The companies racing to build artificial intelligence are not racing each other. They are holding each other up. Every giant you think is fighting to win is quietly dependent on the rivals it claims to be beating. That dependence is not a flaw in the system. It is the system.
I learned this at a dinner table.
A few nights ago I was with friends, the kind of evening where work comes up sideways and nobody means anything by it. One of them works at Google. At some point, almost in passing, he mentioned that he uses Claude. Anthropic's Claude. The product built by one of Google's direct competitors.
He said it the way you'd name the coffee you drink. No edge to it. A working detail.
I let it sit. Then I went home and pulled the thread.
What I found at the dinner table was the whole industry in miniature
The thing my friend said out loud is not a quirk. Bloomberg has reported that some Google engineers prefer Anthropic's Claude Code to Google's own coding tools, including teams inside Google DeepMind, the group that builds Gemini itself. Google officially bars employees from using rival tools like Claude for security reasons. Engineers can apply for an exception if they show a work need.
The company wrote a policy to keep its own people away from a competitor's product. Its own people keep filing paperwork to use it anyway.
Now put the money on top.
In April, Google committed up to $40 billion to Anthropic. Ten billion immediately, another thirty tied to performance targets. Put it plainly: Google is funding the company whose product its engineers route around internal policy to use.
It gets stranger.
This month, Google agreed to pay SpaceX about $920 million a month, from this October through the middle of 2029, for access to roughly 110,000 Nvidia chips. Google owns more data center capacity than almost anyone alive. Google has committed more money to infrastructure this year than most nations spend on anything. And it is still renting compute from a rocket company, because it cannot build fast enough to feed its own demand.
Here is where the line bends back on itself. Those chips sit in data centers owned by xAI, the company behind Grok, which competes head to head with Google's Gemini. SpaceX absorbed xAI earlier this year. So Google is renting its competitor's computers.
And it is not alone in that bed. Anthropic rents from the same landlord, more than a billion dollars a month for the Colossus facility near Memphis that xAI originally built to train Grok.
By most accounts, those machines never performed the way xAI hoped. The architecture was a patchwork, the training stalled, and the capacity sat idle. So xAI now rents the computers it built to win the race to the two companies it built them to beat.
The compute that was supposed to deliver the victory is powering the competition instead.
This is not one strange deal. It is the shape of the whole thing
Pull any thread in this industry and the same pattern appears. Amazon is Anthropic's largest backer and one of its biggest suppliers. Nvidia sells the chips every one of these companies runs on, and holds stakes in several of them. Microsoft funds OpenAI and competes with it in the same breath. The org chart says rival. The wiring underneath says partner.
Nobody is building this alone. Not one of them. Not even the companies sitting on the largest piles of money on earth.
There is a word for this, and it is not competition
It is how rival nations behave.
Think about how two powerful, hostile countries operate when each depends on the other for things it cannot make itself. They do not go to war. They cannot afford to. Their factories need each other's materials. Their economies are stitched together at a hundred seams neither government fully controls. So they posture in public, compete for influence, and trade across the border every single day, because the alternative to trade is mutual ruin.
That is not friendship. It is arithmetic.
The AI giants have arranged themselves the same way. If Anthropic collapsed, Google would lose an investment and a tenant. If the compute rental market seized, the machines running every major model would go dark at once. They are not allies. They are a cartel that cannot afford to let each other fail.
No winner is coming, and that is good news
Here is why this matters to you, and it has nothing to do with which logo ends up on top.
You are being sold a story. The story says this is a race with a finish line. One company pulls ahead, owns the future, and your job as a leader is to pick that winner now or be left behind. Every vendor deck, every "future-proof your stack" pitch, every anxious subject line about falling behind runs on that story.
The codependency is the evidence that the story is false.
No single winner is coming. There cannot be one, because no single company can stand on its own. The race you are told to bet on is a web that none of the players can walk out of.
So stop waiting for the dust to settle. It will not settle. And while you wait, you are missing the gift this arrangement is handing you.
The empires are building your toolkit
When Google's own engineers reach for a competitor's tool, when every giant rents the same chips and trains on the same data, they are telling you something they would never say from a stage. The model is not the moat. The engine is a commodity, and they treat it like one.
Look at what their codependency forces them to do. They build their tools to connect to each other, on shared open standards, because no one can afford a walled garden anymore. They keep the engine cheap and swappable because they are all renting from the same suppliers and hedging against the same risks. Without meaning to, they are pushing the most capable technology ever built out into the open, where anyone can reach it.
That is the part nobody is selling you, because nobody profits from you understanding it.
For most of history, this kind of capability lived behind palace walls. Building a custom system that could think, write, analyze, and help run your operations required an engineering department and a budget only an enterprise could carry. That is no longer true. The same intertwining that keeps these giants alive has spilled their capability outside the walls. A single leader with the right tools can now build what used to require an empire.
This is the democratization of building itself.
So commit, and own what you build
Here is where I part from the usual advice, which tells you to stay detached and refuse to plant a flag. That is not what I do, and it is not what I would tell you to do.
I have planted a flag. My businesses run on Anthropic. I use Claude Code to build the agents that run my operations, wired into the Google Workspace my team already lives in. I committed to the tools that are best for the work in front of me, and I committed hard.
That is not dependence. Here is the difference, and it is the whole difference.
I do not rely on any one company to hand me my value pre-packaged. I use the best tool for each job, the model from one, the workspace from another, and then I build my own layer on top of all of it. The agents, the workflows, the processes, the hard-won knowledge of how my businesses run. That layer is mine. If the best engine changes tomorrow, I carry my layer to the new one and keep moving.
The first real decision in any serious AI build is not which model to use. It is where your source of truth lives.
Rent the engines. Own the architecture.
The enemy here is not Google or Anthropic or any of them. The enemy is the quiet belief that this technology belongs to the resourced few, that real capability is reserved for the giants and the enterprises with engineering armies, and that your job is to wait and buy whatever you are eventually sold. That belief is wrong, and it is expensive. It keeps you a customer of your own potential.
The capability is already in your hands
My friend at Google was not being disloyal when he reached for Claude. He was being honest about how this technology works. One tool among several. None of them sacred. The right one chosen for the job in front of him.
He was building, not waiting.
That option is open to you right now, at whatever scale you operate. Not someday, when a winner is finally crowned. Today, while the giants are too busy holding each other up to notice they have handed you the keys.
The leaders who win the next decade will not be the ones who picked the right empire. They will be the ones who realized they could build their own.
I map this web every week in The Signal, where I track what the giants' moves mean for the people who have to make real decisions while the empires play their game.
