If the artificial intelligence industry were pitched as a television drama, no network executive would buy it. Too many characters. Too many implausible reversals. Too much money. Nobody would believe that a founder fired by his own board could return five days later with more power than he had before. Or that the former colleagues who left to prevent his excesses would build a rival company that eventually stood accused of using safety to suppress competition. Or that the world’s dominant chipmaker would finance customers so they could buy more of its chips.
Then there is Elon Musk, who plays the spurned co-founder, rival suitor, courtroom antagonist and self-appointed guardian of the original faith—frequently within the same episode. Mark Zuckerberg has been recast as the champion of open AI. Satya Nadella is the patient power broker who financed Sam Altman, rescued him, embedded OpenAI across Microsoft and is now quietly making certain Microsoft can survive without him. Andy Jassy is investing billions in Anthropic while ensuring AWS remains the indispensable party in the relationship. Jensen Huang sells the weapons to almost everyone, finances some of the combatants and somehow remains welcome in every room.
Larry Ellison and Masayoshi Son arrive with infrastructure, financing and promises large enough to make yesterday’s enormous commitments look quaint. The supporting cast includes Tim Cook, the Pentagon, the White House, sovereign wealth funds, federal regulators and a growing collection of Chinese model makers who are portrayed as technologically inferior, economically dangerous and existentially threatening—sometimes in the same paragraph.
They compete with one another, invest in one another, sue one another, supply one another and periodically release joint statements about protecting humanity.
Welcome to As the AI World Turns.
This is the first installment of an occasional series about the personalities, shifting alliances, circular financing, convenient principles and increasingly public betrayals shaping the artificial intelligence economy. It will be slightly satirical because pretending none of this is ridiculous would be dishonest. It will also be bitingly serious because this cast of characters is fighting over who will control the infrastructure, models, applications, data and customer relationships at the center of the emerging intelligence economy.
The soap opera is not a distraction from the real story. It is the real story.
The Conscience Has a New Product
Our first episode belongs to Anthropic, the company founded by former OpenAI executives who wanted to build a safer, more principled alternative to the organization they left behind.
That origin story gave Anthropic a valuable role in the AI drama. OpenAI had Sam Altman, the slippery visionary who could turn almost any crisis into further evidence of his destiny. Anthropic had Dario Amodei, the sober scientist warning that the technology was moving too quickly and needed stronger guardrails.
OpenAI wanted to capture the future. Anthropic wanted to make sure the future survived.
It was always a little too tidy.
Now, according to The Wall Street Journal, a backlash is brewing against Anthropic across Silicon Valley. The complaints include Anthropic’s position on open-weight models, its efforts to prevent Chinese competitors from distilling Claude’s capabilities, its data-retention requirements and a growing suspicion that the company is using legitimate safety concerns to strengthen its commercial position.
But the controversy that captures the larger problem involves Figma.
Figma was not merely another company watching Anthropic from across the valley. It was a customer and partner. Figma incorporated Claude into its products and reportedly worked closely with Anthropic on AI-powered design capabilities. Then Anthropic launched Claude Design, a product that moved directly into the territory occupied by Figma and other design-software companies.
The timing made the maneuver harder to dismiss as ordinary competition. Mike Krieger, then an Anthropic executive, resigned from Figma’s board just three days before Anthropic announced Claude Design. Figma CEO Dylan Field subsequently said Anthropic had not been “consistently candid” in its communications, according to Upstarts.
There may have been no contractual violation. There is no public evidence that Anthropic misappropriated private Figma information, and it would be irresponsible to claim otherwise. Anthropic also did not need Figma to teach it that generative AI would eventually move into visual design.
But contracts are not the only things that hold an ecosystem together. Trust matters. So does the reasonable expectation that a strategic partner is not using the relationship to obtain a better view of where it should attack you next.
The uncomfortable accusation is that Anthropic cozied up to partners, absorbed what it could learn about their products, workflows and customers, and then moved into their markets. Figma and other application companies performed the hard work of identifying valuable use cases, designing usable experiences and proving that customers would pay for them. Anthropic occupied a privileged position underneath those applications, watching where the value accumulated.
Today’s strategic partner became tomorrow’s unpaid product-research department.
Moving Up the Stack
This is about more than Anthropic launching a design tool. It is a textbook example of the dynamic I explore in my forthcoming book, The Indispensability Trap.
Being indispensable to another company is not the same as owning the customer.
Anthropic may provide extraordinarily capable models, but if Figma owns the workflow, interface and customer relationship, Anthropic remains a supplier. Suppliers can be squeezed. Their products can be commoditized. Customers can add a second model, route workloads to a cheaper provider or replace the underlying technology without asking the end user’s permission.
Anthropic understands that remaining the intelligence inside somebody else’s product leaves it vulnerable to the company controlling the layer above it. The escape is to move up the stack—from model provider to application platform, and from application platform into the applications themselves.
