AI’s next bill: power, permissions and cash

Sunday brief · October 4

• Power has a price: Oracle’s Point Beach proposal puts the electricity bill alongside the chips. Expected customer savings still require regulatory approval. [1]
• Agents need more than intelligence: Microsoft’s business-data integration and Oracle’s controlled execution tools address the less glamorous work between a demonstration and dependable production. [3,4]
• Wheel desk: Microsoft is watch-only; Oracle and Intel are passes for now. Three visual studies expose the cash commitment and downside behind Friday’s premiums.

The lead · who pays for AI’s electricity?

Oracle announced on October 2 that it plans to subscribe to part of the existing Point Beach nuclear plant’s output. The company expects the arrangement to save Wisconsin utility customers about $300 million in fuel costs. The planned subscription remains subject to state Public Service Commission approval. This is a proposed allocation of existing power costs, not proof of newly commissioned generating capacity. [1]

Our interpretation: AI infrastructure has to earn its return after electricity, buildings, chips and financing. A data-center announcement can be strategically valuable while making the near-term cash bill larger. For shareholders, the question is whether contracted demand will pay for that bill on attractive terms.

The counterargument deserves weight: securing dependable power and addressing local costs may help a developer make progress where competitors face delays. A higher upfront commitment can be rational if it supports durable, profitable utilization. The October 2 announcement does not quantify that return.

The next useful evidence is the regulatory filing, its approved commercial terms and the timing of usable capacity. Keep those separate from promotional economic-impact estimates. For a wheel investor, willingness to own Oracle should survive a slower buildout and a funding surprise; a small option credit cannot carry the entire thesis.

Eddie’s watchlist · three checkpoints

• MSFT — trusted business context: Microsoft made Fabric IQ available in Copilot Chat and Cowork on September 28; IQ Sharing remains a preview. The investment question is whether better data grounding supports paid, repeatable use. Watch customer economics rather than treating product availability as incremental revenue. [4]
• ORCL — execution meets financing: Fusion Claw arrived September 29 with 25 new applications. Its controls may make complex workflows more usable, but the cash comparison below shows why software announcements cannot settle the infrastructure funding question. Watch deployment results and the next cash-flow report. [3]
• INTC — separate recovery from valuation: July’s Q2 report showed revenue up 25% and plans to increase equipment, clean-room and substrate investment. That historical improvement is relevant to AI supply; it does not establish a cheap entry after a strong share-price advance. Next earnings date remains unconfirmed here. [6]

AI buildout & revolution · production is the test

Governed action: Oracle describes Fusion Claw as combining AI reasoning with deterministic computation and explicit organizational permissions. This architecture aims to reduce expensive reasoning on routine execution. Our test is practical: completed work, exception rates and total cost, including human checking. The announcement establishes a product launch, not audited customer savings. [3]

Work between conversations: OpenAI’s September 29 announcement introduced dots, persistent agents with a cloud computer, with a gradual rollout to eligible users. This week’s development shifts the question from a useful answer to reliable ongoing work. Our investment lens: more activity can expand compute demand, but retries and supervision can consume the value. Usage alone is not profit. [21]

One chart · prepayments are not recurring earnings

Oracle’s $23.103 billion operating cash inflow included $11.363 billion of customer prepayments with a significant financing component: 49.2% of the total. Capital spending was $28.499 billion, leaving a $5.396 billion gap. Prepayments support construction but bring future service obligations; subtracting them mechanically does not produce a normalized cash-flow forecast. [2,12]

Macro lens: September payrolls rose 29,000 and unemployment was 4.2%, reported October 2. That leaves a tension for AI stocks: softer demand can pressure customer budgets even if markets hope for easier rates. The report does not identify AI as the cause. [7]

3 Wheel Trade Ideas

PRIOR-SESSION STUDIES · October 16 puts, 12 calendar days. Stock closes and last option trades: October 2. Exact bid/ask clock, timezone, delay, OI date and IV rank remain unconfirmed. Markets are closed Sunday; reprice at the October 5 open. No fill assumed. All amounts USD.

