Oracle is raising $20 billion while Palantir reports a quarterly operating margin of 47%. Both are expanding in AI. Their numbers describe very different experiences of the same boom.
\n\n\n\nAlongside them, Claude developer Anthropic has disclosed enormous computing commitments as demand for its models grows. Read separately, these announcements look like financial news for investors. Read together, they raise a question for anyone using AI at work: can the value customers receive support the cost of supplying it?
\n\n\n\nThe answer will influence which providers survive, what businesses pay and how much control customers retain over increasingly essential tools.
\n\n\n\nAnthropic develops models. Oracle supplies cloud infrastructure alongside its established software businesses. Palantir connects data, software and AI to operational decisions inside organisations.
\n\n\n\nThese are different businesses with different investment needs. Palantir’s margins cannot establish that training frontier models is profitable. Oracle’s financing requirements cannot establish that applying AI inside a business is uneconomic.
\n\n\n\nNor do these announcements describe a single transaction chain linking all three companies. The connection is economic: demand for useful applications supports spending on models and computing capacity. The risk is that spending across those layers grows faster than customers’ willingness to pay.
\n\n\n\nAnthropic’s November 2 registration filing reports approximately $17 billion in third-quarter revenue, compared with $1.4 billion a year earlier. Its operating loss was $567 million.
\n\n\n\nThe filing also discloses approximately $895.4 billion in purchase commitments and lease agreements, substantially related to computing and infrastructure. About $69.8 billion falls due in the twelve months following September 30, 2026. Much of the remainder extends well beyond 2030.
\n\n\n\nComparing the entire commitment with one quarter’s revenue would be misleading. These obligations cover many years, and contractual conditions matter.
\n\n\n\nNevertheless, the mismatch in timing matters. Anthropic is committing to capacity before knowing precisely how demand, model prices and competitive alternatives will develop throughout those years.
\n\n\n\nMy reading is that strong adoption has reduced uncertainty about whether customers want these tools. It leaves substantial uncertainty about the margins available as the market matures.
\n\n\n\nOracle exposes a different part of the problem. According to Reuters’ reporting on the financing, it is raising approximately $10 billion through shares and another $10 billion through debt to support its expansion.
\n\n\n\nThis is an established, profitable company. Its fiscal 2026 results show approximately $17 billion in net income, but negative free cash flow of $23.7 billion as infrastructure investment absorbed cash.
\n\n\n\nAccounting profit and money available after investment measure different things. A company can report healthy earnings while needing external funding for construction and equipment.
\n\n\n\nAt its October investor day, Oracle projected positive free cash flow in the second half of fiscal 2028. Customer prepayments, customer-supplied hardware and vendor financing are expected to reduce upfront funding needs.
\n\n\n\nThat is a plan with assumptions. Its success depends on delivery, utilisation, payment and the eventual relationship between operating income and investment requirements.
\n\n\n\nPalantir’s third-quarter results provide an important counterweight. Revenue rose 96% to $2.31 billion. Operating income reached approximately $1.08 billion, producing a 47% margin under standard US accounting rules.
\n\n\n\nUS commercial revenue reached $984 million, exceeding US government revenue for the first time. The company also generated approximately $1.20 billion in operating cash flow.
\n\n\n\nThese figures demonstrate that a business selling software and AI-enabled applications can generate substantial profits today.
\n\n\n\nThey do not reveal the return achieved by every customer. Revenue records what buyers pay; it does not measure how much additional profit, reduced waste or improved service those buyers obtain.
\n\n\n\nThe distinction matters because renewals ultimately depend on customers finding sufficient reason to keep paying. Supplier profitability is evidence of a functioning business, rather than a complete audit of AI’s economic benefits.
\n\n\n\nPalantir was already an established business before the current results. Its 2023 annual figures showed $2.23 billion in revenue and $210 million in net income.
\n\n\n\nThat history changes the interpretation. Its current performance combines new technological opportunities with capabilities built earlier: customer relationships, software deployment and knowledge of complex organisations.
\n\n\n\nFor entrepreneurs, the implication is practical. Access to a capable model is only one ingredient. Distribution, integration, trust and an understanding of the customer’s problem can determine who captures value.
\n\n\n\nOracle adds another familiar lesson: expansion must often be financed before it pays for itself. The longer that interval, the more important financing terms and reliable customer payments become.
