Oracle stock is being asked to absorb two facts that do not answer the same question. The Financial Times reported on October 1, 2026, that Tencent has taken a five-year lease worth about $7 billion for access to roughly 100,000 advanced AI chips in Oracle data centers in Southeast Asia. Separately, director Stephen H. Rusckowski bought 25,000 shares in the open market on September 29 at a weighted average of $139.352. One number is a multi-year compute contract that Oracle has not identified in its latest quarterly report. The other is a small, coded purchase on a Form 4. The comparison that matters is whether $7 billion is large next to the backlog and the spending Oracle has already disclosed.
Key facts
- The reported Oracle Tencent deal is a five-year lease of about $7 billion, with roughly 30 percent paid upfront, for about 100,000 advanced AI chips in Southeast Asia. The Financial Times is the publication that reported it. Oracle does not mention Tencent in its September 10 earnings release, its September 10 call, or its Form 10-Q for the quarter ended August 31, 2026.
- Those accounts do not name Nvidia or any other chip vendor. “Advanced AI chips” is as specific as the reporting gets.
- Remaining performance obligations were $664 billion on August 31, 2026. About $7 billion is 1.05 percent of that balance. Cash capital expenditure in the quarter was $28.5 billion. On the September 10 call, Chief Financial Officer Hilary Maxson said the company still anticipated $90 billion to $95 billion of capital expenditure for the full year.
- Rusckowski’s Form 4 is transaction code P, an open-market purchase, not a grant. The October 1 close was $138.07, about 0.9 percent below his price. The form does not mention Tencent.
What was reported, and who reported it
The FT story, by Zijing Wu in Hong Kong, is headlined as Tencent leasing 100,000 chips from Oracle. FinanceFeeds could not read the piece behind the FT paywall on October 2. The terms below are the ones Investing.com, carried on Yahoo Finance, attributed to the FT, which it said had cited two people with knowledge of the matter. Reuters ran a headline on the same report. Its page did not open for this desk the morning of October 2.
As that account has it, Tencent agreed to a five-year lease across multiple Oracle data centers in Southeast Asia. The lease is described as Tencent’s largest overseas lease, worth $7 billion, with about 30 percent paid upfront. Thirty percent of $7 billion is about $2.1 billion. The chips are described as advanced processors that were not available for purchase inside China. The same account says the commitment weighed on Tencent’s free cash flow in its second-quarter results. The chip count is in this article because the FT, Reuters, and Investing.com each attached a name to “about 100,000.” It is not in this article because Oracle published a bill of materials.
None of that text names Nvidia, AMD, or an Oracle part. A separate paragraph in the secondary coverage says U.S. rules have limited direct purchases of advanced Nvidia chips by Chinese firms while still allowing some overseas leases. That is backdrop, not a specification. FinanceFeeds has covered Nvidia GPU rentals separately. It is not evidence of the vendor in this contract. An even split of $7 billion would be $1.4 billion a year, or about $350 million a quarter, but the 30 percent upfront means the cash is front-loaded. Oracle’s 10-Q says cloud revenue is generally recognized over the term, or as a usage service is consumed. The newspaper figure is a contract value, not a quarterly revenue print.
The silence is the industry response
Oracle’s September 10 earnings release, the call that afternoon, and the Form 10-Q filed September 11 do not contain the word Tencent. Tencent has not, in any statement this desk could tie to the October 1 report, confirmed the lease, the dollar figure, the term, the sites, or the chip count. A Yahoo Finance rewrite of the wire said Reuters had not verified the FT account and that Oracle and Tencent had not answered comment requests. Because the Reuters page itself did not load here, that non-response sits at one remove. It still matches the documents that did load: unnamed sources in a newspaper story, and no company exhibit.
Until one of the companies identifies the contract, $7 billion is a reported lease, not a named line in the backlog. It can be real and already inside the August 31 balance. The relayed FT account ties the cash to Tencent’s second-quarter free cash flow, the quarter ended June 30, so it was not signed on October 1. Oracle’s fiscal first quarter ran from June 1 to August 31. A June signing could sit there. An April or May signing would sit in the quarter ended May 31. Oracle has not said which.
Set $7 billion next to the $664 billion backlog
Oracle’s Form 10-Q states that remaining performance obligations were $664 billion as of August 31, 2026, and $455 billion a year earlier. The filing says the increase came primarily from significant cloud contracts signed in the period. Oracle expects to recognize about 13 percent of the August 31 balance over the next twelve months, 37 percent in months 13 through 36, 34 percent in months 37 through 60, and the rest later. Thirteen plus 37 is half inside three years: about $86 billion over the coming year, and about $332 billion over 36 months.
