Recently, Konstantine Buhler, a partner at Sequoia Capital, conducted an in-depth analysis of the four core pillars of AI as a national strategic asset in an interview with Bloomberg: computing power, power, data, and algorithms, and emphasized that “AI memory” is rapidly becoming a new key capability outside the pillars. With the evolution of generative AI, agents are no longer just tools for executing commands, but “collaborators” with continuous self-awareness and collaborative capabilities.
He also publicly introduced for the first time the Model Context Protocol (MCP), a protocol that Sequoia is highly concerned about, a basic framework designed to enable language interoperability and task collaboration between AI and AI, AI and software. Through MCP, multiple AI agents with different expertise can collaborate like an interdisciplinary team, opening up processes such as research, decision-making, and generation, truly opening the era of “AI ecosystem”.
In the face of the rapid rise of China’s AI research forces, Buhler said that although China already has a scale advantage in talent density, the United States still maintains a leading position in algorithms and application implementation with an open and collaborative technical culture and a top engineer ecosystem.
What does a product manager need to do?
In the process of a product from scratch, it is not easy to do a good job in the role of product manager, in addition to the well-known writing requirements, writing requirements, writing requirements, there are many things to do. The product manager is not what you think, but will only ask you for trouble, make a request:
View details >
Compere:
What do you think of your reaction to infrastructure agreements in the last 24 hours, with U.S. technology entering the Gulf countries, countries setting up data centers and development projects there?
Konstantine Buhler：
These new agreements prove the fact that AI is not just a matter of business success, it is a strategic need at the national level. This speaks volumes about the importance of this industry. I think back to a decade ago, we were still worried about whether the United States could stay ahead of the curve in AI.
The strength of AI actually has four pillars: computing power, power, data, and algorithms. Looking back at the last wave of AI development, the reason why the United States was far ahead is the fourth pillar – algorithms.
So, as we develop new technological allies globally – whether it’s computing power, power, or data – we must ensure that we remain at the forefront of algorithmic advancements. We have the best researchers and top engineers in the world.
Compere:
We have just received breaking news from the White House: President Trump has secured an economic commitment totaling $1.2 trillion in Qatar. According to the White House, the commitment includes specific economic agreements worth $243.5 billion.
This statement was announced by the President himself in Doha. We just connected with Bloomberg reporter Annmarie Horden, who is currently in Doha, Qatar. She mentioned that the agreements are still in the making, and which industries and companies are involved have not yet been announced. As soon as there are more details, we will bring you updates as soon as possible.
Now, let’s return to the topic of conversation — U.S. leadership in infrastructure. You just mentioned four key categories, and one of the topics that has been popping up frequently on my desktop recently is “AI memory” in the context of generative AI.
Many people say that we have to invest more in this area and find solutions. But personally, I lack a professional background and no academic understanding in this area. Why is this issue being discussed so often in your industry?
Konstantine Buhler：
In addition to computing power, data, power, and algorithms, there is a very critical but often overlooked component, which is “memory”. Because when you interact with a smart agent, you certainly want it to remember you – but more importantly, it must remember itself as well.
For example, when a doctor communicates with a patient, an intelligent agent should be able to help the doctor review past communication records. It’s not just what the patient says, not just their vital signs or medical record data, but the ability to learn and evolve over time. That’s exactly what companies like Open Evidence are working on.
So, “memory” is becoming a new key technology field, especially in the context of “genetic AI” (genetic AI) that you mentioned. If we want to build an AI that not only responds to humans, but also performs self-reflection, then it must have memory capabilities. It’s not just about remembering a few conversations, it’s about building a long-term memory system that is continuous, self-aware.
Because of this, more and more companies are starting to invest heavily in “memory capabilities,” especially companies like Open Evidence, which focus on long-term intelligent interactions. This is also a key step in the transformation of AI from a tool to a “collaborator”.
Taking medical care as an example, with memory ability, doctors can accurately recall the details of every interaction with patients, and even the evolution of communication methods – and these will become particularly important in AI-assisted medicine in the future.
The question arises: where should the infrastructure for all these activities be deployed? Today’s Bloomberg Big Take puts forward the idea that the sudden rise of a company shows that China’s related industries are developing rapidly and are hardly affected by US policy or political uncertainty.
Compere:
Chance Amont told me in March that more than 50% of AI researchers are now in China. Combined with the situations you just mentioned, do you agree with this view? That said, China’s AI industry is really growing rapidly, and they’re working on issues similar to yours?
Konstantine Buhler：
They do have very good researchers, there’s no doubt about that. But it’s also true that we also have some of the most creative and brightest research talents on the planet.
For example, just recently, we hosted the Sequoia Annual Summit, where we invited 150 of the industry’s top experts. From Jensen Huang, Sam Altman to some very promising young rising stars, all present.
The most talked about technology topic at this conference was a concept called “Tool Use”, which is to enable AI to collaborate with each other. We are teaching computers how to use computers. Fortunately, we have made significant progress in this area over the past few months.
At present, we have a new protocol called “MCP (Model Context Protocol)”. You can think of every AI Agent as an “expert”, and similarly, each software can be seen as a manifestation of some kind of professional ability. For example, your CRM system may be very good at recording and understanding historical interactions with your customers. But the problem is that these “experts” do not necessarily speak the same language among themselves. MCP is like a universal “translation protocol” that allows all these agents and software to communicate and share information with each other. For us in the United States, this capability is key to staying ahead of the curve. We must deepen collaboration and promote open innovation.
Let me share a real-world use case for this protocol: we have a portfolio company called Rocks. It helps top salespeople complete very in-depth and accurate background research before meeting with prospects.
Now, they can not only complete the research, but also integrate all the information through the MCP protocol and automatically generate a tailored pitch deck. You can even connect to Cognition or Cloud Code to automatically generate a full product demo.
That’s how we stay ahead of the curve at the AI collaboration level – relying on deep collaboration between researchers and engineers. We must continue to invest in such cooperative mechanisms. We know that China is accelerating in many fields, but in the United States, we have a unique “open innovation culture” that allows talents from different companies and backgrounds to gather, experiment and share results freely.
Compere:
You mentioned Tool Use and MCP – it sounds like these are ways to get AI systems to collaborate with each other. So what do you think will be the next breakthrough in the industry? Are you already seeing some early signals?
Konstantine Buhler：
Yes, what we will see next is the rise of the AI Agent ecosystem. Imagine multiple AI Agents working together to complete tasks like an interdisciplinary team of experts. For example, one agent might be good at financial analysis, another good at writing code, and another good at marketing strategy – they can work together seamlessly through MCP.
The greatest value of this model lies in the ability to scale: you are no longer limited to the capabilities of one model, but can call on the collective intelligence of the entire ecosystem. This also means that AI will become more and more like a partner rather than a tool. It will be able to build knowledge, ask questions, and correct mistakes together with humans. This is the next stage in the development of AI and will be a key factor in determining which country and which company is leading.
