Microsoft shares closed Friday, July 10, at €339.75, up 1.09%, after OpenAI named GPT-5.6 the preferred model for Microsoft 365 Copilot, a reassurance rally that collided with €196.6 billion in future AI-data-center lease commitments and fresh questions about when the spending will pay.
The immediate story is that Microsoft and OpenAI are not breaking up. The more consequential story is that Microsoft can no longer afford to build its AI business as though OpenAI will always provide the best model, the lowest operating cost, and the most strategically convenient answer at the same time.
Microsoft is therefore attempting something more sophisticated than choosing between a partner and a competitor: it is buying OpenAI’s frontier intelligence, building its own MAI alternatives, and turning Copilot into a routing layer capable of deciding which supplier should handle each task. The partnership is deepening at the product level even as dependency is being engineered out of the platform.
Microsoft’s Friday recovery followed a turbulent session in which the shares fell as low as €334.95 before closing at €339.75. The 1.09% daily gain offered some relief, but it did little to repair a year in which the stock has declined 17.01%.
OpenAI’s GPT-5.6 announcement supplied the market with a reassuring counter-narrative. After Bloomberg reported that Microsoft had begun replacing some OpenAI and Anthropic technology with its own models in applications including Excel and Outlook, OpenAI said GPT-5.6 would become the preferred model for Microsoft 365 Copilot.
Microsoft Copilot chief Nitin Agrawal endorsed the arrangement, saying customers could expect “significantly better results.” The message was clearly intended to make one point: Microsoft’s internal model program has not displaced its access to OpenAI’s newest technology.
That distinction matters because the market has repeatedly treated every sign of Microsoft model diversification as evidence that the OpenAI relationship is deteriorating. In reality, the evidence points toward a relationship becoming less exclusive, more transactional, and arguably more durable because neither company must now pretend that one supplier can satisfy every requirement.
OpenAI remains capable of supplying frontier models that Microsoft can rapidly distribute through Word, Excel, PowerPoint, and its collaborative services. Microsoft, meanwhile, owns the applications, customer relationships, identity systems, compliance controls, cloud infrastructure, and billing machinery surrounding those models.
The GPT-5.6 rollout on July 9 therefore did not restore the old partnership. It demonstrated what the newer one looks like: OpenAI competes for workloads inside Microsoft’s products while Microsoft retains the power to route other workloads elsewhere.
The distinction between “preferred” and “exclusive” is the hinge on which the entire strategy turns.
OpenAI wants Microsoft’s distribution and infrastructure, but it does not want its commercial future reduced to being the intelligence supplier buried inside another company’s software. The two businesses remain strategically entangled while steadily creating room to operate independently.
That tension intensified in 2026. In April, the companies revised their partnership, ending Microsoft’s exclusive right to sell OpenAI models while preserving continuing economic and technology ties. OpenAI said revenue-share payments to Microsoft would continue through 2030, subject to a cap, even as the companies moved toward a less exclusive commercial structure.
Bloomberg subsequently reported that Microsoft had spent more than $100 billion across its OpenAI investments, infrastructure construction, and model-hosting obligations. That figure illustrates why Microsoft cannot treat model procurement as a routine software-licensing decision: OpenAI is embedded in the economics of Azure, Copilot, and Microsoft’s wider infrastructure program.
Yet the same scale makes diversification unavoidable. Microsoft cannot credibly promise enterprises stable pricing, predictable latency, regional capacity, and continuous model improvements if every important Copilot workload ultimately depends on one external laboratory.
The strategic options now divide into two complementary tracks:
Microsoft’s seven-model MAI family, introduced at Build 2026, spans reasoning, coding, image, voice, and transcription workloads. The company emphasized efficiency and lower token costs alongside capability, making clear that the project is as much an economic program as a research program.
Its first reasoning model, MAI-Thinking-1, was described as a mid-sized model with 35 billion active parameters and a 256K context window. Microsoft said the model was trained from scratch without distillation, using clean, commercially licensed enterprise-grade data.
Those claims position MAI-Thinking-1 less as an attempt to win every public benchmark than as infrastructure Microsoft can safely place inside commercial products. For Microsoft, a model that is somewhat less capable but materially cheaper, faster, easier to govern, and sufficiently accurate may be more valuable across millions of ordinary prompts than an expensive frontier model used indiscriminately.
This is not unique to Microsoft. The economics of enterprise AI increasingly favor portfolios of models, with premium systems reserved for tasks that justify their cost and smaller models assigned to summarization, extraction, classification, formatting, and routine document work.
What is unique is Microsoft’s ability to implement that portfolio across software already used by hundreds of millions of workers.
It must answer millions of requests inside applications where customers already expect rapid responses and where the cost of each inference can accumulate faster than subscription revenue. A premium model may produce an excellent answer, but sending every prompt to it would be like assigning a senior engineer to every help-desk ticket.
Bloomberg reported that Microsoft has started using its own models for a portion of the work performed in products such as Excel and Outlook. TechCrunch described the shift as part of a wider AI cost-control trend, while Microsoft’s Build announcements stressed first-party models offering enterprise capability at lower operating cost.
