# Resistance Is Futile. Indispensability Is Temporary. Commoditization Is Inevitable > Source: > Published: 2026-08-04 07:47:42+00:00 TL;DR — Key Takeaways - Amazon, Alphabet, Microsoft and Meta are preparing to spend roughly **$700 billion or more** on AI infrastructure this year, including chips, data centers, networking and energy. - The spending is beginning to consume cash: Amazon’s trailing free cash flow fell to negative $7.6 billion, while Alphabet reported negative quarterly free cash flow of $5.9 billion. - AI demand is genuine, but that does not guarantee every data center, GPU cluster or power contract will generate an acceptable return. - The massive buildout will eventually increase capacity, reduce scarcity and commoditize parts of the AI infrastructure market. - The biggest long-term winners may be the companies that recognize where value is moving and shift upward before their current advantage disappears. Resistance is futile. That may sound like surrender, but it is really a description of where the AI infrastructure race stands today. Amazon, Alphabet, Microsoft and Meta are preparing to spend roughly $700 billion or more this year, much of it on data centers, chips, networking and the power required to run them. Amazon has raised its 2026 capital spending plan to approximately $220 billion. Alphabet now expects to spend between $195 billion and $205 billion. Microsoft and Meta are making commitments on a similarly historic scale. As a recent [New York Times examination of the buildout](https://www.nytimes.com/2026/07/30/technology/amazon-google-ai-data-center-spending.html) made clear, this is no longer money being skimmed from the surplus cash generated by some of the most profitable companies in history. The spending is beginning to consume the cash. Amazon reported that its trailing 12-month free cash flow fell from a positive $18.2 billion to a negative $7.6 billion, driven primarily by increased property and equipment purchases related to AI. Alphabet spent $44.9 billion on capital expenditures in the second quarter alone and reported negative quarterly free cash flow of $5.9 billion. Neither company is in imminent financial danger, but that is not the point. The point is that the scale of the AI buildout is now large enough to change the financial character of the companies leading it. [Amazon’s second-quarter results](https://ir.aboutamazon.com/news-release/news-release-details/2026/Amazon-com-Announces-Second-Quarter-Results/) and [Alphabet’s earnings call](https://abc.xyz/investor/events/event-details/2026/2026-Q2-Earnings-Call-2026-GgTAq7Is0z/default.aspx) show just how quickly that shift is occurring. It would be easy to take these figures and write another article asking whether AI is a bubble. It would also be mostly pointless. “Bubble or no bubble?” forces a binary answer onto a situation in which two competing realities can be true at once. The demand is real. AWS grew 37% in the second quarter, its fastest growth in 18 quarters. Google Cloud grew 82%, and its backlog reached $514 billion. Enterprises are deploying AI. Governments consider it a strategic capability. Consumers are using it in search, software, content creation and a rapidly expanding range of everyday activities. The infrastructure being built will not sit entirely idle. But real demand does not prove that every data center, every power contract, every GPU cluster and every dollar of private credit will earn an acceptable return. A technology can transform the economy while the investment cycle surrounding it becomes financially excessive. The railroads did. Telecommunications did. The internet did. AI can too. ### Resistance Is Futile Every major participant in this market believes it cannot afford to sit out the buildout. That belief has become self-reinforcing. If Amazon spends $220 billion, Alphabet cannot simply declare that it will conserve cash and hope its existing infrastructure is sufficient. If Microsoft adds capacity, Google must respond. If Meta secures gigawatts of power and hundreds of thousands of accelerators, every company attempting to build a frontier model has to consider what falling behind would mean. Once competitors commit hundreds of billions of dollars, declining to match them begins to look more dangerous than overbuilding. The same logic is spreading beyond the hyperscalers. National governments view AI infrastructure as a sovereignty and security issue. Enterprises increasingly fear that a lack of access to models, compute or data will leave them structurally disadvantaged. Utilities, real estate developers, private credit funds, chipmakers and data center operators are reorganizing around projections of seemingly inexhaustible demand. The decision is no longer merely whether an individual project offers an attractive return. It is whether a company or country can risk being absent from what may become the defining technology platform of the next several decades. That is why resistance is futile. It does not mean every proposed data center should be approved or that every investment thesis is sound. It means the strategic momentum behind the buildout has become too powerful for conventional financial caution to stop it. Every participant may recognize the possibility of excess, yet each has a rational reason to keep spending. Collectively, those rational decisions can still produce an irrational result. ### Indispensability Is Temporary The buildout is also a scramble to control whichever layer the rest of the market cannot function without. NVIDIA wants its GPUs, networking and software ecosystem to remain essential to AI development. The hyperscalers want their clouds to become the unavoidable homes for training and inference. Model providers want their intelligence embedded in every application, workflow and device. Memory manufacturers, networking suppliers, data-center operators and utilities have each discovered moments when scarcity makes their particular contribution appear indispensable. This is the dynamic at the heart of my forthcoming book, *The Indispensability Trap*. Every company wants to become the layer everyone else needs. Indispensability brings pricing power, influence and the ability to capture an outsized share of the value created by the broader ecosystem. It also puts a target on your back. Customers do not enjoy being dependent on a single supplier. Competitors attack the largest profit pools. Governments fund sovereign alternatives. Cloud