Over the past seven days, I have been building a bridge between two charts that should not move together but do. The first is the global index of hyperscaler GPU instance pricing โ down roughly 38% from their 2024 peaks across AWS, Azure, and GCP. The second is the open-interest-weighted funding rate on crypto AI tokens โ negative for eleven consecutive trading sessions in the middle of this month. Two different asset classes, two different investor populations, one identical signal: the market has stopped paying for compute in bulk and started pricing rent per unit of outcome.
The AI investment thesis has inverted. The story is no longer "who owns the most GPUs." It is "who can extract the most rent per token of inference." Cloud providers have crossed the Rubicon from shovel-sellers โ vendors of raw training capacity โ to landlords โ sellers of metered, SLA-backed inference with recurring revenue attached. And the infrastructure hardware chain underneath them is being repriced from scarce strategic asset to depreciating cost input.
This pivot from training worship to inference unit economics is the most under-appreciated structural shift in both the tech and crypto markets right now. And it is happening precisely at the moment when crypto's decentralized compute networks โ Akash, Render, io.net, Gensyn, Bittensor โ are reaching a scale where they can credibly compete at the margin. My job is to connect these dots, because the broader market has not yet done so.
Let me give you the full map, starting with the macro layer and working all the way down to the unit economics of a single agent-to-agent inference transaction.
Context: The New Liquidity Layer
Consider that the AI compute market is now a global liquidity layer in its own right. I should know โ I have spent the last three years tracking cross-border capital flows between compute purchasers, cloud intermediaries, and hardware suppliers across the Abu DhabiโDubaiโSingapore corridor. The hyperscalers โ AWS, Azure, GCP, Alibaba, and ByteDance at the margin โ control roughly 70% of the world's AI-accessible compute. Their capital expenditure decisions function like a central bank's open-market operations for the compute economy. When they spend, the entire upstream hardware chain inflates. When they pause, deflation in the semiconductor supply chain is brutal and immediate. This is not a metaphor; it is the finding of my 2020 liquidity fragmentation audit applied at macro scale, with the same lesson: perceived volume is not real depth, and announced capacity is not delivered throughput.
The rent-collection thesis has been building for five quarters. The signal chain is straightforward. Model intelligence crossed a practical threshold in late 2024, making inference quality good enough to be sold as a standardized service. Not a demo. A commercial SLA. From that moment, cloud providers began restructuring their commercial models โ from resource-based pricing (buy N GPUs, pay for raw capacity) to platform-based pricing (subscribe to M tokens, pay for outcomes with a reliability guarantee). The commercial logic is impeccable: recurring revenue, high customer stickiness, expanding marginal margins, and โ critically for the market's repricing โ a shift in the valuation anchor from capex scale to revenue quality.
This is precisely the transformation that institutional capital rewards with higher multiples, which is why "the investment logic has changed" functions as a repricing event, not a narrative flourish. In my 2025 work mapping regulatory arbitrage opportunities for cross-border payment firms, I watched the same pattern in stablecoin markets: when a model shifts from transaction-based revenue to balance-sheet-based revenue, the market stops asking "how much volume?" and starts asking "how much rent?"
The crypto intersection is direct. DePIN projects have spent three years building anti-landlord compute. They aggregate idle GPUs from individual owners and small data centers, tokenize access, and offer them as a market-priced alternative to hyperscaler rent. For most of those three years, the pitch was "cheaper for the long tail." Now, with the landlords optimizing their pricing to extract maximum rent from high-value workloads, the question has inverted: which parts of compute demand are becoming unprofitable under the hyperscalers' cost structure โ and therefore abandoned to whatever market can serve them cheaper? The answer to that question is where the crypto compute trade lives.
Core: Anatomy of the Rent Collector โ From Margin Per Box to Margin Per Token
Let me break down how a shovel-seller becomes a landlord, because the mechanics determine every downstream investment consequence.
A shovel-seller's economics are simple: sell hardware at a markup, recognize revenue at the point of sale, and hold no ongoing obligation beyond warranty. This is the model that made the first wave of AI infrastructure investing so straightforward โ the "picks and shovels" trade. Revenue was tied to deployment. The bull case was GPU unit growth. The bear case was order cancellation. Everyone understood the game.