\n\n\n\nThese observations cannot settle whether AI is overvalued. They explain why useful technology, growing businesses and financial stress can coexist.
\n\n\n\nThere is another trap in reading the boom. When a model provider spends subscription revenue on cloud computing, that expenditure becomes revenue for an infrastructure supplier.
\n\n\n\nBoth transactions are real. Adding them together does not establish an equivalent amount of value created for the final customer.
\n\n\n\nThat makes independent customer evidence essential. Are businesses producing more with the same resources? Are errors falling? Do savings remain after implementation, supervision and correction?
\n\n\n\nGrowing supplier revenue answers a different question from whether an entire industry is becoming more productive.
\n\n\n\nFor a small company, the consequences appear in ordinary decisions: subscription budgets, recruitment, supplier contracts and promises made to customers.
\n\n\n\nA cheaper AI service might allow a team to offer something previously unaffordable. A price increase or tighter usage limit could weaken the same business case. A tool that accelerates drafting may still require expensive checking.
\n\n\n\nFor workers, faster task completion creates choices about workload, service quality and staffing. These financial announcements cannot tell us which choice an employer will make.
\n\n\n\nFor business owners, the useful measurement is the full cost of an accepted result, including human review. Counting generated documents or completed prompts says little about the money retained afterwards.
\n\n\n\nThree developments deserve attention.
\n\n\n\nFirst, customer renewals and expansion: whether organisations move beyond experimentation and keep spending because outcomes justify it.
\n\n\n\nSecond, supplier economics: whether model providers improve margins and infrastructure operators generate cash after investment.
\n\n\n\nThird, pricing and dependence: whether competition makes useful AI cheaper, or switching costs leave customers exposed to changing terms.
\n\n\n\nDemand could grow rapidly while prices and margins fall. Applications could flourish while particular infrastructure investments disappoint. Several outcomes can happen together.
\n\n\n\nConsider a hypothetical company that saves ten hours each week with an AI assistant. If checking its work takes eight hours, the real gain is two. If the subscription price doubles, that gain may no longer justify the expense. Another company might use the same model to remove a recurring bottleneck and earn considerably more. The technology is identical; the economics are different. That is why evidence from actual workflows will matter more than broad claims about an industry becoming more intelligent.
\n\n\n\nFor most entrepreneurs, the opportunity is to solve a customer problem using infrastructure someone else has financed. Keeping data portable and measuring results makes that opportunity easier to evaluate as suppliers and prices change.
\n\n\n\nWould your AI investment still make sense if you had to prove its return today?
\n","content_text":"Oracle is raising $20 billion while Palantir reports a quarterly operating margin of 47%. Both are expanding in AI. Their numbers describe very different experiences of the same boom.\n\nAlongside them, Claude developer Anthropic has disclosed enormous computing commitments as demand for its models grows. Read separately, these announcements look like financial news for investors. Read together, they raise a question for anyone using AI at work: can the value customers receive support the cost of supplying it?\n\nThe answer will influence which providers survive, what businesses pay and how much control customers retain over increasingly essential tools.\n\nThree Companies, Three Different Businesses\n\nAnthropic develops models. Oracle supplies cloud infrastructure alongside its established software businesses. Palantir connects data, software and AI to operational decisions inside organisations.\n\nThese are different businesses with different investment needs. Palantir’s margins cannot establish that training frontier models is profitable. Oracle’s financing requirements cannot establish that applying AI inside a business is uneconomic.\n\nNor do these announcements describe a single transaction chain linking all three companies. The connection is economic: demand for useful applications (https://reinouttebrake.com/stories/the-biggest-change-in-ai-is-happening-behind-your-chat-window) supports spending on models and computing capacity. The risk is that spending across those layers grows faster than customers’ willingness to pay.\n\nAnthropic: Growing Demand Meets Long Commitments\n\nAnthropic’s November 2 registration filing (https://www.sec.gov/Archives/edgar/data/2130357/000162828026068676/anthropicpbc-sx1ax1.htm?utm_source=chatgpt.com) reports approximately $17 billion in third-quarter revenue, compared with $1.4 billion a year earlier. Its operating loss was $567 million.