The September 10 earnings release calls that $664 billion an increase of $209 billion from a year earlier, and says Oracle booked more than $30 billion of additional AI cloud contracts in the quarter. Those are different statistics. The $209 billion is the year-over-year change in the stock of obligations. The “more than $30 billion” is new AI contracts in one quarter. On the call, Maxson said the backlog rose $26 billion from the prior quarter, a smaller net change because older obligations were recognized as revenue while new ones were added. Adding $7 billion on top of $664 billion assumes the lease was signed after August 31 and is still unbooked. The second-quarter cash comment points the other way. Inside the balance, $7 billion is 1.05 percent. Outside it, the balance would rise by about 1 percent. Neither version changes the shape of the backlog.
FinanceFeeds went through the 36-month conversion after the quarter. The schedule to keep is the one in the 10-Q. A five-year lease, recognized evenly, would drop a thin slice into each bucket. It would not be the contract that moves the 13 percent.
On revenue, the release reported $19,345 million in the quarter, up 30 percent, cloud revenue of $11,607 million, infrastructure revenue of $7.4 billion, up 121 percent, and applications revenue of $4.2 billion, up 10 percent. An even $350 million a quarter from the reported lease would be about 4.7 percent of one quarter of infrastructure revenue and about 3 percent of cloud revenue. Large for an ordinary order. Small next to a cloud line that just grew triple digits. Oracle has not published this customer’s schedule, and it has not said the customer is new.
The same dollars next to the buildout
The 10-Q cash flow statement shows capital expenditure of $28,499 million for the three months ended August 31, 2026, against $8,502 million a year earlier. Operating cash flow was $23,103 million, against $8,140 million. The release describes free cash flow as negative $5 billion, because Oracle was spending to support cloud infrastructure growth.
On the September 10 call, Maxson said: “Our CapEx for the quarter was $28 billion, leading to negative free cash flow of $5 billion. Our net cash CapEx, so net of prepayments, was $18 billion for the quarter.” Spending, she said, will not be linear. “We continue to anticipate $90 billion-$95 billion in CapEx for the full year, with not more than $70 billion in net cash CapEx.” That range is her forward-looking comment, not a 10-Q line, and it is the yardstick she offered for judging a $7 billion contract.
On the filing figure, the reported lease is about 25 percent of one quarter’s cash capital expenditure. It is about 7 to 8 percent of the full-year range Maxson said the company still anticipated. The upfront 30 percent, about $2.1 billion, sits next to $11.4 billion of customer prepayments in the quarter that included a significant financing component. The cash flow statement records an $11,363 million increase in deferred revenue from those prepayments. A $2.1 billion check would be visible inside $11.4 billion. It would not be most of it. Oracle has not said any Tencent cash is in the line.
The release also says Oracle sold $20 billion of common stock, before commissions, through an at-the-market program in the quarter. Cash proceeds net of issuance costs were $19,909 million. FinanceFeeds has already reported the debt side of the same buildout. Oracle is financing data centers by the tens of billions a quarter. One reported customer of $7 billion over five years does not fund that program.
The release said the quarter’s new AI contracts were structured so there would be “no incremental impact” on plans to raise capital. Maxson said: “The vast majority of those new contracts were via prepay or bring your own hardware or similar mechanic, so will not require incremental capital from Oracle. Also, that new RPO will not impact our CapEx or revenues until fiscal 2028 or beyond.” She also expects “around half of our RPO to convert into sales over the next 36 months,” the 10-Q’s 13 percent plus 37 percent. A prepaid lease can fit that pattern. The FT account sounds like capacity with cash up front, not a customer bringing its own racks. Fitting the pattern is not Oracle naming the customer, and her timing comment argues against counting this headline as fiscal 2027 revenue.
What Clay Magouyrk said, and what he did not
Ken Bond, head of investor relations, introduced co-chief executives Mike Sicilia and Clay Magouyrk, and Maxson. Nobody on the call said Tencent. The call was September 10, three weeks before the FT story. Magouyrk’s sentence on bookings was: “We closed more than $30 billion of additional AI contracts in Q1 without requiring additional capital from Oracle.”
In the same stretch he said Oracle had delivered “850 megawatts of AI capacity containing more than 300,000 GPUs” since the end of the prior quarter, almost three times that quarter’s delivery and 73 percent of last fiscal year’s delivered capacity. The earnings release says 850 megawatts of additional data-center capacity and more than 300,000 GPUs. Neither names the vendor. He also said utilization was 97.9 percent, and that GPUs up for renewal were renewed or resold at a 20 percent premium, most of them four years old or older.
That is the scale check on the chip count. More than 300,000 GPUs delivered in one quarter, against reported Tencent access of about 100,000. If both figures are right, the lease is a large customer and a fraction of one quarter’s deliveries. It is not the cluster that explains a $664 billion backlog. Full-year revenue guidance is at least $90 billion, and non-GAAP earnings per share are guided at $8.10. August-quarter GAAP diluted earnings were $1.56. The Tencent report revises none of those, because Oracle has not tied the report to guidance.