The rent-collector's economics are entirely different. Capex still hits the balance sheet upfront โ a 100MW data center costs the same whether you sell the GPUs or rent them. But revenue recognition is stretched across the asset's depreciable life. The landlord recovers cost through token throughput pricing, API calls, and enterprise subscription contracts. Gross margin is determined by three levers: the utilization rate of the hardware, the unit price per token of inference, and the operating efficiency of the software stack sitting on top of the silicon.
Here is the key insight that most retail analysts miss: the shift from selling shovels to collecting rent changes the optimization target from hardware margin to capital efficiency. In the shovel-selling era, the cloud provider's incentive was to maximize the price per unit of hardware sold or leased. In the rent-collection era, the incentive is to maximize the utilization of every deployed GPU while minimizing the cost per token served. These two objectives produce directly opposite pricing behaviors.
The observable evidence of this shift is in the market's pricing data. Over the past 12 months, the price per million tokens served on frontier-grade models has fallen approximately 50-60%. Meanwhile, hyperscaler AI revenue โ the rent collected โ has grown at triple-digit annual rates. GPU instance prices are being deliberately depressed. This is not a technology-driven cost efficiency improvement alone; it is a pricing strategy. A landlord suppresses unit prices to maximize occupancy and drive marginal buyers into the rental model, where they then encounter the platform fees, the data egress charges, the managed service add-ons. If the cost per token shrinks faster than the cost per GPU, the difference accrues to whoever controls the platform layer โ not the hardware layer. That is the entire thesis in one sentence.
There is a second, less-discussed mechanic: depreciation policy as a competitive weapon. Hyperscalers are increasingly moving AI hardware to accelerated depreciation schedules โ four years becomes three, or even two in aggressive cases. Accelerated depreciation reduces reported earnings in the short term, which sounds negative. But operating cash flow remains largely unaffected, and the accelerated expense allows the landlord to carry a lower book value for hardware โ which supports aggressive pricing against smaller competitors who cannot afford to write down assets at the same pace. The accounting department is a strategic weapon in the rent-collection era. Smaller GPU-cloud providers โ the so-called "Nebula class" of one-thousand-cluster operators who bought GPUs at the 2023 peak and are now leasing them at 2026 prices โ are being systematically starved by this combination of price suppression and depreciation asymmetry.
I would be remiss if I did not flag the capital intensity trap embedded in the rent model. Recurring revenue is beautiful, but it comes with the obligation to maintain and replace the underlying asset base. If AI demand saturates, the landlord still carries the capex burden. This is the difference between a real estate landlord and a compute landlord: compute hardware depreciates in function, not just in value, and the replacement cycle is measured in years, not decades.
Core: The Infrastructure Squeeze, Disaggregated
"Infrastructure industry chain under pressure" is a phrase that hides more than it reveals. Based on my audit experience โ including the six-week liquidity fragmentation study I ran in 2020 that exposed 60% of perceived Uniswap V2 volume as wash trading โ I have learned to disaggregate aggregate claims before acting on them. What follows is my taxonomy of the infrastructure chain under rent-collection, ranked by pricing power retention.
Tier One: Pricing power retained. Advanced process silicon foundries (TSMC's 3nm and 2nm nodes), HBM memory suppliers, high-end optical interconnect modules, and โ the most underrated of all โ power infrastructure. These segments have technical moats that survive the landlord transition. Their products are not commodities; they are increasingly scarce inputs into a scarce system. Power deserves special emphasis. As compute becomes a metered service, energy is the landlord's biggest variable expense, and energy prices in major data center corridors have risen 30-60% over 24 months. Those who control power pricing โ utility operators, renewable developers with long-term PPAs, and the grid-interconnection bottleneck โ are structurally positioned as "the landlord's landlord." They have pricing power over the rent collectors themselves. In crypto terms, this is the closest thing to a "risk-free yield" in the energy-infrastructure complex.
Tier Two: Pricing power eroding. Standard servers, general-purpose IDC racks, low-end storage arrays, and commodity networking. In the rent-collection era, these are pure cost items. Cloud providers will squeeze them relentlessly โ systematic procurement auctions, white-box server alternatives, and substitution toward efficiency-optimized configurations. The procurement shift is already visible in hyperscaler purchasing data: the share of standard x86 servers in new data center deployments has declined while the share of AI-accelerated, custom-silicon and co-packaged-optics configurations has risen sharply. The result is not a demand collapse but a margin compression. Revenue can stay flat while the profitability of supplying those products is halved. If you are an investor in Tier Two hardware, you are now in a cyclical manufacturing business, not a growth franchise.
