Chips

Google sells TPUs into Oracle's cloud for the first time as Oracle's $90 billion chip plan still hedges Nvidia

Google's TPUs enter another company's data center for the first time, beside Cerebras and SambaNova racks and an expanded Nvidia commitment, all announced in one morning. Oracle now carries the incumbent and its challengers on a $90 billion-plus annual spending plan.

Vincent JiangVincent Jiang · 3 min read
Share
Cerebras chief executive Andrew Feldman speaking on a Collision conference stage
1 / 7Slide 1 of 7
Cerebras chief executive Andrew Feldman. His company's shares rose about 9.5 percent on Tuesday after Oracle agreed to carry its inference hardware, though the deal books no revenue until calendar 2027.

At Oracle's+4.44% — Oracle, up 4.44 percent today AI World conference in Las Vegas on October 27, one cloud signed four silicon suppliers in a single morning 1. Google's Tensor Processing Units, the custom AI chips built for Google's+2.10% — Google, up 2.10 percent today own cloud, will run inside Oracle Cloud Infrastructure from calendar 2027, the first time they have ever been sold into another company's data center 12.

Cerebras, whose flagship processor is a single chip the size of a whole silicon wafer rather than one cut from it, follows in the same year 3. SambaNova's$11B — SambaNova, private, latest valuation $11B SN50 racks and a deepened Nvidia+0.84% — Nvidia, up 0.84 percent today commitment complete the set 45.

A Google TPU v4 printed circuit board carrying four liquid-cooled TPU packages
A TPU v4 board from Google's own system. The chips have run only inside Google's data centers; Oracle will host them from calendar 2027, the first outside sale. · Norman P. Jouppi et al. via arXiv and Wikimedia Commons

Oracle pays for both sides of the fight

The money behind the morning is Oracle's own. The company maintained a capital-spending target of about $90 billion to $95 billion for the fiscal year ending May 2027, and said on its earnings call last month that customers increasingly foot the bill for the chips 6.

Oracle's quarterly capital spending has overtaken its revenue

  • Capital spending
  • Revenue
$0B$10B$20B$30BQ4 '24Q1 '25Q2 '25Q3 '25Q4 '25Q1 '26Q2 '26Q3 '26capex passed revenue; FCF rannegative $5.4B
Data
Capital spendingRevenue
Q4 '24$3.97B$14.06B
Q1 '25$5.86B$14.13B
Q2 '25$9.08B$15.9B
Q3 '25$8.5B$14.93B
Q4 '25$12.03B$16.06B
Q1 '26$18.64B$17.19B
Q2 '26$16.49B$19.18B
Q3 '26$28.5B$19.35B
Capital expenditure and total revenue by calendar quarter, in US$ billions, from Oracle's SEC filings. Q3 '26 is the quarter ended August 31, 2026, the quarter the story cites: $28.5 billion of capex, free cash flow of negative $5.4 billion.8

Nvidia, the supplier being hedged, booked $96.2 billion of revenue in the quarter ended July 26, 2026 7 and has said roughly half its revenue comes from hyperscalers such as Oracle 1. Oracle's free cash flow ran at negative $5.4 billion last quarter, on $28.5 billion of capital spending 8.

Nvidia books about $96 billion a quarter, the machine Oracle is hedging

$20B$40B$60B$80B$100BQ3 '24Q4 '24Q1 '25Q2 '25Q3 '25Q4 '25Q1 '26Q2 '26$96.22B
Data
Quarterly revenue
Q3 '24$35.08B
Q4 '24$39.33B
Q1 '25$44.06B
Q2 '25$46.74B
Q3 '25$57.01B
Q4 '25$68.13B
Q1 '26$81.62B
Q2 '26$96.22B
Reported quarterly revenue in US$ billions, from Nvidia's SEC filings. Q2 '26 is the quarter ended July 26, 2026.7

The hedge renews the dependency

The same morning deepened what it hedged. Nvidia and Oracle will expand Vera Rubin, Nvidia's newest generation of AI servers, to multi-gigawatt scale 5, and Nvidia chief executive Jensen Huang has said each gigawatt can mean up to $40 billion of revenue for his company 6.

OpenAI$500B — OpenAI, private, latest valuation $500B has trained and served GPT-6 Astra on an OCI Supercluster of more than 100,000 Blackwell GPUs in Abilene, Texas 5. One carve-out: Oracle's Nvidia systems will skip some Nvidia networking chips in favor of Oracle's in-house Acceleron 6. Oracle is financing Nvidia's challengers out of the same capital-spending plan that buys Nvidia.

