AI model companies want to grab the application layer too

The article, published on July 13, 2026 in Forward Future's weekly 'Zeitgeist' roundup, raises a cross-cutting concern that runs through several of the week's stories: the companies that control AI models, chips and the underlying infrastructure may also be encroaching on the application companies built on top of that infrastructure. According to the program's team, this tension is playing out on at least four distinct fronts: talent moves, disputes over data access, product usage policies, and NVIDIA's backing of open models.

On talent, it is highlighted that Tom Blomfield, managing partner at Y Combinator, is taking a leave of absence to join Anthropic's compute team. The program's team reads this decision as a bet that there is more competitive advantage in controlling models and compute capacity than in building applications on top of them. In the opposite direction, it is noted that Anthropic's head of design has reportedly left the company to join Cursor, a move the team views favorably for a company like Cursor, which is focused on making AI-assisted programming tools more accessible.

Another of the cases discussed is the security dispute between Apple and OpenAI. Apple accuses a former hardware engineer who joined OpenAI of keeping a company laptop and exploiting a vulnerability to access its internal network storage. Apple reportedly detected a message about that access on its internal system before the conversation moved to Signal. The program's team interprets this episode as a sign that the competition between Apple and OpenAI is no longer limited to products, but now extends to employees, intellectual property and computer security.

A third theme is the resharing of a post by Satya Nadella that revisits Arrow's so-called 'information paradox': buyers need to see information in order to value it, but showing it means revealing exactly what one is trying to sell. The program's team applies that idea to the AI context: when the buyer of a service or data is also a model provider, the customer's context can be used to improve that model, and the provider could use what it learns to enter the customer's own market. The article distinguishes this risk from that posed by a conventional cloud provider, which only observes usage patterns, versus a model provider, which could learn patterns directly from the content of the work being done.

It is also mentioned that OpenAI temporarily removed Codex's five-hour usage window just as Anthropic users were facing service outages and stricter limits. The program's team reads the timing as a competitive maneuver aimed at capturing frustrated Anthropic users. One of the hosts, identified as Matthew, further notes that expiring credits and reset windows incentivize people to consume more compute before they lose access to it.

The roundup also picks up a public statement signed by prominent economists and AI researchers, available at wemustactnow.ai, calling on governments and institutions to prepare for the economic effects of AI through better incentives and safeguards. The team feels the list of signatories lends credibility to the effort, but considers the proposal too generic to evaluate: they call for concrete policies, identified people responsible, and measurable objectives, rather than a general call to 'prepare.'

Finally, the article explains why NVIDIA backs the development of open models: the company benefits when cheaper, more accessible models increase demand for inference and, with it, demand for its chips. The team links this to Jevons' paradox, according to which lowering the cost of a resource can increase its total consumption rather than reducing it. In addition, it is noted that open models pose less risk for those building applications, since they are easier to run independently or to replace if the original provider ends up becoming a direct competitor.

Taken together, the through-line of the roundup is that, as providers of models, chips and infrastructure gain market power, structural incentives emerge for them to compete with the very companies that depend on them—whether by poaching their talent, gaining privileged access to their customers' data, making opportunistic product decisions, or controlling how open or closed the available models are.