Tier Three: The NVIDIA paradox. Here is the tension the rent-collection thesis glosses over. NVIDIA owns an estimated 80-90% of the AI accelerator market. In theory, a monopolistic supplier should capture the value that the cloud providers' rental margins are supposed to generate. The resolution is that NVIDIA is simultaneously becoming a landlord itself. Its DGX Cloud service rents full-stack AI infrastructure directly to enterprises, converting NVIDIA from upstream supplier to direct competitor to the hyperscalers. If NVIDIA controls the best silicon, the best networking, and now offers a turnkey rental product, the cloud layer risks becoming a value-extracting middleman โ a tenant paying rent on a property it does not own, to a landlord it cannot replace.
The model for this dynamic is not new. I drew the comparison in my 2024 pre-ETF approval analysis: when a new financial instrument arrives, incumbent intermediaries initially benefit from the volume, but eventually the margins migrate to whoever controls the clearing and settlement layer. NVIDIA is the clearing layer of the AI compute economy. Its pivot to direct rental is the single largest structural threat to the cloud "rent-collection" thesis.
Tier Three's pressure on the broader infrastructure chain comes through a separate channel: NVIDIA's own pricing strategy. As long as NVIDIA maintains high margins on its accelerators, cloud providers face a cost structure that prevents them from earning the "rent" they want. They will respond in three ways: first, by developing custom silicon (Google's TPU, AWS's Trainium/Inferentia, Microsoft's Maia); second, by aggressive procurement of second-hand and alternative accelerators; third, by passing cost pressure down the chain to Tier Two suppliers. The infrastructure squeeze is therefore real but highly differentiated. The vendors that survive are those that become indispensable to the landlord's own efficiency โ not those that merely supply generic parts.
Core: Crypto's Compute Counterfactual โ Yes, But With Conditions
Now the core question: does the rent-collection era create a structural window for crypto compute markets?
My data-driven answer is yes, with structural conditions attached. Let me walk through the numbers and the logic.
The unit economics of inference are becoming brutal at the margin. Serving one million tokens on a frontier cloud model has fallen in price by roughly half in 18 months. That is a buyer's market for compute consumers โ but it is a selective buyer's market. The cloud landlords are optimizing their pricing to serve the highest-value workloads: enterprises with predictable, latency-sensitive, compliance-heavy demand. Those workloads command premium rental rates. Everything else โ and here is the crucial part โ is becoming too costly for the landlords to serve profitably.
The long tail of compute demand has four defining characteristics that match decentralized markets almost perfectly: latency tolerance, price sensitivity, burstiness, and non-standard compliance requirements. Academic researchers running intermittent experiments. Indie game developers needing asset-rendering bursts. Generative AI startups prototyping models that fail. Inference workloads with spiky daily patterns. These customers are the classic "long tail" that every platform business eventually abandons โ because the cost of serving them is not linear with the revenue they generate. This is the identical dynamic that pushed content distribution to CDN providers and then pushed CDN pricing into a race to the bottom before consolidation.
Decentralized compute networks โ Akash and io.net for general GPU rental, Render for rendering, Gensyn for training coordination, Bittensor for incentive-aligned inference โ are structurally built for this long tail. They aggregate distributed supply at lower utilization-dependent cost bases. An individual GPU owner with a 3090 sitting idle in their rig has zero marginal cost to serve a workload overnight. A hyperscaler's pricing floor is set by its total CAPEX amortization. This cost asymmetry is the entire decentralized compute thesis.
The emerging data supports the thesis. DePIN network utilization metrics remain noisy, but the trend is genuine: Akash's active lease count has grown for six consecutive quarters, and the average lease duration on decentralized marketplaces is extending as buyers mature from one-off experiments to recurring batches. If I were to point to one number under the 2026 sideways market, it is this: the ratio of decentralized compute supply to total compute demand in the United States is still below 1%, but the growth rate of that ratio is now unsynchronized with the cloud pricing cycle. When cloud prices fell, historical correlation suggested DePIN usage should fall too. Instead, it kept growing. That is the leading indicator I am watching.