The market paid on Tuesday for 2027 revenue

Cerebras+10.22% — Cerebras, up 10.22 percent today traded up about 9.5 percent and Oracle about 4 percent 1. The Cerebras deal discloses no volumes and books no revenue until calendar 2027 39, and analyst consensus compiled by Bloomberg has the company under $1 billion of sales this year, tripling to almost $3 billion next 9. Oracle, still down about 31 percent this year, grew cloud-infrastructure revenue 121 percent last quarter, with remaining performance obligations at $664 billion 10.

"That's a massive expansion of our customer base of tier-one customers," Cerebras chief executive Andrew Feldman told Reuters, which notes that Nvidia dominates training while inference, the work of answering live requests, is the more open market, where Cerebras and Intel+3.45% — Intel, up 3.45 percent today-backed SambaNova compete 6.

The questions the morning left open

For Nvidia holders, two questions no announcement answered: whether the Acceleron carve-out spreads, and whether Google sells TPUs into any other cloud. Oracle, which plans to hold an investor day this week 6, did not say how it will finance the chip purchases 6. The signal to watch: whether OCI names paying customers for the TPU capacity before the 2027 ramp 10.

Get Chips by email, free

How AI demand is reshaping chipmakers, memory and the fabs behind them. In our daily email, every weekday.

Loading NVDA

More about NVIDIA

NVDA · Fiscal Q2 2027
Fiscal Q2 2027 · quarter to 26 Jul 2026 · changes vs a year earlier

Revenue rose 106% to $96.2B, 92.5% of it from Data Center, “driven by the ramp of our Blackwell Ultra infrastructure”. Gains on equity stakes of $7.8B lifted net income to $59.7B.

Revenue$96.2B▲ 106%
Gross margin75.0%▲ 2.6 pts
Operating margin66.2%▲ 5.4 pts
Net income$59.7B▲ 126%
Hyperscale$48.7B▲ 102%AI clouds & enterprise$40.3B▲ 138%Data Center$89.0B▲ 117%Edge Computing$7.2B▲ 27%Revenue$96.2B▲ 106%Gross profit$72.1B75.0% marginCost of revenue$24.1BOther income$7.8Bequity-stake gainsOperating income$63.7B66.2% marginOperating expenses$8.4BNet income$59.7B▲ 126%Tax$11.8B16.5% rateR&D$7.1BSG&A$1.4B
Show as a table
LineValue
Revenue$96.2B
Gross margin75.0%
Operating margin66.2%
Net income$59.7B
Hyperscale$48.7B
AI clouds & enterprise$40.3B
Data Center$89.0B
Edge Computing$7.2B
Gross profit$72.1B
Cost of revenue$24.1B
Other income$7.8B
Operating income$63.7B
Operating expenses$8.4B
Tax$11.8B
R&D$7.1B
SG&A$1.4B

Reader comments

0 comments

    Chips

    SnapMagic turns text prompts into full circuit schematics; rival Flux has raised $37 million for the same job

    SnapMagic, the Y Combinator-backed company whose parts library sits inside distributor catalogs and PCB tools, is quietly generating whole circuit schematics from text, with no press release and a $37 million-funded rival already in the market.

    Chips

    Qualcomm cuts jobs worldwide, launches Anduril laptop as handset chips fall 20% before its November 4 report

    A layoff memo with no headcount and an American-built laptop landed eight days apart. On November 4, Qualcomm has to show the pivot outgrowing the phone decline.

    Chips

    Chip design software grew 12.2% in Q2. Silvaco, whose CEO read the numbers, guides down 8.5%

    The ESD Alliance's Q2 2026 report shows design-software revenue up 12.2%, with its spokesman, Silvaco CEO Wally Rhines, hailing the growth. His own company's quarter is modeled down 8.5%, and the Nov 4 print decides whether the AI story converts.

    Chips

    Advantest reports October 28 with a 51.7% margin and Bernstein's ¥900 billion operating-profit bar

    Advantest reports October 28 at 3:30pm Japan time, the first hard evidence on AI chip demand since the Nikkei's two-week rout. Bernstein's ¥900 billion full-year operating-profit bar, and a 51.7% margin business underneath, make tomorrow's guidance the AI trade's clearest test.

    Chips

    Intel CEO Lip-Bu Tan discussed collaboration with AMD; outside customers are 5.1% of Intel's foundry revenue

    Lip-Bu Tan put a possible AMD collaboration on the record on October 24, but Intel's own 10-Q shows outside customers are 5.1% of foundry revenue and most of the growth is accounting, while TSMC's A14 just restored cache scaling that Intel cannot yet match.

    Chips

    Monolithic Power is 32% below its high before October 29 earnings, with a fraud suit and Vicor's import bid

    The AI power-chip maker reports October 29 with a securities-fraud suit a federal judge has twice refused to dismiss and an ITC probe seeking to bar its AI-server power modules from US import. Consensus still expects its data-center segment to more than double.

    Get the newsletter

    The day's AI and market news, in your inbox every weekday.