The problem is that Anthropic’s customers understand this too.
Once a model provider begins competing with the companies building on its models, every partnership starts to look like competitive reconnaissance. Every integration gives the model company a better view of customer behavior. Every joint product effort teaches it more about the workflow. Every success reveals another application it might absorb.
Anthropic is hardly alone. OpenAI has expanded from foundation models into coding, search, browsers, shopping, productivity tools and enterprise applications. Microsoft finances OpenAI while developing its own models, copilots and agents. Amazon backs Anthropic while building alternatives and controlling the infrastructure underneath it. Google already owns pieces of nearly every layer. Nvidia is moving from chips into networking, systems, software, models, cloud services and complete AI factories.
Everyone praises the ecosystem while quietly studying which parts of it can be swallowed next.
Chipmakers move into systems. Systems companies move into cloud infrastructure. Clouds move into models. Model companies move into applications. Application companies fight to own workflows, data and customer relationships. Nobody wants to remain the layer that can be squeezed, commoditized or replaced.
That is why this industry produces such strange alliances. Everyone is somebody else’s partner, supplier, customer, investor and existential threat. Frequently, they are all five at once.
Safety, Sincerely and Conveniently
The other complaints against Anthropic fit the same pattern.
Anthropic has faced criticism for declining to join more than 70 technology companies supporting open-weight AI. Amodei responded that Anthropic has never advocated banning open-weight models and described less-capable open models as a public good. His concern is with highly capable downloadable models that could spread dangerous biological, cyber or national-security capabilities without meaningful control. Anthropic’s position is more nuanced than its critics sometimes acknowledge.
It is also remarkably convenient.
Restrictions on the most capable open-weight models would benefit companies whose businesses depend on customers accessing intelligence through interfaces they control. If the best models cannot be downloaded, modified and operated independently, customers remain dependent on Anthropic, OpenAI and the other closed providers.
That does not mean Amodei is fabricating the danger. Principle and commercial advantage are not mutually exclusive. Safety can be a sincere conviction and a formidable competitive moat.
The same tension appears in Anthropic’s campaign against model distillation. The company says DeepSeek, Moonshot AI and MiniMax generated more than 16 million Claude exchanges through roughly 24,000 fraudulent accounts to extract capabilities for their own models. Anthropic’s detailed account describes coordinated account networks, proxy services and efforts to evade regional restrictions.
That is not casual experimentation. Anthropic has a legitimate interest in protecting its systems from industrial-scale extraction.
But distillation is also a standard AI development technique. Frontier labs use stronger models to train smaller ones. The entire AI industry was built by training models on vast quantities of human-created material, much of it collected without individual permission. The companies that absorbed the open web cannot now casually declare that knowledge becomes proprietary the moment it passes through their models.
Apparently everyone is allowed to suck someone else’s brain. The dispute is over whose brain, at what scale and whether the sucking happens to threaten an incumbent’s margins.
Anthropic’s 30-day data-retention requirement for customers accessing some of its most advanced models presents another contradiction. The company says retention is necessary to detect and mitigate jailbreaks and acknowledges that the policy imposes real commercial costs. That may be true. It also asks enterprises to surrender a measure of sovereignty and privacy to Anthropic’s safety apparatus.
The company is effectively saying: Trust us with greater control because we understand the risks better than you do.
Perhaps it does. Concentrated power does not become harmless merely because the people exercising it have good intentions.
Next Week, Someone Else Gets Betrayed
Anthropic deserves credit for taking positions that have cost it politically and commercially. Its willingness to stand behind model safeguards during its battles with the Pentagon required more backbone than most technology companies have demonstrated. Its concerns about bioweapons, cyberattacks and uncontrolled frontier capabilities should not be dismissed as corporate theater.
But Anthropic cannot serve simultaneously as the industry’s conscience, infrastructure provider, application competitor and regulatory gatekeeper without expecting customers to question which role is driving each decision.
The company founded to prevent OpenAI’s excesses increasingly looks like another powerful lab trying to control as much of the AI stack as possible. The conscience did not merely become the gatekeeper. It discovered that controlling the gate was insufficient and began moving into the businesses on the other side.
That is what makes Anthropic the perfect first episode of As the AI World Turns. The industry’s roles remain unsettled. Nobody knows who will ultimately become the platform, supplier, application, utility—or casualty. Every company is racing to become indispensable before the market hardens around someone else.
So the alliances shift. The financing travels in circles. The principles bend under commercial pressure. Yesterday’s rebel becomes today’s incumbent, and today’s customer discovers that its most important supplier has been working on a remarkably familiar product.
In As the AI World Turns, every partnership contains the seeds of the next betrayal. The strategic partner smiling across the conference table may already be writing your product into next season’s script.