash’s permissionless compute—will be the ones that endure the next bear market.
Liquidity is the only truth in a world of noise.
The tracker will create noise. It will generate headlines about model rankings, cost curves, and investment flows. But the truth remains: institutional tools are not neutral. They are lenses that shape what we see. The Bank of America AI tracker is a lens that focuses on cost and performance, while blurring out sovereignty, resilience, and trust. As an analyst who has spent years watching Wall Street repackage crypto as a yield product, I see the pattern. The question is whether the AI community will repeat the same mistakes.
From my cabin in the Bohemian Switzerland National Park, where I retreated during the 2022 bear market, I learned that the best investments are the ones that don’t need a CUSIP. The same applies to AI models. The best models are not the ones that top the Bank of America chart. They are the ones that can run on a Raspberry Pi, or in a permissionless network, without asking for permission.
That is the real intelligence. And it cannot be tracked.