Microsoft is reassessing its earlier strategy of making large bets on external AI compute suppliers. After tightening its review of partnerships with AI "neocloud" service providers, Microsoft has reportedly been phasing out its arrangement to rent AI cloud compute from Oracle.

The news, which surfaced on Friday, November 6, US Eastern Time, jolted market expectations for AI data center leasing demand. During Friday's US trading session, Oracle shares dived quickly at midday, while neocloud concept stocks that had surged in the morning gave back most of their gains. Among them, CoreWeave, Inc. (ASX: CRWV) briefly plunged before Oracle rebounded and ended higher. Neocloud service providers also all closed up.

On Friday, November 6, US Eastern Time, Bloomberg reported that Microsoft had begun shutting down some computing resources it previously rented from Oracle in recent months and decided not to renew the relevant contracts. People familiar with the matter said that over time, the compute Microsoft rented became less and less useful and no longer met Microsoft's technical specifications, because the servers Oracle provided used an older generation of Nvidia chips, reducing their value to Microsoft.

Just the day before, a Thursday report showed that after committing more than $80 billion to rent compute from other cloud companies, Microsoft was dissatisfied with the high costs and delivery delays of neocloud suppliers and regretted its earlier strategy. It was strengthening scrutiny of some data center projects, exiting projects it deemed riskier, and internally discussing whether to pause signing new partnership agreements. Two consecutive days of reports highlighted that Microsoft is re-examining its external compute procurement strategy.

Oracle shares plunge, company stresses AI facility demand remains strong

After Friday's report emerged, Oracle (NYSE: ORCL) shares, which had risen as much as 3.8% early in the session, quickly reversed course, erased their gains, turned negative, and hit a session low of $109.21, down nearly 5.2% intraday. They then fluctuated back into positive territory and closed up about 0.8%.

Other AI compute leasing suppliers also saw notable pullbacks from their gains. CoreWeave, Inc. (NYSE: CRWV), which had risen 9.3% in early US trading, briefly turned lower, falling more than 1% at one point, before quickly recovering but never gaining more than 2%. IREN Ltd (NASDAQ: IREN), which had surged about 24.6% early on, narrowed its gain to about 12% at one point. NEBIUS (NASDAQ: NBIS), which had risen 13.7% early in the session, saw its gain narrow to less than 1.9%.

By the close, CoreWeave rose more than 6%, while IREN and Nebius gained about 7.6% and 20%, respectively.

Regarding the news that Microsoft will not renew its lease, Oracle emphasized that demand for its AI infrastructure remains strong, including demand for older-generation GPUs. An Oracle spokesperson said the company still sees robust market demand for this type of compute. Oracle disclosed in September that it was renewing GPU rental contracts purchased four years ago at a premium of 20% above the original contract price.

Microsoft declined to comment on the matter.

Delivery delays and high costs prompt Microsoft to re-examine neocloud partnerships

Microsoft's shift in attitude toward external compute suppliers did not begin with this adjustment to the Oracle leasing agreement.

On Thursday, Bloomberg reported that Microsoft was stepping up oversight of some data center projects involving AI neocloud service providers such as Nebius, IREN, and Nscale. Because these companies face issues such as project construction delays, regulatory approval obstacles, and high costs, some Microsoft management has grown doubtful about previous large-scale partnership bets.

People familiar with the matter said Microsoft is exiting projects it considers riskier and internally discussing whether to continue signing new neocloud partnership agreements. Microsoft is also increasingly inclined to send employees to partner data center projects to inspect construction progress and ensure projects meet technical specifications and regulatory requirements.

Among them, an IREN data center project in Texas encountered fiber cabling problems, had a previous generator fire incident, and disclosed hardware delivery delays. Microsoft and IREN recently renegotiated the project timeline, and completion of the relevant facilities may be delayed by weeks or even months.

Microsoft Chief Financial Officer Amy Hood has also repeatedly mentioned on earnings calls the impact of third-party compute delivery delays on the company's capital expenditure plans. Microsoft is placing increasing importance on whether external suppliers can deliver compute on schedule and whether projects can meet the company's actual needs.

Nscale is another typical case of Microsoft scaling back cooperation.

Bloomberg reported on Thursday that Microsoft decided not to exercise an additional expansion option in Nscale's data center project in Texas. The two sides had previously reached a compute leasing arrangement of about 240 megawatts, and Microsoft originally had an option to add another 700 megawatts of compute, but because the project faced obstacles in energy access approval, Microsoft ultimately decided not to proceed with the expansion.

Earlier, Microsoft had already exited Nscale's flagship data center project in Loughton, Essex, in the UK. That project originally involved a compute leasing arrangement of about $3 billion, but due to delays in power access, the relevant power supply was expected to become available only in the 2030s.

Losing new cooperation opportunities with Microsoft one after another puts pressure on Nscale's plan to seek a $35 billion IPO valuation. However, Microsoft's decision not to exercise the Texas expansion option does not affect the original contract of about 240 megawatts between the two sides, nor does it affect Nscale's previously disclosed order backlog of about $103 billion.

From scrambling for compute to scrutinizing orders, the logic of AI infrastructure investment is changing

Over the past two years, large technology companies such as Microsoft, Google, and Amazon raced to build AI data centers. But because the pace of building their own infrastructure could not fully keep up with demand, emerging data center operators such as Nebius, IREN, CoreWeave, and Nscale won a large number of long-term compute leasing orders and became important beneficiaries of AI infrastructure expansion.

To meet the compute needs of its Azure cloud services and AI business, Microsoft had previously committed more than $80 billion in compute leasing spending to external cloud computing companies including CoreWeave, Nscale, and Nebius.

But as its own data center construction advanced and problems such as construction delays, equipment delivery issues, and financing costs at external suppliers gradually emerged, Microsoft began to reassess the economics and execution risks of these partnerships.

It is worth noting that Microsoft's adjustment of its external compute procurement strategy does not mean AI compute demand has peaked. Bloomberg previously pointed out that global AI computing capacity is still in short supply, and other large customers such as Anthropic and Google are still accelerating cooperation commitments with emerging data center operators.

For investors, the truly noteworthy change is that large technology companies' procurement standards for AI compute are shifting from simply pursuing scale to placing more emphasis on compute generation, delivery progress, cost efficiency, and project execution capability. For neocloud service providers that rely on orders from a small number of large technology companies to support huge capital expenditures, order size is no longer the sole basis for valuation. Whether contracts can be fulfilled on schedule and whether investment can be converted into stable cash flow will become key to how the market reassesses their business models.