\n\nThe filing also discloses approximately $895.4 billion in purchase commitments and lease agreements, substantially related to computing and infrastructure. About $69.8 billion falls due in the twelve months following September 30, 2026. Much of the remainder extends well beyond 2030.\n\nComparing the entire commitment with one quarter’s revenue would be misleading. These obligations cover many years, and contractual conditions matter.\n\nNevertheless, the mismatch in timing matters. Anthropic is committing to capacity before knowing precisely how demand, model prices and competitive alternatives will develop throughout those years.\n\nMy reading is that strong adoption has reduced uncertainty about whether customers want these tools. It leaves substantial uncertainty about the margins available as the market matures.\n\nOracle: Profitable Does Not Mean Self-Funding\n\nOracle exposes a different part of the problem. According to Reuters’ reporting on the financing (https://www.schwab.wallst.com/schwab/Prospect/research/stocks/news.asp?doc=3200-nL3N46J11L-10&symbol=MSFT&utm_source=chatgpt.com), it is raising approximately $10 billion through shares and another $10 billion through debt to support its expansion.\n\nThis is an established, profitable company. Its fiscal 2026 results (https://investor.oracle.com/investor-news/news-details/2026/Oracle-Announces-Record-Q4-and-FY-2026-Results-Driven-by-Cloud-Infrastructure--Cloud-Applications/?utm_source=chatgpt.com) show approximately $17 billion in net income, but negative free cash flow of $23.7 billion as infrastructure investment absorbed cash.\n\nAccounting profit and money available after investment measure different things. A company can report healthy earnings while needing external funding for construction and equipment.\n\nAt its October investor day (https://investor.oracle.com/investor-news/news-details/2026/Oracle-Investor-Day-2026-Accelerating-Value-Creation/default.aspx?utm_source=chatgpt.com), Oracle projected positive free cash flow in the second half of fiscal 2028. Customer prepayments, customer-supplied hardware and vendor financing are expected to reduce upfront funding needs.\n\nThat is a plan with assumptions. Its success depends on delivery, utilisation, payment and the eventual relationship between operating income and investment requirements.\n\nPalantir: Applications Are Already Making Money\n\nPalantir’s third-quarter results (https://www.sec.gov/Archives/edgar/data/1321655/000132165526000045/a2026q3ex991pressrelease.htm?utm_source=chatgpt.com) provide an important counterweight. Revenue rose 96% to $2.31 billion. Operating income reached approximately $1.08 billion, producing a 47% margin under standard US accounting rules.\n\nUS commercial revenue reached $984 million, exceeding US government revenue for the first time. The company also generated approximately $1.20 billion in operating cash flow.\n\nThese figures demonstrate that a business selling software and AI-enabled applications can generate substantial profits today.\n\nThey do not reveal the return achieved by every customer. Revenue records what buyers pay; it does not measure how much additional profit, reduced waste or improved service those buyers obtain.\n\nThe distinction matters because renewals ultimately depend on customers finding sufficient reason to keep paying. Supplier profitability is evidence of a functioning business, rather than a complete audit of AI’s economic benefits.\n\nWhat The Past Helps Us Understand\n\nPalantir was already an established business before the current results. Its 2023 annual figures (https://investors.palantir.com/news-details/2024/Palantir-Reports-Its-Fifth-Consecutive-Qua?utm_source=chatgpt.com) showed $2.23 billion in revenue and $210 million in net income.\n\nThat history changes the interpretation. Its current performance combines new technological opportunities with capabilities built earlier: customer relationships, software deployment and knowledge of complex organisations.\n\nFor entrepreneurs, the implication is practical. Access to a capable model is only one ingredient. Distribution, integration, trust (https://reinouttebrake.com/stories/ai-is-making-games-easier-to-build-getting-players-is-the-hard-part) and an understanding of the customer’s problem can determine who captures value.\n\nOracle adds another familiar lesson: expansion must often be financed before it pays for itself. The longer that interval, the more important financing terms and reliable customer payments (https://reinouttebrake.com/stories/ai-financing-through-spvs-quietly-shifts-risk-and-debt) become.\n\nThese observations cannot settle whether AI is overvalued. They explain why useful technology, growing businesses and financial stress can coexist.