Oracle found cash. Has AI found returns?

iPrompt Signals

AI & robotics investing, explained so you can actually act on it.

ISSUE 28 · FRIDAY 11 SEPTEMBER 2026 ·

Last Friday we asked who finances the AI buyer. Oracle’s results bring the question closer to shareholders: how much of the buildout can customers fund before the service is delivered? Reuters reports $11.36bn of customer prepayments alongside $28.50bn of quarterly capital expenditure. Free cash flow was still negative $5.40bn.

Reuters · Oracle results · 10 Sep

My read: advance payments strengthen the funding case, but they leave a delivery obligation behind. Before raising an AI position, I want to know what cash has arrived, what the company owes in return, and whether the contract still earns an attractive return after the equipment needs replacing.

WHAT TO DO WITH IT

Add a funding line to the thesis for your largest AI holding. Separate cash from customers, cash from lenders or shareholders, and promises that only become payable later. Then write down the evidence that would justify a higher valuation. A bigger backlog alone doesn’t answer that question.

WEEKLY SCOREBOARD

Thursday 10 September regular-session close. Changes below are for Thursday only, not the full week. Oracle and Adobe reported after the close.

MarketWatch · Thursday semiconductor closing prices · 10 Sep | MarketWatch · Adobe closing price · 10 Sep | Associated Press · Thursday market close · 10 Sep

After hours: Oracle rose and Adobe fell in initial earnings coverage. Those reactions are separate from the closing prices above and are not Friday’s result.

Barron’s · Oracle earnings and trading reaction · 10 Sep | Reuters · Adobe results and outlook · 10 Sep

TOP HEADLINES

1 Oracle improves the funding case

Fiscal Q1 revenue rose 30% to about $19.3bn, while cloud infrastructure revenue reached $7.4bn, up 121%. Remaining performance obligations reached $664bn. Oracle kept its annual capital expenditure outlook at $90bn–$95bn. The scale of contracted demand is considerable; so is the capital required to serve it.

WSJ · Oracle results and outlook · 10 Sep | Business Insider · Oracle capital spending · 10 Sep

The implication: customer-funded construction can reduce Oracle’s need to finance every dollar itself. It cannot tell us the lifetime margin on the resulting contracts. The next useful disclosure is a bridge from gross investment to Oracle’s own cash outlay, together with the terms attached to customer funding.

2 TSMC gives the chip cycle another strong reading

August revenue was NT$514.81bn, up 53.3% year on year and 10.1% from July. Revenue for January through August rose 39.3%. These are company-wide sales in Taiwan dollars, not a separate measure of AI revenue.

TSMC · August revenue release · 10 Sep

The implication: this supports the case that semiconductor demand is reaching manufacturing revenue. It doesn’t establish how much comes from price, mix or volume, or what returns shareholders will earn on new fabs. TSMC remains a useful way to study the economics shared by competing chip designers.

3 Adobe makes software selection harder

Adobe reported $6.76bn of quarterly revenue, up 13%, and more than 150% growth in AI-first annualised recurring revenue. Its fourth-quarter revenue target is $6.80bn–$6.85bn. Reuters reports that the midpoint came in slightly below consensus.

Adobe · Q3 FY2026 results and financial statements · 10 Sep | Reuters · Adobe results and outlook · 10 Sep

The scale matters: management puts AI-first ending ARR above $650m, compared with $27.50bn of total Adobe ARR. Fast growth in that smaller category cannot yet establish the growth rate of the whole business.

Adobe · Q3 prepared remarks and slides · 10 Sep | Adobe · Q3 FY2026 results and financial statements · 10 Sep

Last week’s favourable software signal should stay company-specific. AI product growth can coexist with pressure elsewhere in an incumbent’s business. For Adobe, track whether adoption produces incremental paid demand after free usage, competing tools and inference costs have taken their share.

