It’s undeniable that AI is going to make a significant impact on our economy. But there’s always a chance AI could be another bubble. I lived through the internet expansion of the ’90s, and AI is on pace to be much bigger, more disruptive, and more impactful. AI has the potential to impact every human being and company in the world. In order to achieve this and ensure that Western values drive AI adoption, we have three big things to fix before AI can become the “new internet” that changes every interaction we have. 

Accessing critical AI infrastructure is hindered by the following fundamental problems: crushing costs, extreme concentration, and insufficient capacity. The U.S. AI Action Plan framed it as, “Currently, a company seeking to use largescale compute must often sign long-term contracts with hyperscalers—far beyond the budgetary reach of most academics and many startups.” Further, data transfer fees, the costs a company faces when moving its own data out of a cloud system, strangle competition by creating artificial barriers that have nothing to do with technical merits.

Part of what made the Internet open to all was the common TCP/IP standard, that lets any computer talk to any other computer. This ability to connect outside of the (then) walled garden of AOL opened up new business uses while creativity flourished and capabilities improved rapidly. 

If we want AI to grow and succeed like the Internet, we need  the global standards and open exchange that can make accessing it as easy as sending a packet from one computer to the next. Beijing has already published several action plans that help provide direct subsidies to AI companies and make vast datasets available to national champions through public-private partnerships, providing common approaches for Chinese companies to follow. The Chinese government has announced a state-run exchange to catalog and market compute. For too long, America’s fragmented approach and lack of national strategy has risked ceding our current AI leadership permanently. By setting the standards that will help us stay ahead in AI, we can create the type of market-driven solutions that will quickly bring the power of American AI to more of the world’s population.

The ’40s, the ’60s, the ’90s and today

Before the internet, we can look further back to the age of the mainframe and the transition to the PC for insight. In the 1940s, computing power was so expensive that only governments could afford it, housing room-sized machines in hangars and partnering with universities for access. Today, my pocket-sized phone is more powerful than those early supercomputers. What changed? IBM led the commoditization of computer hardware while building a profitable business. Microsoft and Linux created operating system standards. These changes rolled out over decades, but today’s technology moves so much faster, and calls for a faster response to stay competitive. 

Today, AI sits where computing did in the 1960s — powerful but accessible only to those with massive resources. We need our own IBM PC revolution for artificial intelligence, with open standards and exchange, so we can move as quickly as possible to win the global war for AI.

This shift won’t happen just through government mandate, nor should it. It requires market forces guided by open standards that make AI resources truly accessible and interoperable. When companies can easily compare compute options and move workloads between providers, competition will drive market-efficient prices and improve service. When developers can access standardized AI infrastructure without months of advance planning, innovation accelerates.

This approach serves America’s strategic interests. When our allies and our adversaries can easily adopt US-developed AI standards and infrastructure, we strengthen the entire democratic technology ecosystem. President Trump’s move to allow China to purchase NVIDIA H20 processors helps us in the long-term, because it helps encourage Chinese developers to build on American technology. When adversaries rely on an American technology stack, we gain influence far beyond what export controls can achieve while promoting our democratic values.

The government has a role here to ensure markets function competitively. Strategic compute reserves could provide emergency capacity during crises, similar to our strategic oil reserve. Procurement policies can prioritize small and medium enterprises over concentrated providers, especially government procurement. Standards development can prevent the kind of fragmentation that has historically hampered American technology leadership.

This is the time for swift action.

President Trump has promised to keep America “the hottest country” in technology and his AI Action Plan positions us for success. China has a plan that they’ve been working on for years. Delivering on the promise of AI for the Western world requires not only a government commitment to leadership but also creating the market conditions where American innovation can flourish – open, competitive, and accessible to everyone willing to build the future.

The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.