The significant word is portion. Microsoft does not need MAI to replace GPT-5.6 across all of Copilot to alter the economics of the service.
It only needs to divert enough routine traffic away from expensive third-party models. A simple spreadsheet-formatting request, email summary, document classification, or meeting follow-up may not require the same reasoning capacity as a complex financial model or a multi-document legal analysis.
That creates a model-routing problem. Copilot must determine the nature of a request, evaluate the available models, consider latency and cost, apply organizational policy, and then send the task to an appropriate system without making the experience feel fragmented.
If Microsoft gets that layer right, the identity of the underlying model becomes less important to most users. People will purchase Copilot because it works inside their documents, messages, meetings, permissions, and business processes—not because every response carries the label of a particular AI laboratory.
If Microsoft gets it wrong, users may see inconsistent quality across applications or even across similar prompts. A model selected to reduce costs could misunderstand a complex spreadsheet, produce weaker prose, or respond differently from the system employees were trained to use.
That is why GPT-5.6 becoming the “preferred” model should not be read as confirmation that all Microsoft 365 Copilot requests will be processed by GPT-5.6. Preferred status establishes the flagship experience; it does not necessarily disclose how Microsoft will allocate every workload behind the interface.
For Microsoft, intelligent routing is not merely an engineering optimization. It may determine whether Copilot can become a software-margin business rather than a heavily subsidized showcase for data-center spending.
According to Bloomberg data cited in the source reporting, Microsoft added €41 billion in the latest measurement period. Across the cloud industry, cumulative commitments have climbed beyond €850 billion.
These are future obligations rather than a single immediate cash payment. Even so, they represent real commitments to facilities, power, networking, cooling, and computing capacity whose economics depend on future demand developing broadly as expected.
Microsoft argues that this infrastructure is necessary to support future cash-flow growth. That case is plausible: Azure demand remains strong, model usage is compute-intensive, and shortages of suitable capacity can prevent the company from serving customers even when those customers are ready to spend.
But there is a structural difference between being demand-constrained and being profit-constrained. Microsoft may fill its data centers while still discovering that the price of providing AI services is too high relative to the revenue those services generate.
The industry’s challenge is that AI capacity must often be arranged before demand becomes fully visible. Data centers take time to secure, equip, connect, and energize; waiting until every workload is contracted would mean arriving years too late.
That forces companies to make irreversible decisions under uncertain conditions. If demand exceeds forecasts, Microsoft owns valuable capacity in a scarce market. If demand disappoints, it holds expensive commitments to infrastructure whose most profitable use cases have not yet materialized.
The scale also changes the meaning of Microsoft’s in-house models. MAI is not simply a hedge against disagreement with OpenAI. It is an attempt to extract more useful work from every unit of infrastructure Microsoft has already committed to obtain.
A cheaper model can increase gross margin, but it can also increase effective capacity. If Microsoft can process a task using fewer computational resources, the same data-center footprint can serve more customers without a proportional increase in spending.
That is the deeper connection between the Bloomberg report and the GPT-5.6 announcement. Microsoft is not choosing between high-end external models and efficient internal ones; it needs both because the infrastructure bill has become too large for a one-model strategy.
The decisive AI contest may be fought not over which company has the smartest model, but over which company can deliver sufficient intelligence at a sustainable cost.
Azure and other cloud-services revenue grew 40%, while the Intelligent Cloud segment expanded 30% to $34.7 billion. Those figures confirm that customers continue moving workloads to Microsoft’s cloud and that the company is converting at least part of the AI cycle into measurable revenue.
The unresolved question is whether growth is accelerating enough to justify the magnitude and timing of the infrastructure commitments. A company can produce excellent revenue growth while still earning an inadequate return on the incremental capital required to generate it.
Microsoft’s roughly €2.5 trillion market capitalization makes that distinction especially important. Investors are not valuing the company merely as a mature software vendor with a strong cloud division; they are pricing expectations that AI will expand revenue, defend Microsoft 365, deepen Azure usage, and create new categories of paid work.
Those expectations raise the burden of proof. Azure growing 40% is powerful evidence of demand, but investors also need to know how much capital was required to support that growth and whether cloud margins can remain resilient as AI workloads consume more expensive infrastructure.
Some analysts reportedly expect the most tangible returns from the current buildout to emerge only from 2028 onward. If that assessment is correct, Microsoft faces an awkward interval in which commitments continue rising faster than the market’s ability to verify their ultimate profitability.
That interval explains the cooling enthusiasm. The market has moved beyond asking whether generative AI will be important and has begun asking who will retain the economic value after model providers, chipmakers, data-center operators, utilities, and cloud platforms all take their share.
Satya Nadella continues to describe generative AI as a fundamental computing shift. The argument is consistent with Microsoft’s history: the company has repeatedly benefited from owning the platform through which a major transition reaches business customers.
Yet AI differs from previous software transitions because its variable operating costs can remain substantial after the product has been built. Microsoft could distribute Office to an additional customer at minimal incremental cost; supplying that customer with continuous model inference is a different economic proposition.
