Microsoft Corp CEO Satya Nadella warned that enterprises using closed artificial intelligence models may effectively pay for intelligence twice. Social Capital founder Chamath Palihapitiya endorsed this view on October 4, 2026, stating that companies risk exposing valuable institutional know-how alongside monetary costs.

The reverse information paradox

Nadella argues that businesses incur a dual cost when relying on proprietary AI systems. The first expense is direct, through token or subscription fees. The second is indirect, arising from the proprietary knowledge required to make the intelligence useful.

"You essentially pay for intelligence twice, once with money, and again with something even more valuable: the proprietary knowledge you must reveal to make that intelligence useful," Nadella wrote in July. This knowledge manifests in prompts, agent tool calls, evaluations, and human corrections. Each correction contributes to what Social Capital describes as a learning loop, where repeated interactions become increasingly valuable to the model provider.

Open-weight models and data control

Palihapitiya has previously argued that cheaper open-weight models are rapidly closing the capability gap, raising questions about the premium paid to closed-model providers. Social Capital research from September indicated that leading open-weight models were roughly four months behind the best publicly evaluated closed frontier systems. However, these open alternatives offer businesses greater control over data, infrastructure, and customization.

Enterprise policies and economic reversals

The "pay twice" argument does not imply that every enterprise prompt automatically trains a provider’s foundation model. Microsoft states that prompts, responses, and organizational data in Microsoft 365 Copilot are not used to train foundation models. Similarly, OpenAI confirms that ChatGPT Business, Enterprise, and API data are excluded from training by default unless customers explicitly opt in. Reuters reported last week that OpenAI reiterated business data is not used for training without consent.

However, the economics can shift based on voluntary data sharing. OpenAI offers complimentary API tokens to customers who voluntarily share eligible inputs and outputs for model improvement. This incentive includes up to one million daily tokens on certain larger models, potentially reversing the traditional cost structure for participating enterprises.

What the numbers show

The divergence between capability lag and cost incentives highlights a strategic trade-off for enterprises. While open-weight models trail closed frontier systems by approximately four months according to September data, they offer significant advantages in data sovereignty. Conversely, closed models offer immediate access to top-tier performance but require careful navigation of data-sharing agreements to avoid unintended intellectual property leakage. The availability of complimentary tokens for data sharing in closed ecosystems suggests providers are actively monetizing enterprise data flows, validating the concern that knowledge itself is a currency in the AI economy.

Microsoft Corp. (NASDAQ: MSFT) is working with OpenAI to integrate the latter's new autonomous "Dots" agents into Microsoft's enterprise security and governance stack. This partnership allows companies to manage AI workers that operate across corporate systems without constant prompting, leveraging Microsoft's existing control plane tools.

The integration centers on "specialist Dots," which can be assigned unique identities, credentials, and access to internal systems. OpenAI stated it is collaborating with Microsoft to link these agents with Agent 365, a service described as a control plane for managing and securing AI agents. Agent 365 utilizes Entra for identity and access management, Defender for security, and Purview for data governance. The standalone Agent 365 service costs $15 per user per month with an annual commitment.

OpenAI's autonomous agent capabilities

OpenAI unveiled Dots at DevDay on Tuesday. Powered by GPT-6 Astra, these agents possess their own cloud computers and can connect to more than 4,000 apps. They are designed to maintain continuity across multiple projects without requiring user direction at every step. For instance, a developer Dot can monitor customer feedback, build and test fixes, and return completed pull requests for review. Communication channels include ChatGPT, Slack, and Microsoft Teams.

This development follows Microsoft's recent redesign of its own Copilot platform, which introduced Home, Code, and Autopilot capabilities. Autopilot is a persistent AI agent designed to handle recurring tasks within a company's Microsoft environment. The dual approach suggests Microsoft is positioning itself both as a builder of AI workers and as the infrastructure provider for managing agents created by others.