providers design their own chips. Model builders look for alternatives to expensive accelerators. Enterprises adopt multicloud and multimodel strategies. Open-weight models place pressure on proprietary pricing. Efficiency improvements allow smaller models to perform tasks that once required access to the frontier. The more profitable and strategically important a bottleneck becomes, the more incentive the market has to engineer around it. Scarcity attracts capital. Capital creates capacity. Capacity creates competition. Competition erodes the scarcity that made the provider indispensable in the first place. Today’s choke point becomes tomorrow’s interchangeable component. That does not mean NVIDIA, Amazon, Microsoft, Google or the leading model companies are destined to disappear. Indispensability can last long enough to create enormous companies and enduring advantages. But it is never a comfortable permanent state. The market is always working to reduce its dependence on whoever currently controls the bottleneck. Indispensability comes with an expiration date, even when no one can say precisely when it will arrive. ### Commoditization Is Inevitable Here is the paradox. The companies spending hundreds of billions of dollars to protect their control over AI infrastructure are also financing the capacity that will eventually weaken its scarcity value. More accelerators will come to market. More custom chips will challenge general-purpose GPUs. More data centers will open. More energy projects will be built. Models will become more efficient. Inference prices will decline. Performance at the lower and middle sections of the market will converge. Customers will gain more choices. The current buildout is creating the conditions for the commoditization of compute and intelligence. Commoditization does not mean that everything becomes cheap, perfectly competitive or equally good. AI infrastructure will remain enormously capital-intensive. Power access, supply chains, distribution and scale will continue to favor large companies. Frontier development may remain concentrated among a relatively small number of players. What will change is where the value accumulates. As chips become more available, chipmakers move into systems, networking and software. As cloud compute becomes more interchangeable, hyperscalers move into data platforms, models, agents and business applications. As models become less differentiated, model providers move toward proprietary data, workflows, distribution and measurable business outcomes. Everyone moves up the stack because the layer beneath them is becoming a commodity. This is the recurring escape attempt within the indispensability trap: become essential, harvest the advantage and climb into the next layer before the market commoditizes the one you currently control. The companies that succeed will not necessarily be those that own the most infrastructure. They will be those that recognize when their present advantage is starting to decay and move before their customers do. The [second New York Times article](https://www.nytimes.com/2026/07/30/technology/ai-bubble-venture-capital.html) adds another, more troubling dimension. Some in Silicon Valley are already arguing that an AI bubble might be acceptable. Bubbles, in this telling, finance infrastructure faster than a cautious market would, eliminate weaker companies and leave society with useful assets. The fiber-optic overbuild of the dot-com era is usually offered as the exhibit. There is truth in that argument. The telecom bust destroyed investors and companies, but much of the fiber remained and eventually supported the modern internet. A failed AI investment cycle could leave behind power generation, transmission connections, data-center shells and a much larger base of technical expertise. But GPUs are not fiber. They become obsolete much more quickly. Data centers designed around one generation of power density and cooling do not automatically retain their economics through the next. Even where the infrastructure remains useful, that does not make the losses harmless. A bubble may be “just fine” for venture funds with diversified portfolios, founders who sold shares along the way and hyperscalers positioned to acquire distressed assets. It may look different to utility customers paying for grid upgrades, communities that made long-term water and power commitments, pension funds and bondholders financing the projects, or employees and contractors caught in the retrenchment. The risk is already spreading. Hyperscalers had issued roughly $194 billion in bonds through early July, up 79% from the previous year, and Goldman Sachs estimated that debt is financing about one-third of current capital spending. Investor demand has begun to cool even as projected issuance continues to rise. [Reuters reported](https://www.reuters.com/business/hyperscaler-debt-binge-pushes-yields-up-investor-demand-cools-2026-07-29/) that the buildout is placing increasing pressure on credit markets. That raises the question Silicon Valley’s bubble rationalization too often skips: If AI infrastructure becomes indispensable to the economy but uneconomic for many of its original investors, who ultimately owns the risk? Strategic necessity can quickly become an argument for rescue. Once data centers, power projects and model providers are considered essential to national competitiveness, allowing them to fail becomes politically difficult. The gains remain private during the boom, while the losses migrate outward when the economics fail. That does not mean we should stop building. It means we should stop pretending that a bust would be painless simply because some useful infrastructure would survive it. All three parts of the title can be true at the same time. Resistance is futile because the AI buildout has acquired overwhelming strategic momentum. Indispensability is temporary because every bottleneck attracts competitors, substitutes and customers determined to escape dependency. Commoditization is inevitable because massive investment eventually turns scarcity into capacity. The buildout may succeed beyond today’s most optimistic forecasts, and many of the investments financing it may still fail. The long-term winners will not necessarily be the companies that spend the most money or build the largest clusters. They will be the ones that see where value is moving next and escape their moment of indispensability before it becomes their trap.