THE MACRO UNDERNEATH IT

The ten-year Treasury yield reached 4.95% on Thursday. That raises the return investors can earn elsewhere and makes distant cash flows less valuable today. Strong AI sales and falling AI shares can therefore coexist without either being irrational.

Associated Press · Thursday market close · 10 Sep

This edition is dated before the scheduled 11 September US CPI release at 08:30 New York / 14:30 Bratislava. The FOMC meets on 15–16 September. Neither outcome is assumed here.

BLS · September release calendar | Federal Reserve · FOMC calendar

AI IN OPERATIONS AND ROBOTICS

4 Nvidia and Palantir put a customer inside the story

The companies announced a supply chain collaboration combining Nvidia Nemotron models with Palantir Foundry and AIP. Its first deployment is in Nvidia’s own operations. The announcement describes human experts retaining final control of decisions; it provides no quantified contract value or independently verified return on investment.

Nvidia · Palantir supply chain collaboration · 10 Sep

The implication: this is a more concrete enterprise use case than a generic partnership announcement. Its investment value will depend on repeatability. Watch for unrelated customers buying similar deployments, shorter implementation times and measurable savings after integration costs. A prestigious reference customer is useful evidence, but it isn’t a revenue forecast.

5 Skild offers a commercial robotics signal worth testing

On 9 September, Skild said it had crossed a $100m annual revenue run rate within ten months of its first commercial deployment, with more than 60 paying customers. It also described work with Nvidia and Foxconn on Blackwell assembly. These are company-reported figures. A run rate extrapolates a current pace; it is not $100m of revenue already earned over a year.

Skild AI · Commercial deployment update · 9 Sep

Nvidia’s 10 September account of Skild’s S1 model describes learning new tasks from a video demonstration. The model launched the previous week. Reported test performance includes roughly 66% success at each step on new multistep tasks, not a 66% rate for completing an entire production job.

Nvidia · Skild S1 research and deployment · 10 Sep

The implication: the commercial update is encouraging, while deployment economics remain the test. Ask about paid installations, hours without intervention, installation labour and service costs. Public-market exposure through a supplier is indirect; a successful private robotics company may still be too small to move that supplier’s earnings.

NEW TO INVESTING

There are several stages between an impressive demonstration and shareholder returns. A customer must adopt the product, pay enough to cover delivery and support, then renew or expand. A stock price can already assume success at every stage. Your job is to identify which stage the evidence actually supports.

OUR INVESTING ANGLE

Keep last week’s distinction between pricing power and financing exposure. Add a third question: how quickly does each contract turn into cash the company can retain? This is a way to compare the quality of growth across chips, cloud services, software and robotics without pretending their accounting is identical.

THREE IDEAS TO RESEARCH THIS WEEKEND

Research priorities, not trade recommendations. Each needs its own valuation work.

1 Oracle and the cost of financing growth

Why now: its latest results make customer funding central to the thesis. The case improves if cash arrives ahead of equipment spending and service delivery generates enough margin to fund renewal. The strongest objection is that attractive near-term funding can accompany expensive long-term obligations.

What would change my view: two successive quarters showing a smaller funding gap on comparable definitions, with no offsetting deterioration in payment terms or delivery obligations. Tripwire: advances weaken while committed construction and hardware payments continue. Start with the cash-flow statement and contract notes, not the backlog multiple.

2 TSMC and the return on manufacturing capacity

Why now: the August sales release gives another current checkpoint. The case is that competing AI architectures still require advanced manufacturing. The strongest objection is that a successful buildout may demand so much capital that revenue growth overstates the improvement in shareholder returns.

What would change my view: sustained sales strength alongside resilient gross margin and disciplined capital intensity in the next quarterly report. Tripwire: manufacturing investment rises while utilisation or margin deteriorates. Value the business using cash generation after investment, and explicitly allow for Taiwan exposure.