Market sentiment on AI agent leadership

Despite OpenAI's push into autonomous agents, prediction-market traders currently favor Anthropic in the race for the best AI agent. A Polymarket contract asking which company will have the best AI agent by the end of October shows Anthropic leading with 83% odds, compared to 11% for OpenAI. Meta Platforms Inc. sits at approximately 1%, reflecting market focus on agent capabilities rather than adoption rates.

Microsoft's strategy may rely less on having the single best underlying model and more on becoming the default management layer for whichever agents enterprises choose to deploy. By embedding governance into Agent 365, Microsoft secures a role in the workflow even when competitors build the core intelligence.

Copilot's new capabilities

The redesigned Copilot brings three distinct additions to the platform. Home combines Chat and Cowork while adding Word, Excel and PowerPoint directly to Copilot. Code lets users build apps, dashboards and automated workflows using natural-language prompts, running on technology that also powers GitHub Copilot. Autopilot is designed as a persistent AI agent that can handle recurring tasks, follow up on projects and continue working without constant prompts, operating within a company's Microsoft environment with defined permissions and governance.

Microsoft said Home and Code will begin rolling out through its Frontier program in the coming weeks. Autopilot will expand to private preview at the end of September. The company also introduced new FinOps tools aimed at helping businesses manage AI spending.

Deep discounts for large enterprises

The product launch follows a report by The Information that Microsoft plans to cut Copilot prices for large corporate customers. The base price for Copilot remains at $30 per user per month. Discounts could begin as soon as October and may be tied to large seat commitments and additional usage-based spending.

The proposed discount tiers are structured as follows:

This represents a significant shift from previous offers, where discounts were capped at approximately 10% for customers buying at least 2,000 seats. In July, Microsoft reported that Microsoft 365 Copilot surpassed 30 million paid seats, with net seat additions more than doubling quarter over quarter.

AI spending set to surge

Microsoft's monetization push comes as hyperscaler AI spending accelerates. Goldman Sachs strategists led by Ryan Hammond expect capital spending by Alphabet Inc., Amazon.com Inc., Microsoft, Meta Platforms Inc. and Oracle Corp. to rise more than 50% to $1.2 trillion in 2027, according to Bloomberg. The group is on track to spend about $800 billion in 2026. Goldman estimates hyperscalers need roughly $300 billion in annual AI revenue to break even on those investments.

Technical indicators

Microsoft remains in a long-term uptrend. The stock is trading 3.1% above its 20-day simple moving average of $500.01, 8.4% above its 50-day SMA of $475.80, and 19.3% above its 200-day SMA of $432.06. The 20-day SMA remains above the 50-day average. The 50-day SMA crossed above the 200-day SMA in August, forming a bullish golden cross. However, the MACD is below its signal line and the histogram remains negative, suggesting the broader trend is still positive but the pace of gains has slowed. Near-term resistance sits around $518, with support near $486.

Analyst outlook and ETF exposure

Microsoft carries a Buy consensus rating and an average price forecast of $567.04. Recent analyst actions include:

• Stifel: Upgraded to Buy on Sept. 23; raised price target to $575
• Oppenheimer: Maintained Outperform on Sept. 22; lifted target to $570
• Cantor Fitzgerald: Maintained Overweight on Sept. 21; raised target to $608

Microsoft is a major holding in several large growth and technology ETFs. It represents about 9.59% of the Vanguard Growth ETF, 9.83% of the Technology Select Sector SPDR Fund and 9.75% of the iShares Core S&P U.S. Growth ETF. Its large weighting means fund flows into or out of those ETFs can also influence trading in Microsoft shares.

What the numbers show

The combination of integrating OpenAI's Dots into Agent 365 and launching Copilot's own Autopilot highlights a bifurcated strategy: Microsoft is competing to build AI workers while simultaneously building the essential infrastructure to manage agents from any provider. The $15 monthly fee for Agent 365 creates a recurring revenue stream independent of which AI model dominates the capability race, evidenced by prediction markets favoring Anthropic (83%) over OpenAI (11%) for best agent performance. This positions Microsoft as a neutral governance layer, potentially insulating its enterprise software dominance from volatility in underlying model performance.