The gross-margin advantage of software does not disappear, but it becomes dependent on architecture. Model selection, caching, infrastructure utilization, custom silicon, scheduling, and prompt efficiency all influence whether an AI subscription behaves financially like software or like a metered utility.
That is why the MAI program belongs in the same discussion as Azure growth and lease obligations. It is one of the mechanisms Microsoft can use to make revenue growth economically meaningful rather than merely impressive.
The combination is telling. A lowered target with a positive rating does not amount to a rejection of Microsoft’s strategy; it reflects diminished confidence in the speed and valuation multiple of the payoff.
Investors should also avoid comparing Bonner’s dollar-denominated target directly with the euro-denominated market figures without accounting for the different currencies and trading venues. The important signal is the direction of the revision, not a simplistic subtraction between the two numbers.
Technically, Microsoft remains under pressure. Friday’s €339.75 close left the shares 2.4% below their 50-day moving average of €348.21 and 10% below their 200-day moving average of €378.99.
The stock has recovered 10.6% from the 52-week low of €307.10 reached in late June. It nevertheless remains 28.9% below the October 2025 high of €478.10.
An RSI reading of 50.6 is effectively neutral. The stock is neither signaling obvious technical exhaustion nor producing the momentum that would settle the debate in favor of a durable recovery.
Those indicators describe sentiment rather than business value, but they show that the market is waiting for a more decisive catalyst. GPT-5.6 may reduce anxiety about the OpenAI relationship, yet it cannot answer whether Microsoft’s AI infrastructure will generate an adequate return.
That leaves the July 29 fiscal fourth-quarter earnings report as the next major test. Investors will be looking beyond the headline revenue figure to Azure growth, cloud capacity, capital requirements, margins, management guidance, and any evidence that Copilot is becoming a material contributor rather than an expensive feature bundled into Microsoft’s wider ecosystem.
April 29, 2026 — Microsoft reports fiscal third-quarter revenue of $82.9 billion, up 18%, with Azure and other cloud-services revenue growing 40%.
June 2, 2026 — Microsoft uses Build 2026 to introduce its expanded family of seven in-house MAI models across reasoning, coding, image, voice, and transcription workloads.
Late June 2026 — Microsoft shares reach a 52-week low of €307.10 as concern grows over AI spending and the time required to earn a return.
July 7, 2026 — Bloomberg reports that Microsoft is replacing some OpenAI and Anthropic model usage with its own technology in applications including Excel and Outlook.
July 9, 2026 — OpenAI rolls out GPT-5.6 and identifies it as the preferred model for Microsoft 365 Copilot.
July 10, 2026 — Microsoft closes at €339.75, Argus cuts its price target from $620 to $510 while retaining a Buy rating, and the market continues weighing growth against infrastructure commitments.
July 29, 2026 — Microsoft is scheduled to report fiscal fourth-quarter results after the closing bell.
That flexibility can benefit customers. Microsoft can introduce stronger models without forcing every organization to rebuild integrations, and it can use lower-cost systems for routine tasks while reserving more expensive reasoning for genuinely difficult requests.
It can also complicate testing. If a finance team validates a spreadsheet workflow against one model, a later routing change could alter how Copilot interprets instructions, handles formulas, summarizes ranges, or explains its result.
The same concern applies to Word and PowerPoint. A new model may produce better documents overall while changing tone, structure, citation behavior, or the way it follows established prompt templates.
IT teams should therefore treat Copilot as a continuously serviced platform, not as conventional software whose behavior remains fixed until the next installed version. Model changes require regression testing even when Microsoft presents them as transparent upgrades.
Administrators also need to distinguish between model identity and data governance. The appearance of a new model does not, by itself, prove that an organization’s contractual protections, processing locations, retention behavior, or compliance controls have changed—but those assumptions should be verified rather than inferred.
Microsoft’s multi-model direction makes vendor review more important because the underlying supply chain can expand. OpenAI, Microsoft’s own MAI systems, and other model providers may appear in different Copilot experiences, depending on product design and organizational configuration.
The practical objective is not to block every model change. It is to know which business processes are sensitive enough that a change in model behavior must be tested before broad deployment.
The first test is Azure’s trajectory. Microsoft previously guided toward fiscal fourth-quarter Azure growth of 39% to 40% in constant currency, making any material deviation important to the market’s assessment of AI demand.
The second test is margin. If revenue grows rapidly while the expense of providing AI capacity rises faster, investors will remain skeptical about the quality of that growth.
The third is capacity utilization. Microsoft has repeatedly framed infrastructure investment as a response to constrained demand, but the market will want signs that new capacity is coming online in the right regions and being consumed by paying customers.
The fourth is Copilot monetization. Microsoft does not need to disclose every inference cost or model-routing decision, but it does need to show that its AI products are strengthening subscriptions, increasing usage, attracting premium workloads, or expanding customer commitments.
Finally, management must explain the €196.6 billion-equivalent lease burden in a way that connects specific obligations to a credible revenue opportunity. Broad declarations about the arrival of a new computing era are no longer enough on their own.
