WALL STREET IS COUNTING EVERYTHING AGAIN—EXCEPT THE CASHOracle CorporationBATS:ORCLTrade-TechniqueThe 2026 numbers investors cannot stop quoting: AI revenue, RPO, backlog, bookings, users and gigawatts—and what each one is actually worth Evidence date: 17 July 2026 Every market cycle develops a favourite number. During one cycle it is users. In another, total addressable market. Then it becomes subscribers, gross merchandise value, orders, capacity or adjusted EBITDA. In 2026, the fashionable numbers are even larger: Oracle has $638 billion of remaining performance obligations. CoreWeave has almost $100 billion of revenue backlog. GE Vernova has $163 billion of backlog. Uber processed $53.7 billion of Gross Bookings in one quarter. Snowflake has $9.21 billion of RPO. Broadcom expects billions of dollars of AI semiconductor revenue every quarter. The numbers are real. The mistake begins when a real operating number is treated as if it were already shareholder value. Operating metric → Revenue → Operating profit → Taxes → Reinvestment → Free cash flow → Risk and probability → Present value That chain is the framework. This time, instead of revisiting NVIDIA, Palantir, Amazon, Meta or Tesla, we will apply it to seven additional US stocks that sit directly inside the hottest 2026 themes. This is not a list of stocks to buy or avoid. It is a test of evidence: how far has each headline metric actually travelled towards free cash flow per share? 1. ORACLE: $638 BILLION OF RPO—AND NEGATIVE $23.7 BILLION OF FREE CASH FLOW Oracle may be the cleanest 2026 example of why backlog and value are not the same thing. At the end of FY2026, Oracle reported: Remaining performance obligations of $638 billion, up 363% year over year FY2026 revenue of $67.4 billion Cloud revenue of $34.0 billion Cloud Infrastructure revenue of $18.1 billion, up 77% GAAP operating income of $20.6 billion Operating cash flow of $32.0 billion Capital expenditure of $55.7 billion Free cash flow of negative $23.7 billion The cloud demand is genuine. Q4 IaaS revenue grew 93%, and management said much of the recent RPO increase came from large AI contracts. But recognizing that RPO will require data centres, GPUs, power and years of execution. Oracle also disclosed a useful detail: $75 billion of its large AI-contract commitments involved hardware prepaid or supplied by customers. That reduces Oracle's capital burden. It does not eliminate the need to examine the timing, margins and financing attached to the rest of the backlog. The bridge currently looks like this: AI contracts → $638B RPO → gradual cloud revenue → operating cash flow → massive data-centre capex → currently negative FCF What matters next: How much RPO converts within 12, 24 and 36 months? What operating margin does the new AI infrastructure earn after depreciation? Does capex growth slow before operating cash flow does? How much funding comes from customers versus Oracle's own balance sheet? Does free cash flow turn positive as commissioned capacity begins generating revenue? Framework verdict: Demand pass; reinvestment bridge under maximum stress. Human version: a signed cloud contract is economically useful, but the data centre must still be built and paid for before the contract becomes shareholder cash. 2. COREWEAVE: THE BACKLOG IS ENORMOUS—SO ARE THE CAPITAL REQUIREMENTS CoreWeave reported Q1 2026 revenue backlog of $99.4 billion and quarterly revenue of $2.08 billion, more than double the prior-year period. It had surpassed 1 GW of active power and reported more than 3.5 GW of contracted power. Those numbers place CoreWeave directly at the centre of AI-infrastructure demand. But its income and cash-flow statements show why backlog cannot be valued in isolation: GAAP operating loss: $144 million Net interest expense: $536 million GAAP net loss: $740 million Adjusted EBITDA: $1.16 billion Cash from operations: $2.98 billion Purchases of property and equipment: $7.70 billion Subtracting reported property purchases from operating cash produces a simplified quarterly cash deficit of roughly $4.7 billion before considering financing flows. The company also had approximately $24.9 billion of current and non-current debt at quarter-end. Why can adjusted EBITDA look strong while GAAP profit and post-capex cash look weak? Because depreciation, financing and infrastructure investment are central—not peripheral—to an AI cloud. The company itself defines backlog broadly enough to include RPO plus other estimated future revenue under committed contracts, subject to delivery and service availability. Therefore, power, GPU supply, construction timing and customer concentration all sit between backlog and recognized revenue. What matters next: Backlog conversion rather than backlog growth alone Revenue produced per active megawatt Utilization after new capacity comes online Interest expense as a percentage of revenue Post-capex cash flow and new financing required per dollar of growth Customer concentration and contract protections Framework verdict: Revenue conversion visible; equity-cash bridge incomplete and financing-dependent. Human version: the demand may be extraordinary, but lenders, equipment suppliers and construction spending reach the cash flow before common shareholders do. 