Most people drive by a building in Menlo Park without giving it much thought. According to those who have worked there, engineers spend their days fine-tuning systems that are so sensitive to human psychology that the difference between versions can be measured in minutes of sleep lost. These changes are not announced. A press conference is never held. Someone scrolls a bit longer somewhere, and the algorithm simply becomes a little more precise.
That’s the subdued version of the situation. The more dramatic version includes White House executive orders, classified government reviews, hundreds of billions of dollars, and a race so fierce that the participating companies have started employing actual military strategists. The algorithm wars in Silicon Valley are no longer limited to boardrooms. Geopolitics has been affected by them.
Even if the specifics remain hidden, the main conflict is easy to explain. A small number of businesses are vying to develop AI systems that can control vast portions of daily life, including Google, Meta, Microsoft, Amazon, and a number of well-funded rivals like Anthropic and xAI. which loans are accepted. which job applications are processed by a human. Who gets the news? Which weapon targeting system is approved? These decisions are made by algorithms that are not impartial. They show decisions about power, priorities, and the interests that the system ultimately serves.
The extent to which the government has been drawn into the conflict—not as a regulator standing outside the arena, but as a participant with its own bets placed—is what sets this moment apart from previous tech rivalries. The most important frontier-model decisions were transferred to a classified review process by President Trump’s June executive order, with the companies that developed the systems seated at the same table as the officials tasked with evaluating them. Before a previous draft was changed, Elon Musk and Mark Zuckerberg allegedly had direct conversations with the White House. Whatever was said during those discussions, the dynamic that emerged is worth considering: the officials tasked with examining the algorithms and the engineers who write them are, in a significant way, the same individuals.
In the meantime, military applications are progressing more quickly than nearly everyone admits in public. The Pentagon’s first attempt to use AI for drone targeting analysis, Project Maven, was controversial when it first started in 2017, in part because Google employees voiced their disapproval so loudly that the company ultimately withdrew from the contract. Now, that controversy seems almost nostalgic. Investors and defense executives described AI integration into warfare as something that is already happening at the Fortune Brainstorm Tech summit in Aspen last week. They cited the recent use of U.S. missiles against Iran as proof that current supply chains and decision-making structures are already being strained by the speed of it all.
Observing all of this, it seems as though the algorithm wars have gained momentum of their own, with the competition itself now making choices that no individual would have made. Anthropic has admitted that Claude Mythos, its most sophisticated model, is too potent to make available to the general public. Instead, private businesses are testing it. When a competitor is several months behind and a government contract is at stake, it might be more difficult to exercise such restraint.
Although they don’t often make headlines, the civilian costs of this race are quietly mounting. Persuasive design, which was first developed at Stanford’s Persuasive Technology Lab and involves engineering apps to maximize psychological engagement, has effectively redefined what it means to grow up with a phone, according to research dating back years. Systems that determine creditworthiness and medical eligibility are now using the same optimization logic that keeps teenagers on Instagram at midnight. In both situations, the algorithm is carrying out its intended function. That’s exactly why it’s something to be concerned about.
It’s difficult to ignore the fact that either those developing the systems or those purchasing them have the loudest public voices in this discussion. As of right now, everyone else is awaiting word on the decision.