The stock’s weakness suggests that investors have already discounted part of the slower-payback risk. That does not automatically make Microsoft a buy, just as the enormous lease figure does not automatically make it a sell.
A bearish interpretation sees a company committing unprecedented sums to infrastructure before the unit economics of AI software are proven. A bullish interpretation sees one of the few companies with the applications, cloud, capital, distribution, and engineering capacity required to turn that infrastructure into a global platform.
Both readings can be supported by the current evidence. The July 29 report matters because it may indicate which side is gaining strength.
The immediate story is that Microsoft and OpenAI are not breaking up. The more consequential story is that Microsoft can no longer afford to build its AI business as though OpenAI will always provide the best model, the lowest operating cost, and the most strategically convenient answer at the same time.
Microsoft is therefore attempting something more sophisticated than choosing between a partner and a competitor: it is buying OpenAI’s frontier intelligence, building its own MAI alternatives, and turning Copilot into a routing layer capable of deciding which supplier should handle each task. The partnership is deepening at the product level even as dependency is being engineered out of the platform.
The Relief Rally Hid a More Expensive Truth
Microsoft’s Friday recovery followed a turbulent session in which the shares fell as low as €334.95 before closing at €339.75. The 1.09% daily gain offered some relief, but it did little to repair a year in which the stock has declined 17.01%.OpenAI’s GPT-5.6 announcement supplied the market with a reassuring counter-narrative. After Bloomberg reported that Microsoft had begun replacing some OpenAI and Anthropic technology with its own models in applications including Excel and Outlook, OpenAI said GPT-5.6 would become the preferred model for Microsoft 365 Copilot.
Microsoft Copilot chief Nitin Agrawal endorsed the arrangement, saying customers could expect “significantly better results.” The message was clearly intended to make one point: Microsoft’s internal model program has not displaced its access to OpenAI’s newest technology.
That distinction matters because the market has repeatedly treated every sign of Microsoft model diversification as evidence that the OpenAI relationship is deteriorating. In reality, the evidence points toward a relationship becoming less exclusive, more transactional, and arguably more durable because neither company must now pretend that one supplier can satisfy every requirement.
OpenAI remains capable of supplying frontier models that Microsoft can rapidly distribute through Word, Excel, PowerPoint, and its collaborative services. Microsoft, meanwhile, owns the applications, customer relationships, identity systems, compliance controls, cloud infrastructure, and billing machinery surrounding those models.
The GPT-5.6 rollout on July 9 therefore did not restore the old partnership. It demonstrated what the newer one looks like: OpenAI competes for workloads inside Microsoft’s products while Microsoft retains the power to route other workloads elsewhere.
The distinction between “preferred” and “exclusive” is the hinge on which the entire strategy turns.
Microsoft and OpenAI Are Becoming Customers, Suppliers, and Competitors
The Microsoft-OpenAI relationship has always contained conflicting incentives. Microsoft wants access to the most capable models available, but it also wants to control the cost, reliability, product roadmap, and negotiating leverage behind Copilot.OpenAI wants Microsoft’s distribution and infrastructure, but it does not want its commercial future reduced to being the intelligence supplier buried inside another company’s software. The two businesses remain strategically entangled while steadily creating room to operate independently.
That tension intensified in 2026. In April, the companies revised their partnership, ending Microsoft’s exclusive right to sell OpenAI models while preserving continuing economic and technology ties. OpenAI said revenue-share payments to Microsoft would continue through 2030, subject to a cap, even as the companies moved toward a less exclusive commercial structure.
Bloomberg subsequently reported that Microsoft had spent more than $100 billion across its OpenAI investments, infrastructure construction, and model-hosting obligations. That figure illustrates why Microsoft cannot treat model procurement as a routine software-licensing decision: OpenAI is embedded in the economics of Azure, Copilot, and Microsoft’s wider infrastructure program.
Yet the same scale makes diversification unavoidable. Microsoft cannot credibly promise enterprises stable pricing, predictable latency, regional capacity, and continuous model improvements if every important Copilot workload ultimately depends on one external laboratory.
The strategic options now divide into two complementary tracks:
| Dimension | OpenAI model track | Microsoft MAI track | Strategic consequence |
|---|---|---|---|
| Primary role | Frontier intelligence for demanding Copilot workloads | First-party models optimized for Microsoft products and infrastructure | Microsoft can select models by task rather than brand |
| Current signal | GPT-5.6 becomes the preferred Microsoft 365 Copilot model | Seven in-house MAI models unveiled at Build 2026 | Partnership continues while internal substitution expands |
| Main advantage | Rapid access to advanced general-purpose capabilities | Greater control over cost, deployment, tuning, and roadmap | Lower dependence on any single provider |
| Main risk | Supplier cost and strategic reliance | Performance may not match frontier models in every workload | Model routing becomes essential |
| Customer effect | Potentially stronger results on complex work | Potentially faster or cheaper responses for routine work | Copilot behavior may vary by application and task |
Its first reasoning model, MAI-Thinking-1, was described as a mid-sized model with 35 billion active parameters and a 256K context window. Microsoft said the model was trained from scratch without distillation, using clean, commercially licensed enterprise-grade data.
