When a law firm with $10 billion in annual revenue announces it is setting aside half a billion dollars to build its own AI platform from scratch, the conversation changes. Not because the number is shocking, though it is. Because it forces every other firm to answer a key question: what is our AI strategy, and is it actually a strategy or just a collection of vendor subscriptions? 

The announcement I am referring to landed in May 2026. The firm committed $500 million over three to four years to develop a proprietary AI platform built around its own lawyers’ combined expertise. Two hundred and fifty lawyers contributed to the design. One hundred and eighty technology professionals were assigned to build it. The message was unambiguous: AI is not a tool you buy. It is a capability you own. 

But that firm is not the only one making moves. Across the top tier of global law, at least four other distinct architectural approaches have emerged in the last eighteen months. Each reflects a different theory of where value accrues, what you should build versus buy, and how much control is worth paying for. 

Approach 1: The Blank-Check Build 

The half-billion-dollar commitment I opened with is the most extreme example. The logic is straightforward: if your firm’s shared know-how is your competitive moat, then encoding that know-how into a proprietary platform creates a durable advantage no competitor can replicate by purchasing the same software. 

This approach treats AI as a strategic asset, not a productivity tool. The firm is not buying seats on a platform its rivals also use. It is building something that captures how its lawyers think, work, and make decisions. The risk is equally clear: half a billion dollars and three to four years before the platform delivers, with technology that may look very different by the time it ships. 

Approach 2: The Infrastructure Sovereigntist 

A different top-tier firm, the second-largest in the US by revenue with over $8 billion in annual receipts, recently took a different path. It purchased its own GPU servers from Nvidia and installed them in a secure data centre accessible only to its personnel. Its engineering teams are now fine-tuning open-weight models internally, on hardware the firm controls. 

The motivation is client confidentiality. Certain matters are simply too sensitive to route through a third-party cloud provider. But the strategy is not isolationist. The firm still uses commercial AI platforms for other tasks. It has built an architecture that lets it choose between internally operated models and third-party services depending on the sensitivity and economics of each task. 

This is a hedge against dependence on a single provider and pricing changes. If a commercial AI vendor raises prices or alters terms, the firm has a fallback. It also employs over 900 technology specialists, with roughly 100 dedicated to AI. That is not a technology budget. That is a technology division. 

Approach 3: The Deep Vendor Partnership 

Another global firm, nearly 4,000 lawyers across dozens of offices, has taken the vendor relationship to its logical conclusion. It partnered with a specialised legal AI platform early, back in 2023. What started as a firmwide chatbot deployment has evolved into something more ambitious: the two organisations are now building agentic AI tools together, focused on complex practice areas like antitrust filing analysis, cybersecurity, fund formation, and loan review. 

These tools are not just for internal use. The firm is selling them to clients and to other law firms. Internal adoption reportedly exceeds 80 percent, with thousands of lawyers using the platform daily and saving two to three hours per week each. 

The architectural logic here is different. Rather than building from scratch, the firm is leveraging a vendor’s specialised legal AI infrastructure and contributing its own domain expertise to make the product better. The risk is that the platform’s core capability is the vendor’s intellectual property, not the firm’s. The opportunity is that the firm has turned a cost centre into a revenue stream. 

Approach 4: The Model-Maker Alliance 

A fourth firm, recently signed a multi-year agreement directly with a frontier AI lab. Not with a legal tech vendor but with the company that builds the underlying models. 

The partnership involves co-developing legal AI workflows, deploying the lab’s models across the firm globally, and feeding the lab with feedback on how their models perform in legal contexts. The firm also becomes the lab’s legal services provider. 

This is a subtly different bet. The deep vendor partnership bets on a legal AI specialist. The model-maker alliance bets on the foundation layer itself. The firm gains early access to frontier capabilities and a voice in how those capabilities evolve. The lab gains a sophisticated enterprise customer with deep domain expertise to help refine its products for the legal market. 

Approach 5: The Pragmatic Aggregator 

Not every firm is making nine-figure investments. Several top-tier firms are taking a more measured approach: buying broadly, building narrowly, and experimenting at scale. 

One global firm launched an internal AI sandbox that generated over 75 ideas, built a centre of excellence for AI-supported due diligence, rolled out a commercial legal AI tool firmwide, and assembled a dedicated squad of 20 AI-focused lawyers to push adoption. Another firm earmarked $25 million per year to buy and develop AI, starting with a proprietary tool built on a commercial model for its venture financing practice. A third built a single proprietary platform for its funds practice, focused narrowly on document search and drafting in one high-value area. 

The common thread: experiment, deploy what works, and build proprietary tools only where the practice area justifies the investment. This does not require a technology division or GPU servers. It requires a clear view of where your firm’s differentiated knowledge actually lives. 

What All Five Approaches Share 

Strip away the dollar figures and the hardware, and every one of these strategies is grappling with the same underlying question I wrote about in a previous piece: your law firm needs an architecture, not a product. 

The firm building a $500 million platform is making an architecture bet. The firm buying GPU servers is making an architecture bet. The firm co-developing with a vendor is making an architecture bet. They have all concluded that the real value is not in any single tool but in how the pieces connect: where the data lives, who controls the knowledge layer, how AI outputs flow back into the firm’s systems, and what happens to all of it when a vendor changes their pricing or a model gets superseded. 

The firms buying tools without answering those questions are not making a strategy. They are making a pile. 

The Question Nobody Is Asking 

The half-billion-dollar announcement and the GPU servers make headlines because the numbers are dramatic. But the most important architectural decision is not how much you spend or whether you own the hardware. It is whether your firm’s institutional knowledge is being captured in a form you control. 

I argued this point in a piece about building AI systems that retain what your firm learns over time. The firms building proprietary platforms are betting that their combined expertise, once encoded, becomes an asset that grows more valuable with every matter handled. Every partner’s strategic insight is captured and reused. When a senior lawyer retires, their workflow adaptations do not leave with them. 

But you do not need $500 million to achieve that. You need a knowledge layer that lives in files you own, connected to AI tools that can read and update it, with a human approval gate that keeps lawyers in control. The firms buying GPU servers and the firms subscribing to commercial platforms face the same test: is the knowledge compounding inside the firm, or is it evaporating into someone else’s product? 

What This Means for the Rest of Us 

Most law firms cannot write a $500 million cheque. Most cannot justify a dedicated data centre or 100 AI engineers on staff. But the architectural principles are the same at any scale. 

Start with your data. Understand where it lives and who controls it. Identify your layers: your systems of record, your automation engine, your AI capability. Keep your proprietary knowledge in a form you own, separate from whichever tool you are renting this quarter. Document one workflow, end to end, and build from there. 

The firms making headlines are not doing something fundamentally different. They are doing the same things, with more money and more engineers. The question for everyone else is not whether you can match their budget. It is whether you can match their intent.