"We're a Microsoft Shop." — Don't Let Big Tech's Death Grip Stall the AI Revolution

I’ve been thinking about a couple experiences I’ve had in the last few years, demonstrating some of the most innovative AI tools I’ve seen in my career — tools that genuinely made the people in the room lean forward, eyes wide, asking how is this even possible.

And then two separate organizations — one a government agency, one a non-profit — shrugged.

Not because they weren’t impressed. They were. But one said, “We’re a Microsoft shop,” and the other said, “We’re a Google shop.” And that was the end of the conversation.

Maddening.

It reminded me of the old IBM commercial. No one ever got fired for buying IBM. Decades later, we laughed at how much innovation that one unspoken rule suffocated. How many better, cheaper, faster tools were left on the shelf because they didn’t have the IBM logo on the box.

We are watching that same dynamic play out in real time — at a moment when the stakes couldn’t be higher.

AI is not just moving fast. AI is moving at a speed we don’t have a good word for. The gap between what startups are building right now and what Microsoft Copilot or Google’s bundled tools can do is not a gap — it’s a canyon. And yet institutional inertia has become a more powerful force than innovation itself.

Here’s what troubles me most: the organizations most dependent on best-in-class AI tools are the ones least likely to use them.

Government agencies manage public health, infrastructure, education, emergency response. Non-profits serve vulnerable populations on razor-thin budgets. These are exactly the organizations that need every efficiency gain, every insight, every capability edge they can get. And they are systematically locked out of accessing it — not by law, not by budget, but by culture.

“We’re a Microsoft shop” is not a technology decision. It’s a risk-avoidance decision. It’s a career-protection decision. Nobody gets fired for buying Microsoft. But plenty of constituents go underserved because the tools that could have helped never got a fair hearing.

The competitive AI landscape has never been more alive with genuine innovation outside of Big Tech. Platforms like Together AI are letting organizations deploy and host open-source large language models without being locked into Google, AWS, or Microsoft infrastructure — giving enterprises control over their own data and models for the first time. Hugging Face, with its massive open-source model repository, has become what one analyst called “GitHub for AI” — a thriving ecosystem where thousands of models, datasets, and tools live outside the walled gardens of trillion-dollar companies. (Sources: AIMMediaHouse, Renewator)

There is even progress at the federal level. The U.S. General Services Administration’s OneGov initiative recently announced support for Meta’s open-source Llama models across federal agencies — specifically because open-source AI allows government teams to build, deploy, and scale applications while maintaining full control over sensitive data. That’s a meaningful crack in the wall. But it shouldn’t take a federal initiative to give agencies permission to try something new. (Source: Meta)

Meanwhile, antitrust experts and regulators are beginning to name what many in the industry have long feared: the Big Five tech companies have accumulated control over the key inputs into modern AI — specialized chips, compute, and talent — in ways that make it extraordinarily difficult for new entrants to compete. Already, trillion-dollar companies are more powerful than many nation-states. They control significant portions of our economy and the flow of information. (Source: ProMarket)

And when government employees reflexively default to the biggest players, they don’t just miss out on better tools — they actively deepen that concentration of power.

I want to be clear about what I’m not saying. Microsoft and Google have built real, useful AI capabilities. I’m not arguing they’re incompetent or malicious. I’m arguing that innovation never comes exclusively from incumbents — and that defaulting to them by policy rather than by merit is a different kind of problem.

The most transformative AI tools being built right now are coming from startups. Vertically focused companies building tools specific to healthcare, legal, education, and civic government. Agentic AI systems that don’t just answer questions but take action. Open-source models that can be deployed on sovereign infrastructure, customized, and controlled — no vendor lock-in, no data harvesting by a platform with competing commercial interests.

These tools exist. They work. And in many cases, they outperform what the big platforms offer.

So how do we fix this?

A few things I believe would make a real difference:

Mandate pilot programs. Government agencies and large non-profits should be required to run structured pilots of non-incumbent AI tools before renewing major technology contracts. Not an endless procurement process — a real, time-boxed evaluation. If the startup tool doesn’t perform, fine. But give it a fair test.

Protect the evaluators. The “no one gets fired for buying IBM” problem is fundamentally about career risk. We need explicit institutional protections for the people who recommend innovative tools — so that trying something new isn’t a career-limiting move, even if it doesn’t pan out.

Fund open-source AI infrastructure. The federal government funds roads and broadband because they’re shared infrastructure that enables everything else. Open-source AI deserves the same treatment. Tools built on Llama, Mistral, or other open models can be customized for public-sector use cases, run on government-controlled infrastructure, and made available without enriching a single private monopoly.

Change the procurement culture. Procurement rules designed for 10-year software contracts are simply incompatible with an AI landscape that changes every six months. We need new frameworks — faster, more adaptive, more willing to accept the possibility of iteration and change.

The AI revolution is not going to wait for institutional comfort to catch up.

The startups building the most remarkable tools I’ve ever seen are not standing still. They are iterating at a pace the big platforms structurally cannot match. Every month that government agencies and non-profits spend defaulting to their incumbent vendors is a month of capability left on the table — and a month of compounding advantage handed to companies that are already more powerful than they should be.

We cannot afford to let trillion-dollar incumbency determine the future of public-sector technology.

I’d love to hear your thoughts. How do we break this open? Share your ideas in the comments.

If your organization is ready to explore what a fleet of AI agents could do for your team, we’d love to start the conversation. Visit soar.com to find the right subsidiary for your needs.

Paul Allen writes about the intersection of AI, human performance, and the future of competitive advantage.