Those claims position MAI-Thinking-1 less as an attempt to win every public benchmark than as infrastructure Microsoft can safely place inside commercial products. For Microsoft, a model that is somewhat less capable but materially cheaper, faster, easier to govern, and sufficiently accurate may be more valuable across millions of ordinary prompts than an expensive frontier model used indiscriminately.
This is not unique to Microsoft. The economics of enterprise AI increasingly favor portfolios of models, with premium systems reserved for tasks that justify their cost and smaller models assigned to summarization, extraction, classification, formatting, and routine document work.
What is unique is Microsoft’s ability to implement that portfolio across software already used by hundreds of millions of workers.
Model Routing Turns Copilot Into a Margin Engine
The public conversation around AI models remains fixated on benchmark leadership: which system writes the best code, solves the hardest mathematics, or performs most impressively in a staged demonstration. Microsoft’s commercial problem is more prosaic.It must answer millions of requests inside applications where customers already expect rapid responses and where the cost of each inference can accumulate faster than subscription revenue. A premium model may produce an excellent answer, but sending every prompt to it would be like assigning a senior engineer to every help-desk ticket.
Bloomberg reported that Microsoft has started using its own models for a portion of the work performed in products such as Excel and Outlook. TechCrunch described the shift as part of a wider AI cost-control trend, while Microsoft’s Build announcements stressed first-party models offering enterprise capability at lower operating cost.
The significant word is portion. Microsoft does not need MAI to replace GPT-5.6 across all of Copilot to alter the economics of the service.
It only needs to divert enough routine traffic away from expensive third-party models. A simple spreadsheet-formatting request, email summary, document classification, or meeting follow-up may not require the same reasoning capacity as a complex financial model or a multi-document legal analysis.
That creates a model-routing problem. Copilot must determine the nature of a request, evaluate the available models, consider latency and cost, apply organizational policy, and then send the task to an appropriate system without making the experience feel fragmented.
If Microsoft gets that layer right, the identity of the underlying model becomes less important to most users. People will purchase Copilot because it works inside their documents, messages, meetings, permissions, and business processes—not because every response carries the label of a particular AI laboratory.
If Microsoft gets it wrong, users may see inconsistent quality across applications or even across similar prompts. A model selected to reduce costs could misunderstand a complex spreadsheet, produce weaker prose, or respond differently from the system employees were trained to use.
That is why GPT-5.6 becoming the “preferred” model should not be read as confirmation that all Microsoft 365 Copilot requests will be processed by GPT-5.6. Preferred status establishes the flagship experience; it does not necessarily disclose how Microsoft will allocate every workload behind the interface.
For Microsoft, intelligent routing is not merely an engineering optimization. It may determine whether Copilot can become a software-margin business rather than a heavily subsidized showcase for data-center spending.
The Lease Bill Moves AI Risk Off the Demo Stage
The largest number in Microsoft’s AI story is no longer a model parameter count or benchmark score. It is the company’s €196.6 billion in future lease commitments, primarily associated with the infrastructure needed to supply cloud and AI capacity.According to Bloomberg data cited in the source reporting, Microsoft added €41 billion in the latest measurement period. Across the cloud industry, cumulative commitments have climbed beyond €850 billion.
These are future obligations rather than a single immediate cash payment. Even so, they represent real commitments to facilities, power, networking, cooling, and computing capacity whose economics depend on future demand developing broadly as expected.
Microsoft argues that this infrastructure is necessary to support future cash-flow growth. That case is plausible: Azure demand remains strong, model usage is compute-intensive, and shortages of suitable capacity can prevent the company from serving customers even when those customers are ready to spend.
But there is a structural difference between being demand-constrained and being profit-constrained. Microsoft may fill its data centers while still discovering that the price of providing AI services is too high relative to the revenue those services generate.
The industry’s challenge is that AI capacity must often be arranged before demand becomes fully visible. Data centers take time to secure, equip, connect, and energize; waiting until every workload is contracted would mean arriving years too late.
That forces companies to make irreversible decisions under uncertain conditions. If demand exceeds forecasts, Microsoft owns valuable capacity in a scarce market. If demand disappoints, it holds expensive commitments to infrastructure whose most profitable use cases have not yet materialized.
The scale also changes the meaning of Microsoft’s in-house models. MAI is not simply a hedge against disagreement with OpenAI. It is an attempt to extract more useful work from every unit of infrastructure Microsoft has already committed to obtain.
A cheaper model can increase gross margin, but it can also increase effective capacity. If Microsoft can process a task using fewer computational resources, the same data-center footprint can serve more customers without a proportional increase in spending.
That is the deeper connection between the Bloomberg report and the GPT-5.6 announcement. Microsoft is not choosing between high-end external models and efficient internal ones; it needs both because the infrastructure bill has become too large for a one-model strategy.
The decisive AI contest may be fought not over which company has the smartest model, but over which company can deliver sufficient intelligence at a sustainable cost.
