Summer of 2026: AI Wrap

Here's what hasn't changed in the AI market, yet still fills the news cycle:

• There could be an AI bubble. Companies such as Anthropic and OpenAI, as well as big tech, are spending unprecedented amounts of money to build AI infrastructure, train models, and more. The publicly traded tech giants are growing their spending faster than their revenue. Both Anthropic and OpenAI have filed confidentially to go public, and both have had to share more financials. Anthropic told investors its annualized revenue run rate reached $65B in July. OpenAI's is reportedly about $70B. Neither is profitable.
  

• Large tech firms are spending a lot (see above) on AI infrastructure. Big tech's spending on data centers and AI has roughly quintupled in three years, from about $150B in 2023 to more than $700B planned for 2026. R&D has grown too, but far more slowly. Most of the AI money goes into buildings, chips, and power (capital spending), not research staff. They are issuing bonds to finance the spending.
  

• The cost of capital is increasing. Large tech companies, especially hyperscalers, used to put about half of their operating cash flow toward capital spending. Now that share is approaching 100%. Many were able to raise tens of billions through bonds at a modest cost, given their high credit ratings. Costs are rising because of higher Treasury yields, geopolitical risks, and uncertainty about returns. Here's roughly what they have raised, mostly in the past two years: Amazon ($79B), Alphabet ($50B), Meta ($65B), and Oracle (>$40B).
  

• Circular funding agreements continue, with Nvidia and the hyperscalers at the core. The potential of the AI market is still uncertain, even if everyone agrees it is large. Partly as a result, insiders are funding the growth, and they have the balance sheets to do so. Nvidia's revenue grew from $26.9B in FY2022 to $215.9B in FY2026. Over the same period, net income skyrocketed from $9.8B to $120.1B. Nvidia agreed to buy Hugging Face for $12.9B. It has also invested tens of billions in Anthropic and OpenAI, money that essentially comes back as chip purchases. More recently, Nvidia announced partnerships with Wall Street firms to mobilize more than $500B for AI infrastructure (i.e., in part, buying chips).
  

• Open-weight models are threatening OpenAI's and Anthropic's business models. Both OpenAI and Anthropic have sought protection from "China." Essentially, a handful of Chinese companies (e.g., DeepSeek, Moonshot AI) are accused of distilling their models to build LLMs at a far lower cost than it took to create the originals. As a result, these same companies can sell access to their models at far lower prices. Both Washington and the tech giants try to position China as the common or greater enemy, even an existential threat to our economy. Most of Silicon Valley wants more competition, i.e., open-weight models.
  

• AI can be unsafe. When a CEO's primary job is to deliver returns to shareholders, most discerning folks find it difficult to trust the altruism in their essays and statements calling for safeguards. At the UN in September, Dario Amodei called on AI companies to slow the pace of capability improvements, and Sam Altman agreed. Neither has said what that will mean in practice. AI is considered unsafe because, to completely oversimplify, we don't fully understand why it does what it does. While most of us may not care if OpenAI's agents hack Hugging Face, we will care if it affects our banks, water, personal data, and more.
  

• Governments are not equipped to create the guardrails citizens need. To be fair, AI is evolving quickly. Our governments have a lot of competing priorities. The United States has midterm elections in about four weeks. Local or even state-level regulations would create a piecemeal system that is difficult for any company to navigate. The current administration doesn't seem to have an appetite to slow down progress; again, the threat is China. Citizens want accountability. Tech giants want Section 230-style protection from what their models may do. And the optics are tough when tech moguls donate huge sums to Donald Trump or his ballroom. Greg Brockman gave $25M to Trump's super PAC, MAGA Inc. Musk gave more.
  

• Citizens' trust is low in both Washington and Silicon Valley. Tech giants made huge fortunes by monetizing consumer data after giving consumers free services such as Email, calendar, social media, etc. Then new and existing tech giants trained their foundation models on Internet content they found but did not have explicit permission to use. A 2025 Pew Research Center study found that just 17% of Americans trust the federal government to do what is right "just about always" or "most of the time." A June 2026 Gallup poll reported that confidence in large tech companies fell to 20%, a record low. Confidence is different from trust, but the two are related when we're talking about confidence in an entity to do the right thing.
  

• Populations are pushing back against data centers. Governors in several states, including New York, Texas, and Pennsylvania, have paused or restricted large-scale data center projects until they can better assess the impact on water, energy, and the environment. Maine's legislature passed a statewide moratorium, but the governor vetoed it. Dozens of municipalities have their own moratoriums. In case you don't read a lot about data centers, the other issue is that they don't create many jobs beyond the initial construction. The business of data centers is still strong. I spoke to the CMO of a company that provides key components. I asked if they were just going to stop spending money on marketing because they are growing so fast. The answer was "no."