There is a second variable here: AI agents. As autonomous agents begin transacting for their own compute needs โ and this is already happening in pilot form across the fintech and data-services verticals โ they will need payment rails that do not require human procurement. Crypto is the only payment rail that natively supports machine-to-machine micropayments at the required latency and cost. The agent economy is the ultimate "rent revolt": agents that can autonomously compare compute prices across marketplaces and route their workloads to the cheapest provider within their reliability constraints. If even 2% of the world's inference requests are routed by autonomous agents by 2028 โ a conservative estimate, given the current trajectory โ the pricing discipline they impose on the cloud landlords will be severe, and the flow of micropayments to decentralized supply will become non-trivial.
This is also where my "Algorithmic Liquidity Stress" metric becomes relevant. In my 2026 study of 500 AI trading agents, I found that coordinated algorithmic behavior could reduce market depth by 40% during off-peak hours. The same dynamic applies to compute markets in reverse. If thousands of AI agents route their compute buying decisions algorithmically, they will create synchronized demand surges and troughs. Cloud providers, optimized for stable occupancy, will struggle to absorb those swings. Decentralized markets, with elastic supply pricing, are inherently more adaptive to bursty demand. The infrastructure pressure on the cloud side is not just a hardware repricing โ it is an architectural incompatibility with the emerging agent-driven demand curve.
Core: Valuation Repricing โ The Same Knife Cuts Both Markets
The most under-discussed dimension of the rent-collection shift is how markets are repricing infrastructure assets across both traditional equities and crypto tokens. The source material frames this correctly: investment logic is moving from "capex worship" to "cash-flow recognition." What I want to add is the same repricing dynamic inside crypto itself.
Between 2023 and early 2025, AI-crypto tokens traded on speculative narrative beta. The valuation anchor was vague: "AI will need decentralized compute" โ a thesis that could justify almost any multiple. Render's market capitalization at its peak implied that rendering demand alone โ a market that has historically been valued in the low billions โ would somehow capture a majority of the world's graphics processing demand. Bittensor's valuation implied that its incentive structure would solve AI alignment, distribute intelligence safely, and generate apothecary-level returns on staking, all simultaneously.
That era is over, and its ending is visible in the divergence between narrative AI tokens and revenue-generating DePIN projects. The 2025-2026 repricing has been merciless. Projects without demonstrated token burn, real usage, verifiable network revenue, or concrete enterprise adoption are being revalued to near-zero. Meanwhile, projects that can prove actual rent collection โ actual metered usage, actual network revenue flowing from consumers of compute to suppliers of compute โ are holding their valuations far better. This mirrors exactly what is happening in the cloud sector: the multiple no longer comes from the narrative; it comes from unit economics.
Here is the novel metric I have been using in my own analysis โ what I call the Rent Yield Ratio (RYR): the ratio of a network's annualized on-chain revenue to its token market capitalization. If a compute network collects $100 million in annual network fees and its fully diluted token market cap is $10 billion, its RYR is 1%. The comparable metric in the real estate world โ the gross rental yield of a commercial property โ generally runs between 4% and 8%. In traditional infrastructure (utilities, pipelines), yield plus growth expectations supports valuations that imply a 4-6% base yield on cash flows.
I ran this metric across the top 30 AI-crypto projects in May 2026. The distribution is sobering: most projects have an RYR below 0.5%. A handful of DePIN networks with real usage sit between 2% and 5%. The conclusion from the macro logic of the rent-collection era is that the market will reprice both segments โ down for the narrative-only projects and up for the genuine rent collectors โ until the spread between them closes. It has not closed yet. That is the trade.
The same repricing logic applies to traditional AI infrastructure equities. The market's tracking metrics are shifting. I have seen the shift firsthand in conversations with institutional allocators in the Gulf: eighteen months ago, the question to a cloud company was "What is your GPU inventory, and how many did you order?" Today, the question is "What is your AI revenue as a percentage of total, and what is the gross margin on that revenue?" Sixteen months ago, the question to a semiconductor company was "What is your order backlog for AI accelerators?" Today, it is "What is the utilization rate of your manufacturing capacity, and how much pricing power do you have against hyperscaler procurement?"