Strong Cloud Growth Has Not Settled the Payback Debate
Microsoft’s operating results do not resemble those of a company whose AI strategy has failed. In the fiscal third quarter of 2026, revenue reached $82.9 billion, an increase of 18%.Azure and other cloud-services revenue grew 40%, while the Intelligent Cloud segment expanded 30% to $34.7 billion. Those figures confirm that customers continue moving workloads to Microsoft’s cloud and that the company is converting at least part of the AI cycle into measurable revenue.
The unresolved question is whether growth is accelerating enough to justify the magnitude and timing of the infrastructure commitments. A company can produce excellent revenue growth while still earning an inadequate return on the incremental capital required to generate it.
Microsoft’s roughly €2.5 trillion market capitalization makes that distinction especially important. Investors are not valuing the company merely as a mature software vendor with a strong cloud division; they are pricing expectations that AI will expand revenue, defend Microsoft 365, deepen Azure usage, and create new categories of paid work.
Those expectations raise the burden of proof. Azure growing 40% is powerful evidence of demand, but investors also need to know how much capital was required to support that growth and whether cloud margins can remain resilient as AI workloads consume more expensive infrastructure.
Some analysts reportedly expect the most tangible returns from the current buildout to emerge only from 2028 onward. If that assessment is correct, Microsoft faces an awkward interval in which commitments continue rising faster than the market’s ability to verify their ultimate profitability.
That interval explains the cooling enthusiasm. The market has moved beyond asking whether generative AI will be important and has begun asking who will retain the economic value after model providers, chipmakers, data-center operators, utilities, and cloud platforms all take their share.
Satya Nadella continues to describe generative AI as a fundamental computing shift. The argument is consistent with Microsoft’s history: the company has repeatedly benefited from owning the platform through which a major transition reaches business customers.
Yet AI differs from previous software transitions because its variable operating costs can remain substantial after the product has been built. Microsoft could distribute Office to an additional customer at minimal incremental cost; supplying that customer with continuous model inference is a different economic proposition.
The gross-margin advantage of software does not disappear, but it becomes dependent on architecture. Model selection, caching, infrastructure utilization, custom silicon, scheduling, and prompt efficiency all influence whether an AI subscription behaves financially like software or like a metered utility.
That is why the MAI program belongs in the same discussion as Azure growth and lease obligations. It is one of the mechanisms Microsoft can use to make revenue growth economically meaningful rather than merely impressive.
The Stock Is Waiting for Proof, Not Another Promise
Argus Research analyst Joseph Bonner reduced his Microsoft price target from $620 to $510 on July 10 while retaining a Buy rating. Bonner praised Microsoft’s revenue and margin momentum but cited a market reassessment of how quickly AI investments are likely to produce returns.The combination is telling. A lowered target with a positive rating does not amount to a rejection of Microsoft’s strategy; it reflects diminished confidence in the speed and valuation multiple of the payoff.
Investors should also avoid comparing Bonner’s dollar-denominated target directly with the euro-denominated market figures without accounting for the different currencies and trading venues. The important signal is the direction of the revision, not a simplistic subtraction between the two numbers.
Technically, Microsoft remains under pressure. Friday’s €339.75 close left the shares 2.4% below their 50-day moving average of €348.21 and 10% below their 200-day moving average of €378.99.
The stock has recovered 10.6% from the 52-week low of €307.10 reached in late June. It nevertheless remains 28.9% below the October 2025 high of €478.10.
An RSI reading of 50.6 is effectively neutral. The stock is neither signaling obvious technical exhaustion nor producing the momentum that would settle the debate in favor of a durable recovery.
Those indicators describe sentiment rather than business value, but they show that the market is waiting for a more decisive catalyst. GPT-5.6 may reduce anxiety about the OpenAI relationship, yet it cannot answer whether Microsoft’s AI infrastructure will generate an adequate return.
That leaves the July 29 fiscal fourth-quarter earnings report as the next major test. Investors will be looking beyond the headline revenue figure to Azure growth, cloud capacity, capital requirements, margins, management guidance, and any evidence that Copilot is becoming a material contributor rather than an expensive feature bundled into Microsoft’s wider ecosystem.
Timeline
April 27, 2026 — Microsoft and OpenAI announce a revised phase of their partnership, reducing exclusivity while preserving continuing economic and technology ties.April 29, 2026 — Microsoft reports fiscal third-quarter revenue of $82.9 billion, up 18%, with Azure and other cloud-services revenue growing 40%.
June 2, 2026 — Microsoft uses Build 2026 to introduce its expanded family of seven in-house MAI models across reasoning, coding, image, voice, and transcription workloads.
Late June 2026 — Microsoft shares reach a 52-week low of €307.10 as concern grows over AI spending and the time required to earn a return.
July 7, 2026 — Bloomberg reports that Microsoft is replacing some OpenAI and Anthropic model usage with its own technology in applications including Excel and Outlook.
July 9, 2026 — OpenAI rolls out GPT-5.6 and identifies it as the preferred model for Microsoft 365 Copilot.