3. BROADCOM: WHEN THE AI NUMBER ALREADY REACHES FREE CASH FLOW Broadcom offers the opposite comparison. In Q2 FY2026 it reported: Total revenue of $22.19 billion, up 48% AI semiconductor revenue of $10.8 billion, up 143% GAAP net income of $9.31 billion Adjusted EBITDA of $15.24 billion Operating cash flow of $10.49 billion Free cash flow of $10.26 billion, or 46% of revenue Management expected Q3 AI semiconductor revenue of approximately $16 billion. More importantly, AI revenue is not merely a pipeline or capacity target. It is already included in reported semiconductor revenue and is accompanied by substantial free cash flow. The bridge is comparatively short: Custom AI accelerators and networking demand → semiconductor revenue → operating profit → operating cash → FCF The remaining risk is not whether AI has started monetizing. It is whether current growth is durable. Custom accelerators can create deep customer relationships, but they can also produce concentration, product-cycle and bargaining-power risk. Investors must also separate AI growth from the infrastructure-software economics acquired with VMware. What matters next: AI revenue growth by accelerator and networking demand Customer concentration and the number of hyperscale programmes GAAP versus adjusted operating economics FCF conversion after integration costs, interest and working capital Whether growth remains broad when individual chip programmes change generation Framework verdict: Strong operational and cash-flow pass. Human version: Broadcom's AI claim is easier to audit because the story has already arrived in the cash-flow statement. 4. SNOWFLAKE: RPO AND RETENTION LOOK STRONG, BUT DILUTION CANNOT DISAPPEAR Snowflake reported Q1 FY2027: Product revenue of $1.33 billion, up 34% Net revenue retention of 126% Remaining performance obligations of $9.21 billion, up 38% 779 customers generating more than $1 million of trailing product revenue GAAP operating loss of $326 million Free cash flow of $233 million Adjusted free cash flow of $266 million RPO indicates contracted future business, while 126% net revenue retention means the existing customer base is spending more after churn and contraction. Both are valuable signals. But Snowflake uses a consumption model. A contract does not guarantee that revenue will arrive evenly, and optimization by customers can affect usage. More importantly, Q1 stock-based compensation was approximately $402 million—larger than reported free cash flow. SBC is non-cash in the quarter, but not economically free. If it increases diluted shares, part of the company's future cash flow belongs to additional shares. RPO and consumption → product revenue → GAAP operating result → operating cash → FCF → divide by diluted shares What matters next: Product-revenue growth relative to RPO growth Net revenue retention without excessive discounts GAAP operating-loss improvement SBC as a percentage of revenue FCF per diluted share—not only company-level adjusted FCF AI workloads that create paid consumption rather than demonstrations Framework verdict: Commercial bridge passes; GAAP profitability and dilution remain the missing pieces. Human version: cash flow is more valuable when the same shareholders still own roughly the same percentage of it. 5. GE VERNOVA: THE POWER BACKLOG IS STARTING TO SHOW MARGINS The AI buildout is not only a semiconductor story. Data centres need turbines, transformers, grid equipment and reliable electricity. GE Vernova's Q1 2026 numbers included: Orders of $18.3 billion, up 71% organically Total backlog of $163 billion Gas Power backlog and slot reservations of 100 GW Revenue of $9.3 billion Adjusted EBITDA of $0.9 billion, with a 9.6% margin Free cash flow of $4.8 billion Raised FY2026 FCF guidance of $6.5–7.5 billion This is a stronger backlog bridge than a company that has not yet demonstrated execution. Revenue, adjusted margin and cash generation are all moving in the right direction. Still, two adjustments matter. First, reported Q1 net income included $4.5 billion of pre-tax M&A gains, primarily related to Prolec GE; that should not be treated as recurring operating profit. Second, quarterly free cash flow can be affected by customer advances and working-capital timing in long-cycle businesses. What matters next: Backlog conversion schedule by equipment and services Price, cost and warranty assumptions on long-duration contracts Adjusted EBITDA margin as deliveries increase Customer advances versus sustainable cash generation Service revenue attached to the installed equipment base Whether data-centre demand adds capacity without damaging project discipline Framework verdict: Strong backlog-to-revenue bridge; normalize gains and working capital before valuing cash flow. Human version: a turbine slot has more value when the manufacturer can deliver it at an attractive margin and later earn service revenue from the installed base. 6. UBER: $53.7 BILLION OF BOOKINGS DOES NOT BELONG TO UBER Uber completed 3.64 billion trips in Q1 2026 and reported 199 million monthly active platform consumers. Gross Bookings reached $53.7 billion. But Gross Bookings represents the total value moving through the platform. Drivers, couriers, merchants, taxes and other parties receive much of that amount. Uber