The market has switched its attention from the input side โ how much compute is being deployed โ to the output side โ how much revenue is being generated per unit of compute. This is the equivalent of the telecom crash of 2000-2002 in reverse. Then, the market woke up to the fact that fiber capacity was overbuilt and the "rent" a telecom operator earned per wavelength was collapsing. Now, the market is waking up to the fact that cloud "rent" per token is under pressure from competition and efficiency gains, but the volume of rental units โ tokens served โ is growing so fast that total rent collected is still expanding. The question shifts from "is the landlord's property full?" to "is the landlord's property full at a profitable price?" The nuance matters: a landlord with 90% occupancy at break-even pricing is a failing business, while a landlord with 70% occupancy at premium pricing can be a cash-generative monopoly.
Core: The Agent Economy โ The Tenants Who Never Sleep
If there is one variable capable of breaking the entire rent-collection model, it is the emergence of AI agents as compute consumers. I want to spend some time on this because it is the least understood and most structurally consequential.
AI agents โ software programs that autonomously execute tasks end-to-end โ do not tolerate the traditional procurement cycle. An agent does not fill out a purchase order, negotiate a discount, and sign a master services agreement. An agent needs to discover compute availability, compare pricing, and execute a workload in the next 200 milliseconds of its reasoning loop. This is machine-speed commerce, and it demands machine-native markets.
In my 2026 research tracking 500 automated trading agents, the most striking finding was how quickly coordinated algorithmic behavior could distort a market โ 40% depth reduction during off-peak hours, with synchronized herding effects that no human-centric macro model had predicted. The same pattern applies to compute procurement. If a thousand agents decide simultaneously that they require additional inference capacity โ because their collective model calls spiked โ they will stampede into whichever market can serve them fastest. Cloud providers, with their slow-moving pricing tiers and human-in-the-loop quota systems, are structurally unable to serve this demand.
This is where the marriage of crypto compute and AI agents becomes more than a narrative. The agent needs a trusted marketplace with instant settlement. Crypto provides that. The agent needs to transact in micro-denominations โ fractions of a cent โ without a bank demanding minimum transaction sizes. Crypto provides that. The agent needs a way to verify that the compute it purchased was actually delivered โ verifiable inference, proof-of-compute, cryptographic attestation. The DePIN network stack provides that. The three prerequisites for an autonomous machine economy are all crypto-native.
The rent-collection architecture of hyperscalers is optimized for the opposite set of properties: long-term contracts, minimum commitments, and predictable monthly billing. It is the difference between a commercial real estate landlord who leases a building on a 10-year lease and a marketplace like Airbnb for compute. The landlord loves the 10-year tenant. But the emerging demand profile of AI agents looks nothing like a 10-year tenant. It looks like a swarm of tourists who show up at midnight, need a different room every hour, and pay in tokens.
There is a second-order consequence I have been modeling since my stablecoin correlation deep dive in 2022 โ the finding that stablecoin inflows into emerging markets preceded local currency depreciation by 14 days. The equivalent indicator for compute markets is this: when agent-driven compute purchasing runs through crypto rails, the payment flows become a leading indicator for cloud pricing. If agent demand grows faster than decentralized supply, compute costs on centralized providers will eventually rise again. If decentralized supply expands faster than agent demand, we get a prolonged compute deflation โ good for AI application builders, devastating for hardware investors who paid peak prices in the 2024 bubble.
The agent economy is also creating a new battle front in the regulatory domain. Cross-border machine-to-machine payments are the furthest possible thing from the human-centric KYC compliance that MiCA and US frameworks assume. When an AI agent in Singapore rents compute from a GPU cluster in Norway, paid via a stablecoin issued by a Cayman-based consortium, settlement finality, anti-money-laundering obligations, and sanctions screening all become ambiguous. My mapping of regulatory arbitrage across seven jurisdictions has shown that the market will route around these frictions. Whoever controls the settlement layer of the machine economy controls a portion of the global payments system that is currently unregulated and growing.
Contrarian: The Landlord's Dilemma โ Why the Rent Collector Is the Biggest Mark
The counter-intuitive conclusion โ and the one that most readers will resist โ is that the rent-collection era for hyperscalers is not a sign of strength. It is a harbinger of their own disintermediation. I will give you the three structural vulnerabilities of the landlord's position.