One standard 100-share put; verify its deliverable. Capital = strike × 100; reference credit = bid × 100; breakeven = strike − bid. Maximum loss assumes shares fall to zero. Delta is model sensitivity, not an assignment promise. Daily Wilder RSI14 and SMA20/50/200 use adjusted closes; relative volume compares the full session with the prior 20-session average. [14–20]

MSFT · Microsoft

Watch-only · quality needs a sensible entry · Prior-session study: the $495 put, expiring October 16, 2026. One contract means agreeing to buy 100 shares at that price if assigned. Prior-session reference; not a current offer.

• Why own the shares: Recurring software and cloud demand support an ownership case; tighter customer budgets remain a macro risk. FY26 GAAP EPS was $17.95, including investment effects. The $493 breakeven is 27.5× that annual figure, not a forward multiple or proof of cheapness. [5]
• Trend and levels: Close $517.53; RSI 62.7; SMA20/50/200 $501.67/$487.39/$431.46; relative volume 0.83×. Above all averages, not oversold. Strike lies between the 20- and 50-day references. [15]
• Contract quality: October 2 bid/ask $2.00/$2.39; spread $0.39; delta −0.1685; IV 24.68%; volume 313/OI 3,759. Size 35/71; last trade 15:57:41, provider clock. Wide spread warrants patience. [16]

ORCL · Oracle

Pass · attractive business, unsettled entry · Prior-session study: the $130 put, expiring October 16, 2026. One contract means agreeing to buy 100 shares at that price if assigned. Prior-session reference; not a current offer.

• Why own the shares: Cloud infrastructure grew strongly, but power and financing costs matter when rates stay high. Q1 GAAP EPS $1.56 implies 20.7× four times that quarter at breakeven; this is a quarterly proxy, not normalized or forward P/E. Negative free cash flow weakens the bargain case. [12]
• Trend and levels: Close $142.30; RSI 48.3; SMA20/50/200 $145.98/$143.59/$162.21; relative volume 0.99×. Below every average, not oversold. A lower strike alone does not establish support. [17]
• Contract quality: October 2 bid/ask $1.07/$1.14; spread $0.07; delta −0.1557; IV 46.98%; volume 1,997/OI 10,675. Size 60/65; last trade 15:59:38, provider clock. [18]

INTC · Intel

Pass · turnaround is not a valuation cushion · Prior-session study: the $108 put, expiring October 16, 2026. One contract means agreeing to buy 100 shares at that price if assigned. Prior-session reference; not a current offer.

• Why own the shares: CPU demand supports recovery, but capital intensity and cyclical spending remain risks. Q2 GAAP EPS was −$2.16 versus adjusted $0.42; escrowed-share valuation charges explain much of the gap. Neither a loss multiple nor annualizing one adjusted quarter establishes cheapness. [6]
• Trend and levels: Close $119.33; RSI 60.6; SMA20/50/200 $111.64/$100.87/$81.18; relative volume 0.85×. Above all averages and far from oversold; the $108 strike is below the 20-day but above the 50-day. [19]
• Contract quality: October 2 bid/ask $1.30/$1.36; spread $0.06; delta −0.1743; IV 58.20%; volume 408/OI 793. Size 18/60; last trade 15:49:10, provider clock. Modest OI needs rechecking. [20]

Next on the radar · Toronto time

• October 5, 09:30: next regular U.S. stock/options open. Reprice, check deliverables and refresh all earnings/dividend calendars before considering any study.
• October 9: Oracle dividend record date; confirm exchange ex-date and entitlement rules. This is not automatic dividend income from selling a put. [12]
• October 14, 08:30: September CPI, before these contracts expire. Watch the implications for financing costs and customer budgets. [8]

From YieldCove · start with ownership

In YieldCove, record the business case and cash required before the premium. Covered calls cap upside without removing share-price losses; three technology names still share macro risks.

Photo: Point Beach, September 8, 2023; archival context, not a new Oracle facility. Yodel2010, CC BY-SA 4.0; original unaltered. [9,10]

AI-assisted editorial by Codex. Educational information, not personalized investment advice. Scenarios exclude fees and taxes; early assignment is possible. No trades executed.