\n\nWhy The Numbers Cannot Simply Be Added Together\n\nThere is another trap in reading the boom. When a model provider spends subscription revenue on cloud computing, that expenditure becomes revenue for an infrastructure supplier.\n\nBoth transactions are real. Adding them together does not establish an equivalent amount of value created for the final customer.\n\nThat makes independent customer evidence essential. Are businesses producing more with the same resources? Are errors falling? Do savings remain after implementation, supervision and correction?\n\nGrowing supplier revenue answers a different question from whether an entire industry is becoming more productive.\n\nHow This Reaches Our Businesses And Working Lives\n\nFor a small company, the consequences appear in ordinary decisions: subscription budgets, recruitment, supplier contracts and promises made to customers.\n\nA cheaper AI service might allow a team to offer something previously unaffordable. A price increase or tighter usage limit could weaken the same business case. A tool that accelerates drafting may still require expensive checking.\n\nFor workers, faster task completion creates choices about workload, service quality and staffing. These financial announcements cannot tell us which choice an employer will make.\n\nFor business owners, the useful measurement is the full cost of an accepted result (https://reinouttebrake.com/ai-for-business#accepted-output), including human review. Counting generated documents or completed prompts says little about the money retained afterwards.\n\nWhat To Watch Next\n\nThree developments deserve attention.\n\nFirst, customer renewals and expansion: whether organisations move beyond experimentation and keep spending because outcomes justify it.\n\nSecond, supplier economics: whether model providers improve margins and infrastructure operators generate cash after investment.\n\nThird, pricing and dependence: whether competition makes useful AI cheaper, or switching costs leave customers exposed to changing terms.\n\nDemand could grow rapidly while prices and margins fall. Applications could flourish while particular infrastructure investments disappoint. Several outcomes can happen together.\n\nConsider a hypothetical company that saves ten hours each week with an AI assistant. If checking its work takes eight hours, the real gain is two. If the subscription price doubles, that gain may no longer justify the expense. Another company might use the same model to remove a recurring bottleneck and earn considerably more. The technology is identical; the economics are different. That is why evidence from actual workflows will matter more than broad claims about an industry becoming more intelligent.\n\nFor most entrepreneurs, the opportunity is to solve a customer problem using infrastructure someone else has financed. Keeping data portable and measuring results makes that opportunity easier to evaluate as suppliers and prices change.\n\nWould your AI investment still make sense if you had to prove its return today?","referenced_links":[{"url":"https://reinouttebrake.com/stories/the-biggest-change-in-ai-is-happening-behind-your-chat-window","label":"demand for useful applications"},{"url":"https://www.sec.gov/Archives/edgar/data/2130357/000162828026068676/anthropicpbc-sx1ax1.htm?utm_source=chatgpt.com","label":"November 2 registration filing"},{"url":"https://www.schwab.wallst.com/schwab/Prospect/research/stocks/news.asp?doc=3200-nL3N46J11L-10&symbol=MSFT&utm_source=chatgpt.com","label":"Reuters’ reporting on the financing"},{"url":"https://investor.oracle.com/investor-news/news-details/2026/Oracle-Announces-Record-Q4-and-FY-2026-Results-Driven-by-Cloud-Infrastructure--Cloud-Applications/?utm_source=chatgpt.com","label":"fiscal 2026 results"},{"url":"https://investor.oracle.com/investor-news/news-details/2026/Oracle-Investor-Day-2026-Accelerating-Value-Creation/default.aspx?utm_source=chatgpt.com","label":"October investor day"},{"url":"https://www.sec.gov/Archives/edgar/data/1321655/000132165526000045/a2026q3ex991pressrelease.htm?utm_source=chatgpt.com","label":"third-quarter results"},{"url":"https://investors.palantir.com/news-details/2024/Palantir-Reports-Its-Fifth-Consecutive-Qua?utm_source=chatgpt.com","label":"2023 annual figures"},{"url":"https://reinouttebrake.com/stories/ai-is-making-games-easier-to-build-getting-players-is-the-hard-part","label":"Distribution, integration, trust"},{"url":"https://reinouttebrake.com/stories/ai-financing-through-spvs-quietly-shifts-risk-and-debt","label":"financing terms and reliable customer payments"},{"url":"https://reinouttebrake.com/ai-for-business#accepted-output","label":"the full cost of an accepted result"}],"social_image":"https://reinouttebrake.com/assets/editorial/servers-1280.webp","citation":"Reinout te Brake. AI Is Growing Fast. Who Will Pay for It?. 2026-11-03. https://reinouttebrake.com/stories/ai-is-growing-fast-who-will-pay-for-it"}