The Form 4 is a purchase, and a small one
Oracle filed Rusckowski’s Form 4 on October 1, the same calendar day as the FT story. The trade date on the form is September 29, which is before that headline. He is listed as a director, not an officer. The definitive proxy filed September 25, 2026, identifies him as an independent director and a member of the compensation committee, and as the former chief executive and president of Quest Diagnostics. The proxy says he was appointed in the board’s fiscal 2026 refreshment.
The form shows one non-derivative transaction: 25,000 shares of common stock, transaction code P, acquired, on September 29. Code P is an open-market purchase. It is not code A, a grant or award, and it is not an option exercise. The shares are held indirectly, “By The Stephen H. Rusckowski Living Trust.” After the trade the trust held 25,390 shares. A separate line shows 60 shares held directly. The form does not indicate a Rule 10b5-1 plan.
The price is $139.352. Footnote F1 says the order was executed in multiple trades from $138.96 to $139.74, and that the reported figure is the weighted average. The footnote uses the phrase “weighted average sales price.” The codes do not describe a sale. The acquired-or-disposed flag is A, for acquired, and the trust’s share count rose. Footnote boilerplate sometimes keeps the word “sales” on a buy. The codes govern. This was a purchase. At the weighted average, 25,000 shares cost $3,483,800.
The earnings release shows 3.000 billion diluted shares and 2.966 billion basic shares. Twenty-five thousand shares are about 0.0008 percent of the diluted count. At $138.07, that basic count implies an equity value near $410 billion, so $3.5 million is a personal ticket. September 29 volume was 45.7 million shares. The director’s lot was about 0.05 percent of the session. The day’s range was $132.53 to $143.67, and the $138.96 to $139.74 fills sat inside it, above the $137.79 close.
The form names no customer and gives no reason. A September 29 purchase cannot be a reaction to an October 1 headline, and the form does not say what Rusckowski knew. The fact on paper is an independent director adding 25,000 shares in the open market, through his living trust, at about $139.35.
The price path around that print
Nasdaq’s daily history puts the October 1, 2026, regular-session close at $138.07. That is $1.28 under the weighted average purchase price, or about 0.9 percent. “About $139” was a fair description of the buy. The stock has not run away from the fill. This article was finished before the U.S. cash open on October 2, so it uses the October 1 close rather than a premarket quote that would be stale by afternoon.
Stooq’s orcl.us history file returned “Access denied” that morning, so the chart uses Nasdaq closes. The marker is the Form 4 price, not the September 29 close of $137.79. A separate quote page showed the same $138.07 regular-session close.
Over those six months the highest close was $248.15 on June 1 and the lowest was $114.99 on July 24. October 1 is about 44 percent below the June close and about 20 percent above the July low, and about 5 percent under the April 1 close of $145.23. September 10 closed at $152.94, from $161.63 the day before. The headline arrived after the June spike was already gone.
Why the two records should not be averaged
The contract is still a newspaper account. The FT, Reuters, and Investing.com are real publications, and the terms they attribute to the FT are specific: about $7 billion, five years, Southeast Asia, about 30 percent upfront, about 100,000 chips. Oracle knows how to book contracts. It reported a $664 billion backlog and, on the call, a $26 billion sequential increase, and it has not named Tencent on either. Maxson’s comment that this kind of new backlog should not hit revenue or capital expenditure until fiscal 2028 or later was not about this customer. It is still the pattern the company has described.
The purchase is a narrow fact about one director, not a substitute for that disclosure. Code P, an indirect buy, a price inside the day’s range, under $3.5 million, filed within two business days. A grant is not a buy, and this was not a grant. A buy is not a press release. The reported lease is small next to August-quarter infrastructure revenue, August-quarter capital expenditure, and the backlog. The purchase is invisible next to the share count. Neither document shows that the other is true. A named confirmation, a named chip vendor, or a sentence on whether the lease is already inside the August 31 balance would change the read. None of those has been published.
What the filings answer
Did Oracle confirm the $7 billion Tencent lease?
No. Not in the September 10 release, the call, or the Form 10-Q. The figure is the Financial Times’, as carried by Investing.com and headlined by Reuters.
How large is $7 billion next to Oracle’s own numbers?
About 1.05 percent of the $664 billion backlog at August 31, 2026. About a quarter of that quarter’s $28.5 billion of capital expenditure. About 7 to 8 percent of the $90 billion to $95 billion full-year range the CFO still anticipated. Spread evenly, about $350 million a quarter, against $7.4 billion of infrastructure revenue. Those ratios use the headline value. They do not assume the contract is new.
Who bought the shares, and was it a grant?
Stephen H. Rusckowski, an independent director, bought 25,000 shares on September 29 at a $139.352 weighted average, through his living trust. Code P is an open-market purchase, not a grant. The October 1 close of $138.07 was about 0.9 percent below that price.
Does the reporting name Nvidia?
Not for these chips. The count is attributed to the Financial Times. The vendor is not, and Oracle’s own comment about more than 300,000 GPUs delivered in the quarter does not name one either.
This article is not financial advice and is not a recommendation to buy or sell Oracle stock.