July 10, 2026 — Microsoft closes at €339.75, Argus cuts its price target from $620 to $510 while retaining a Buy rating, and the market continues weighing growth against infrastructure commitments.
July 29, 2026 — Microsoft is scheduled to report fiscal fourth-quarter results after the closing bell.
For IT, “Preferred” Does Not Mean “Only”
For enterprise administrators, the model competition introduces a governance issue that is easy to miss when announcements focus on improved output. Organizations are no longer deploying a single, static Copilot model; they are adopting a service whose underlying intelligence may change by workload, feature, region, availability, or Microsoft’s internal routing decisions.That flexibility can benefit customers. Microsoft can introduce stronger models without forcing every organization to rebuild integrations, and it can use lower-cost systems for routine tasks while reserving more expensive reasoning for genuinely difficult requests.
It can also complicate testing. If a finance team validates a spreadsheet workflow against one model, a later routing change could alter how Copilot interprets instructions, handles formulas, summarizes ranges, or explains its result.
The same concern applies to Word and PowerPoint. A new model may produce better documents overall while changing tone, structure, citation behavior, or the way it follows established prompt templates.
IT teams should therefore treat Copilot as a continuously serviced platform, not as conventional software whose behavior remains fixed until the next installed version. Model changes require regression testing even when Microsoft presents them as transparent upgrades.
Administrators also need to distinguish between model identity and data governance. The appearance of a new model does not, by itself, prove that an organization’s contractual protections, processing locations, retention behavior, or compliance controls have changed—but those assumptions should be verified rather than inferred.
Microsoft’s multi-model direction makes vendor review more important because the underlying supply chain can expand. OpenAI, Microsoft’s own MAI systems, and other model providers may appear in different Copilot experiences, depending on product design and organizational configuration.
The practical objective is not to block every model change. It is to know which business processes are sensitive enough that a change in model behavior must be tested before broad deployment.
Action checklist for admins
- Review Microsoft 365 service communications and roadmap notices for the GPT-5.6 rollout and any tenant-specific availability details.
- Identify critical Word, Excel, PowerPoint, Outlook, and collaborative workflows that depend on repeatable Copilot output.
- Re-run established prompt and workflow tests when a new model becomes available or preferred.
- Confirm which model providers and subprocessors apply to each enabled Copilot experience.
- Verify that organizational data-handling, retention, regional-processing, and compliance requirements remain satisfied.
- Document unexpected output changes and compare them across applications before expanding deployment.
- Brief help-desk and training teams so users understand that Copilot behavior may evolve without a traditional desktop-software upgrade.
July 29 Must Connect Infrastructure to Revenue
Microsoft’s coming earnings report will not settle the entire AI investment debate, but it can narrow it. The company needs to demonstrate that the demand represented by Azure’s prior 40% growth is durable and that capacity expansion is producing revenue rather than merely future availability.The first test is Azure’s trajectory. Microsoft previously guided toward fiscal fourth-quarter Azure growth of 39% to 40% in constant currency, making any material deviation important to the market’s assessment of AI demand.
The second test is margin. If revenue grows rapidly while the expense of providing AI capacity rises faster, investors will remain skeptical about the quality of that growth.
The third is capacity utilization. Microsoft has repeatedly framed infrastructure investment as a response to constrained demand, but the market will want signs that new capacity is coming online in the right regions and being consumed by paying customers.
The fourth is Copilot monetization. Microsoft does not need to disclose every inference cost or model-routing decision, but it does need to show that its AI products are strengthening subscriptions, increasing usage, attracting premium workloads, or expanding customer commitments.
Finally, management must explain the €196.6 billion-equivalent lease burden in a way that connects specific obligations to a credible revenue opportunity. Broad declarations about the arrival of a new computing era are no longer enough on their own.
The stock’s weakness suggests that investors have already discounted part of the slower-payback risk. That does not automatically make Microsoft a buy, just as the enormous lease figure does not automatically make it a sell.
A bearish interpretation sees a company committing unprecedented sums to infrastructure before the unit economics of AI software are proven. A bullish interpretation sees one of the few companies with the applications, cloud, capital, distribution, and engineering capacity required to turn that infrastructure into a global platform.
Both readings can be supported by the current evidence. The July 29 report matters because it may indicate which side is gaining strength.
The Numbers That Now Define Microsoft’s AI Bet
The short-term debate is about GPT-5.6, MAI routing, and Microsoft’s next earnings report. The longer-term verdict will rest on whether the company can combine frontier capability with first-party efficiency without degrading the products on which enterprises depend.- Microsoft closed July 10 at €339.75, up 1.09% for the session but down 17.01% year to date.
- GPT-5.6 becoming Microsoft 365 Copilot’s preferred model confirms that OpenAI remains strategically important.
- Seven in-house MAI models show that Microsoft is simultaneously building an alternative supply chain.
- Future lease obligations have reached €196.6 billion, including an additional €41 billion in the latest measurement period.
- Fiscal third-quarter revenue rose 18% to $82.9 billion, while Azure and other cloud-services revenue grew 40%.