3 Adobe and the price of a slower growth path

Why now: a profitable incumbent offers a different question from a young AI application. Does its current valuation leave room for a slower, cash-generative business? At $248.83, the midpoint of Adobe’s FY2026 adjusted EPS guide ($24.475) implies about 10.2 times earnings; the GAAP midpoint ($18.145) implies about 13.7 times. Those are current-year price-to-earnings ratios, not a fair-value estimate.

Adobe · Q3 FY2026 results and financial statements · 10 Sep | MarketWatch · Adobe closing price · 10 Sep

The strongest objection is structural erosion of paid demand. What would change my view: conversion and retention improving while cash generation holds up. Tripwire: AI engagement rises but paid growth and cash per share deteriorate. Reconcile adjusted earnings to GAAP, and check how much buyback spending merely offsets employee equity awards.

WHAT COULD GO WRONG WITH OUR ANGLE

We could become too cautious during a profitable investment cycle. Customers paying early may demonstrate scarcity and bargaining power. High capital spending can precede years of attractive returns. Free products can become paying users. The correction is to demand evidence of those outcomes, rather than treating financing, investment or freemium distribution as defects by themselves.

AI INVESTMENT FRAMEWORK

Business conviction is separate from valuation. Risk labels concern the theme, not a forecast of share-price volatility. This is a research map, not a model portfolio or allocation instruction.

CHANGES THIS WEEK

Oracle receives closer scrutiny because funding terms can materially alter its investment case. TSMC becomes an explicit manufacturing watch. Adobe joins the valuation research list. The software signal is selective, and robotics gets a stronger commercial signal with no automatic increase in conviction about any listed stock.

Portfolio discipline: owning a chip designer, its foundry and its cloud customer can concentrate exposure to the same AI spending cycle. TSM appears in infrastructure and global analysis but is one economic exposure. Before adding a name, map the end customers your existing holdings depend on.

WHAT WE ARE WATCHING

BLS · September release calendar | Federal Reserve · FOMC calendar | Micron · 30 September earnings announcement

YOUR MOVE

Pick one holding. Write a one-page answer to these questions before you change its size:

• What evidence shows customers are paying for something economically useful?

• How much investment is needed before those customer payments arrive?

• Which payments are advances against work still owed?

• What would make the business worth less even if reported revenue keeps growing?

The companion deep dive, Customer Prepayments and the AI Cash Test, turns this into a reusable review. It includes a funding comparison, a worked timing example and a checklist for reading a cloud company’s next filing.

My preference this week is to improve the underwriting before expanding the watchlist. Oracle deserves a closer look. Adobe deserves a valuation check. Neither earns a buy decision from one headline.

Know someone building an AI portfolio? Forward this issue and ask which company they think converts growth into cash most convincingly.

Stay curious, and stay qualified.

P.S. Last week we mapped who bears the default. This week we ask what the customer has already paid for, and what still has to be delivered.

QUICK GLOSSARY

Customer prepayment Cash received before the related goods or services are delivered. Contract terms determine the remaining obligation.

RPO Remaining performance obligations: contracted revenue not yet recognised. It is not a cash balance.

Free cash flow Usually operating cash flow less capital expenditure. Always inspect the issuer’s definition and adjustments.

Revenue run rate An annualised extrapolation from a recent pace of revenue. It is not necessarily annual recurring revenue.

Capital intensity How much investment a business needs to support its sales and growth.

SOURCE AND METHOD NOTE

Links beside each story identify the evidence used. Oracle’s new-quarter figures rely on dated financial reporting from Reuters, the Financial Times and the WSJ; its detailed prepayment terms are not established here. Company announcements are attributed, not independently audited by iPrompt. Calculations and investment judgements are our own. Coverage ends before the scheduled Friday CPI release; no Friday market result is included.

Disclaimer: For information and education only, not financial advice. iPrompt Signals is not a registered investment adviser. Conduct your own research and consult a qualified financial professional before making investment decisions.

iPrompt Signals · Issue 28 · 11 September 2026 ·