itself reported: Revenue of $13.2 billion GAAP operating income of $1.92 billion Adjusted EBITDA of $2.5 billion Operating cash flow of $2.4 billion Free cash flow of $2.3 billion That gives us a visible conversion chain: Consumers and trips → Gross Bookings → Uber's revenue/take → operating income → FCF Uber One had reached 50 million members, with members driving half of Mobility and Delivery Gross Bookings. That can improve frequency and retention, but investors should still test the cost of membership benefits and incentives. Autonomous vehicles add a separate layer. A partnership-based, capital-efficient AV strategy can expand the network without Uber owning every vehicle. But future AV value should depend on actual paid trips, utilization, insurance/liability economics and the share retained by Uber and its partners. What matters next: Gross Bookings growth versus revenue growth Operating income and FCF as percentages of bookings Insurance, incentives and driver/courier economics Uber One retention and incremental profitability Paid autonomous trips and economics per trip—not announcements alone Framework verdict: Strong marketplace conversion; AV optionality remains probability-weighted. Human version: the money passing through the app is not the money Uber keeps. 7. APPLOVIN: THE AI ADVERTISING STORY ALREADY PRODUCES CASH—TRANSPARENCY IS THE NEXT TEST AppLovin's Q1 2026 results were financially powerful: Revenue of $1.84 billion, up 59% Net income of $1.21 billion Adjusted EBITDA of $1.56 billion Operating cash flow of approximately $1.3 billion Free cash flow of approximately $1.3 billion Unlike an AI company selling only a future vision, AppLovin's advertising technology is already producing revenue, profit and cash. The analytical challenge is different: ad platforms are less transparent than order-book businesses. Investors need evidence that improved advertiser outcomes—not temporary pricing, customer mix or traffic-acquisition choices—drive the economics. What matters next: Revenue growth alongside advertiser retention and diversification Incremental revenue converted into GAAP operating profit Difference between adjusted EBITDA and GAAP earnings Traffic-acquisition, platform and privacy/regulatory risks Durability of FCF after taxes, working capital and capital returns Framework verdict: Strong reported cash bridge; durability and transparency require continued testing. Human version: an algorithm can be valuable without revealing every detail, but investors still need repeatable financial evidence that the advantage is durable. THE 2026 EVIDENCE SCORECARD These scores grade the completeness of the metric-to-cash evidence. They do not say whether a share is cheap or expensive. Broadcom — 9/10: AI revenue already converts into substantial FCF Uber — 8/10: bookings-to-revenue-to-FCF bridge is visible GE Vernova — 8/10: backlog conversion and margins improving; normalize working capital and gains AppLovin — 8/10: strong profit and cash; durability is harder to observe externally Snowflake — 7/10: strong commercial metrics and positive FCF; GAAP loss and SBC remain material Oracle — 6/10: extraordinary contracted demand; capex currently overwhelms operating cash CoreWeave — 4/10: revenue growth is real; debt, interest and construction consume the economics THE PRACTICAL TEST: FIVE NUMBERS TO WRITE DOWN BEFORE EVERY EARNINGS CALL For any stock promoted using AI revenue, backlog, RPO, bookings or capacity, record these five figures before reading management commentary: Conversion ratio: recognized revenue divided by the relevant operating metric over a sensible period GAAP operating margin: after normal operating costs, before unusual investment gains Reinvestment rate: capex plus working-capital needs relative to operating cash Financing burden: interest and net debt relative to revenue and cash generation Per-share result: normalized FCF divided by the diluted share count Then write one sentence that can be proven wrong: “If the metric is genuinely creating value, revenue conversion, normalized margins and free cash flow per share should improve by ______ within ______ quarters.” If a thesis cannot be expressed with a measurable deadline, it may be a story rather than an analysis. FINAL CONCLUSION The hottest US-stock numbers of 2026 are not imaginary. AI revenue is growing. Cloud contracts are being signed. Power equipment is being reserved. Consumers are completing more trips. Enterprises are consuming more data. But the framework reveals that these companies occupy very different economic positions. Broadcom's AI number already reaches free cash flow. Uber's bookings pass through a visible take-rate and profit bridge. GE Vernova's backlog is converting with improving margins. Snowflake produces cash but must account for dilution. Oracle has extraordinary demand while current capex overwhelms cash generation. CoreWeave shows most clearly that backlog, EBITDA and equity cash can tell three different stories. A huge number is not automatically a valuable number. Its value depends on how much reaches free cash flow per share, when it arrives, and what must be risked to produce it. That is the discipline for 2026: count the contracts, bookings, users and gigawatts—but finish by counting the cash.