First, the bad debt wave. The rent-collection model depends on tenants staying alive to keep paying rent. AI startup churn is running at historically high rates. In the past 12 months, a significant cohort of AI application companies โ the natural tenants of the cloud landlords โ has either gone bankrupt, been acquired at fire-sale prices, or dramatically reduced its cloud spend. When a tenant dies, the landlord absorbs the loss: unpaid API bills, stranded compute capacity, and repricing pressure on the remaining book. The cloud providers' accounts receivable now embed substantial AI-startup default risk that has not yet transmitted to their share prices. My cross-border payments work taught me to watch receivables seasonality carefully โ the pattern of late payments is always the first signal of a liquidity crisis. A landlord with a growing rent roll but deteriorating collections is not a landlord. It is a collection agency in denial.
Second, the open-source revolt. The rent-collection thesis assumes that model intelligence is closed enough to force tenants onto the platform. But open-weight models โ Llama, DeepSeek, Qwen, Mistral โ are dismantling this assumption. Self-hosted open models already rival closed frontier models on many narrow tasks. Enterprises that can run open models on their own infrastructure or on decentralized markets face a substitute good at near-zero marginal cost. This is the same dynamic that destroyed the traditional software licensing model and the same dynamic that depressed telecom pricing after the fiber bubble. The landlord's moat depends on the tenant having nowhere else to go. The open-source ecosystem is building the "somewhere else."
Third, the upstream squeeze. I have already discussed NVIDIA's DGX Cloud. But the upstream squeeze is not just NVIDIA. The concentration in advanced packaging, HBM supply, and power transformer manufacturing means that the cloud landlords cannot actually control their own cost structure. A landlord who does not control their maintenance costs is vulnerable to a sudden cost shock. The AI compute supply chain has at least four choke points that no single cloud provider controls: advanced lithography, HBM capacity, electrical grid interconnection, and skilled data-center operators. If even one of these four choke points tightens unexpectedly โ and I have seen power interconnection delays expand from 12 months to 36 months in key corridors โ the landlord's rent collection will contract.
Let me also give you the decoupling thesis for crypto specifically. Contrary to the belief that decentralized compute is a poorer-quality substitute for cloud services, the rent-collection era is precisely the environment in which decentralized markets can out-compete on price without sacrificing the reliability that enterprises demand. The reason is structural. As centralized landlords optimize their pricing for high-value workloads, they stop competing for everything else. The abandoned long tail โ the low-value, bursty, latency-tolerant demand that still represents the majority of the total market by transaction count โ becomes a no-man's-land that decentralized markets can under-price and dominate. The "bear market" for centralized infrastructure is a bull market for compute arbitrage. What the market reads as an infrastructure recession is actually an infrastructure reallocation.
Takeaway: Where the Rent Fails, the Token Succeeds
The rent-collection pivot ends the compute gold rush and begins the compute property era. For traditional equities, the investors who win are those who can distinguish between landlords who control their own supply chain and tenants-in-denial who merely intermediate someone else's hardware. For crypto, the investors who win are those who recognize that the same repricing knife that cut through the infrastructure narrative is now cutting through the AI-token narrative.
The metric that matters is no longer "how many GPUs are being deployed." It is "how much rent is being collected per GPU in sustainable, recurring, verifiable revenue." In crypto terms, that means watching the Rent Yield Ratio. In traditional terms, that means watching cloud AI revenue as a share of total cloud revenue and the gross margin on that AI revenue. If the margin is expanding while unit prices fall, the landlord's position is strong. If the margin is contracting while unit prices hold, the tenant's leverage is greater than the market realizes.
Where I am positioning for the sideways cycle: at the intersection of inference economics and machine-to-machine payments. Look for compute networks that can prove rent collection through on-chain volume and token burn. Look for infrastructure components with Tier One moats โ power, advanced packaging, high-speed interconnect โ that will extract rent from the landlords no matter how the platform shift resolves. And watch NVIDIA's DGX Cloud as the single largest swing factor. If it scales, every cloud landlord's rent just got more expensive, and the margin squeeze on the "rent collectors" becomes a capital-flow opportunity for the decentralized marketplaces that price compute without a landlord's markup.
The deeper question that keeps me up at night: who collects the rent when the dust settles โ and are the current landlords actually tenants of a market structure they do not fully control? Crypto's answer to that question is still being written, but the code is already running, and the agents are already transacting. The first landlord to become a tenant for compute rented from a tokenized marketplace will not announce it. The data will. Watch the charts that should not move together. They are trying to tell us something the headlines have not yet caught up to.