- The July 29 earnings report must show whether cloud growth and Copilot demand can justify the infrastructure burden.
References
- Primary source: AD HOC NEWS
Published: 2026-07-10T16:24:08.869509
Microsoft's AI Paradox: Soaring Leasing Costs and a Deepening OpenAI Dance
Microsoft shares recover after OpenAI confirms GPT-5.6 for Copilot, but $196B AI lease obligations and analyst caution cap gains. Stock down 17% YTD.www.ad-hoc-news.de - Official source: blogs.microsoft.com
Microsoft Build 2026: Be yourself at work - The Official Microsoft Blog
Platforms shift when developers build. We explore, choose tools, dream, create. This platform shift comes with more information than ever, ready at your fingertips. This shift, it’s about building fast AND THEN: it’s about building, operating, optimizing and observing. Securing your...blogs.microsoft.com - Official source: techcommunity.microsoft.com
Loading…
techcommunity.microsoft.com - Official source: devblogs.microsoft.com
What's new in Microsoft Foundry | Build Edition | Microsoft Foundry Blog
Microsoft Build 2026 brings a major set of Microsoft Foundry updates for developers building agents: hosted runtimes, Toolboxes, memory, Voice Live, Foundry IQ, new models, managed compute, and trust, evaluation, and observability tools.devblogs.microsoft.com - Official source: news.microsoft.com
Microsoft Build 2026
news.microsoft.com
- Official source: build.microsoft.com
Your home for Microsoft Build
Go deep on real code and real systems with the teams building and scaling AI at Microsoft Build, June 2–3, 2026, in San Francisco and online.build.microsoft.com
- Official source: azure.microsoft.com
3 things leaders need to know from Microsoft Build 2026 | Microsoft Azure Blog
At Microsoft Build 2026, AI moved from experimentation to execution—learn how organizations can scale AI systems, connect data and workflows, and drive measurable business outcomes.azure.microsoft.com - Official source: microsoft.com
MSR at BUILD 2026 - Microsoft Research
Microsoft Research is at BUILD 2026 this week, giving developers a hands-on look at some of the many AI-based models and tools they can use to accelerate innovation, enhance their capabilities, and quickly transform ideas into prototypes.www.microsoft.com - Official source: marketingassets.microsoft.com
- Official source: download.microsoft.com
- Official source: cdn-dynmedia-1.microsoft.com
- Related coverage: windowscentral.com
Microsoft drops $9.7B on AI cloud power to keep up with demand | Windows Central
The $9.7B deal highlights how AI demand is outpacing compute supply — and why Microsoft keeps signing billion‑dollar contracts.www.windowscentral.com - Related coverage: axios.com
OpenAI releases GPT-5.6 and ChatGPT Work tool
The leading players in AI are all speeding forward at breakneck speed.www.axios.com
- Related coverage: bloomberg.com
- Related coverage: news.bloomberglaw.com
Microsoft Replaces OpenAI, Anthropic With Own AI in Some Apps
Microsoft Corp., looking to reduce AI costs, is starting to replace OpenAI and Anthropic with its own models in software products like Excel and Outlook.news.bloomberglaw.com
- Related coverage: techcrunch.com
Microsoft joins AI cost-cutting trend by relying more on its own models | TechCrunch
Microsoft is the latest Silicon Valley giant to cut back on its AI spending.techcrunch.com - Related coverage: bloomberght.com
Microsoft ve OpenAI ortaklık yapısını revize etti | Son dakika ekonomi haberleri
Microsoft ile OpenAI arasındaki anlaşma güncellenirken, OpenAI'ın ürünlerini farklı bulut sağlayıcıları üzerinden sunmasının önü açıldı ve gelir paylaşımına üst sınır getirildi.
www.bloomberght.com
- Official source: openai.com
The next phase of the Microsoft OpenAI partnership | OpenAI
OpenAI and Microsoft announce an amended agreement that simplifies the partnership, adds long-term clarity, and supports continued AI innovation at scale.openai.com - Related coverage: news.bloombergtax.com
Microsoft Makes Big AI Inroads in China by Selling OpenAI Models
Microsoft Corp. has built a big business selling AI models to Chinese companies despite the growing rivalry between the US and China over artificial intelligence.news.bloombergtax.com
- Related coverage: techradar.com
'We don’t believe this kind of government access process should become the long-term default': OpenAI unveils big GPT-5.6 upgrades for ChatGPT, but you can't use them yet | TechRadar
Only select partners for nowwww.techradar.com - Official source: help.openai.com
GPT-5.6 in ChatGPT | OpenAI Help Center
Learn how GPT-5.6 Sol works in ChatGPT, which options are available by plan, and how usage limits and availability work.
help.openai.com
- Official source: learn.microsoft.com
Microsoft 365 Roadmap | Microsoft 365
The Microsoft 365 Roadmap lists updates that are currently planned for applicable subscribers. Check here for more information on the status of new features and updates.learn.microsoft.com
- Official source: developers.openai.com
Page not found | OpenAI API
developers.openai.com
- Related coverage: wienerborse.at
