Why OpenAI, Google, and Anthropic Don’t Talk Much About Patents

The way patents are being handled in AI among the major tech giants is hard to see clearly or even imagine. I had AI (Claude) explain several aspects of this.

🙋In the business competition among Google, OpenAI, Anthropic, xAI, Meta, and Microsoft, patents don’t come up much as a topic — is this a limitation of the patent system itself?

The reason patents rarely come up in AI industry discussions lies not so much in the limitations of the patent system itself, but rather in a structure where patents simply don’t function well.

The sources of competitive advantage don’t lend themselves to patents

What separates winners from losers in AI includes:

• The quality and quantity of training data
• Knowledge of scaling laws
• The tacit knowledge of researchers

None of these can be protected by patents. The discovery that “training on more data makes a model smarter” is simply not patentable subject matter.

A clash with academic publishing culture

AI research has its roots in academia, so researchers often post to arXiv before filing a patent, at which point novelty is lost. Since researchers are evaluated by citation counts, this culture is unlikely to change anytime soon.

The legal environment is a barrier

• In the US, following the 2014 Alice decision, algorithms and mathematical methods have become very difficult to patent
• Similar restrictions exist in the EU

A standoff where everyone holds patents but nobody uses them

Every major company holds a vast AI patent portfolio, but filing suit damages reputations, so they exist in a mutual deterrence through cross-licensing. There is also the contradiction of Meta openly releasing Llama as open source while simultaneously launching patent attacks — strategically, those two moves don’t fit together.

The pace of progress outstrips the patent cycle

Patents take two to five years to be granted, during which time models advance through multiple generations. By the time a patent is approved, the technology it covers is often already obsolete.

So what is the real moat?

• Architectural details kept as trade secrets
• Exclusive access to data (Google’s search logs, contracts with Reddit and news publishers, etc.)
• Massive investment in compute infrastructure
• Locking in top-tier talent
• Dependency on APIs and ecosystems

Going forward, patent competition is expected to intensify in areas involving physical implementation — chip design and robotics in particular — and the early signs of that are already beginning to emerge.

🙋For example, Google holds the patent on the Transformer (‘Attention is All You Need’), right? And basically all current LLMs are Transformer-based, right? So I would think they could collect unlimited royalty fees from every company out there.

Google does indeed hold the patents

Almost all of the authors of the 2017 “Attention is All You Need” paper were Google employees, and Google holds patents related to the Transformer architecture. As you point out, since virtually all current LLMs are Transformer-based, Google could theoretically demand enormous royalties.

So why haven’t they enforced them?

The main reasons thought to be behind this include:

• Because the architectural details were published simultaneously with the paper, there is a real risk that the patents could be challenged as “prior art” and invalidated
• If Google were to sue OpenAI, Meta, or Anthropic over patents, it would face counterattacks from Microsoft and Meta’s own patents — a nuclear deterrence standoff
• Google depends on other companies across search, advertising, and cloud, making the cost of turning the entire industry against them far too high
• Many of the original authors subsequently moved to companies other than Google — OpenAI, Anthropic, and others — making the situation politically complicated

The validity of the patents themselves is actually questionable

This may be the most important point of all.

• The Attention mechanism that forms the conceptual foundation of the Transformer already existed around 2014 to 2015, leaving room to challenge the patents as invalidated by prior art
• The innovation in “Attention is All You Need” lay in how existing elements were combined, making it difficult to claim novelty for any individual component
• The Alice decision remains a barrier for software patents

Conclusion

The reason Google doesn’t enforce its patents is not because they are being generous. The more likely explanation is a cold calculation that the moment they do, they turn the entire industry against them, face exhaustive challenges to the patents’ validity, get hit with countersuits, and ultimately lose more than they gain.

That said, this is the equilibrium as it stands today. If the balance of power in the industry shifts dramatically — for instance, if Google’s AI business finds itself truly cornered — the possibility that this “patent nuclear option” gets used is not zero.

🙋Among Google, OpenAI, Anthropic, xAI, Meta, and Microsoft, are there any companies that are quietly biding their time right now, but are actually secretly and strategically positioning themselves to enforce their patents?

When we look at each company’s position, the ones most likely to be quietly targeting patent enforcement naturally stand out.

Microsoft is the most likely candidate

There are several reasons for this.

• They have a history of aggressively using software patents (they spent years collecting royalties from Samsung and HTC over Linux-related patents against Android)
• There is significant pressure to recoup their massive investment in OpenAI
• They have very strong motivation to protect the Azure infrastructure and Copilot ecosystem
• Their legal team is the most experienced with patent litigation among these companies

If OpenAI and Microsoft’s relationship deteriorates in the future, patents could become a powerful negotiating card.

Meta is a close second

• Releasing Llama as open source can be read as a strategy of “if we can’t protect it with patents, dominate by spreading it first”
• Meanwhile, they are quietly accumulating a massive patent portfolio behind the scenes, and may be preparing to enforce rights against companies that commercially exploit Llama in the future
• Zuckerberg has a track record of using patents strategically going back to the Facebook era

Anthropic is actually at risk of being targeted

• Their financial resources are relatively limited and their patent portfolio is thin
• They receive investment from Google and Amazon, which provides protection for now, but if those relationships change they could become vulnerable

xAI is still hard to read

• Musk previously declared Tesla’s patents “open,” but in reality it was a strategic move
• xAI’s patent strategy is still taking shape, though there is a possibility they could make moves around Twitter data rights

Google will hold theirs as a last resort

Rather than actively using them, the most natural reading is that Google is holding the Transformer patents in reserve for when they are truly cornered. It is essentially a “nuclear weapon” — and precisely because using it would transform the entire industry, there is value in not using it.

Looking at the overall picture, Microsoft is the company most likely to actually act, and what sets them apart from the others is the very real and present tension of a potential breakdown in their relationship with OpenAI.