Big Law Firms Likely Next to Go Anthropic Deep

Freshfields announced its multi-year Anthropic partnership on April 23, 2026: 5,700 lawyers, 33 offices, +500% adoption in six weeks. The procurement template is now public. The next question every legal-tech analyst gets asked: who's next? This is a structural pattern question, not a rumor question. No public reporting confirms specific firms in talks with Anthropic. What's possible is to identify which firms fit the structural template (revenue tier, global footprint, AI deployment posture) and where the procurement logic points. Here's the data-backed read on the firms that fit, and the ones that don't.

The three structural criteria that gate co-build deals

Co-build with a foundation model provider isn't a deal every firm can negotiate. Three structural criteria narrow the field, and all three need to be present:

- Revenue tier above ~$2B annually. Co-build commits the firm to a multi-year contract value that likely lands in the $15M-30M+/year range based on the 5,700-seat Freshfields scope. Firms below $2B revenue can't underwrite that without restructuring partner profit-share. Per American Lawyer's 2025 Global 200, about 30 firms globally cleared the $2B threshold last fiscal year. - Global office footprint with sophisticated AI risk-and-ethics function. Co-build deployment across 33 offices like Freshfields requires a centralized AI governance function that can negotiate data residency, jurisdictional regulatory posture, and matter-conflict checks across geographies. Firms with offices concentrated in 2-5 cities can't operate the deployment surface a co-build deal requires. - Existing public AI deployment posture. A firm that's never publicly committed to AI deployment can't absorb the political risk of a Magic Circle-scale rollout. The firm's risk-and-ethics committee, partner board, and clients all need to be on record as accepting AI in workflow before a co-build deal makes sense politically. Firms that have been quiet about AI procurement face an internal change-management problem co-build amplifies.

The second-order point: these criteria aren't independent. They correlate. The firms with $2B+ revenue and global footprint generally also have public AI deployment posture, because their clients (multinational corporates, sovereign wealth funds, major financial institutions) demand transparent AI governance. The qualifying set is therefore tighter than the union of any single criterion.

Allen Overy Shearman: structural fit, vendor incumbency complication

Allen Overy Shearman fits all three criteria. Revenue cleared $2.7B in fiscal 2024 per public reporting. Global footprint of ~50 offices. Public AI deployment posture is the strongest in BigLaw. A&O announced its Harvey AI partnership in February 2023, the first BigLaw firmwide rollout, and has iterated on Harvey through four generations of the platform.

The complication: A&O's relationship with Harvey is structural. Three years of contracts, deployed workflows, integrated training programs, and partner-level political capital invested in the Harvey bet. A pivot to a co-build with Anthropic would either replace Harvey (politically expensive) or run alongside (operationally expensive). Neither is impossible, but both have switching costs.

The second-order angle: Harvey itself runs on Anthropic models for many tasks. A&O is therefore already an indirect Anthropic customer through Harvey. A co-build deal for A&O would likely look different from Freshfields': possibly a Harvey-plus-direct-Anthropic dual-track structure, or a Harvey-on-Anthropic embedded co-development arrangement.

Probability read: A&O is more likely to deepen the Harvey relationship with explicit Anthropic-coupled terms than to announce a Freshfields-style direct co-build in 2026. The procurement contracts have momentum; the political capital is sunk into the Harvey track.

Latham Watkins: strongest open-vendor fit

Latham Watkins fits all three criteria with no incumbency complication. Revenue cleared $6.5B+ in fiscal 2024 per public reporting (largest law firm globally by revenue). Global footprint of 30+ offices. AI deployment posture is public but not vendor-locked. Latham has used multiple legal AI tools (publicly known: CoCounsel, Spellbook in pilots, internal API tooling) without committing to a flagship vendor relationship.

The structural argument for Latham going Anthropic-deep:

- The firm's transactional and capital markets practices are exactly the workload where multi-session memory and task budgets (Opus 4.7 features) compound hardest. M&A diligence, capital markets disclosure work, complex transactional drafting all benefit from foundation model access more than vertical vendor workflows. - Latham's internal engineering capacity is the deepest in BigLaw. Multiple public engineering hires from tech and legal-tech in 2024-2025 signal a build-side procurement preference. - No vendor incumbency means no political cost to a foundation model bet. Latham can announce a co-build without unwinding a Harvey-style relationship.

Probability read: Latham is the most structurally probable next named co-build deal. No public reporting confirms talks, but the fit is clean. Watch the next 6-12 months.

DLA Piper: scale fit, posture question

DLA Piper fits the revenue and footprint criteria. Revenue around $3.5B fiscal 2024. Global footprint of 90+ offices across 40+ countries, actually larger than Freshfields. AI deployment posture is the open question.

DLA has been publicly committed to legal AI but somewhat vendor-quiet. Per public coverage, DLA has invested in internal AI tooling and partnered with several legal AI vendors (specific deployments not always public). The firm hasn't named a flagship co-build relationship the way Freshfields has, but the infrastructure is in place to absorb one.

The second-order complication for DLA: the firm's verein structure (multiple member firms operating under a unified brand) makes firm-wide AI deployment more complex than Freshfields' more centralized partnership structure. Co-build requires uniform governance across member firms, which DLA's structure historically makes harder. Not impossible (DLA has solved similar coordination problems for global compliance) but the political and procurement work is heavier.

Probability read: DLA is structurally fit but operationally complex for co-build. More likely to pursue multi-vendor enterprise procurement than single-vendor co-build in the 2026-2027 window.

Kirkland Ellis: fit, but disputes-heavy practice mix shapes the read

Kirkland Ellis fits the revenue and footprint criteria. Revenue cleared $7B+ in fiscal 2024 (second-largest globally by revenue). Global footprint of 20+ offices. AI deployment posture is growing publicly. Kirkland has been more open about AI deployment in 2025-2026 than historically, partly responding to client demand for transparent governance.

The structural argument for Kirkland: the firm's massive disputes and litigation practice is exactly the workload where Anthropic's calibration improvements compound. Litigation strategy work demands models that don't proceed confidently with bad plans; calibration is a malpractice-relevant feature, not an academic benchmark.

The complication for Kirkland: disputes work also benefits structurally from CoCounsel-style Westlaw-grounded research, which is now Anthropic-powered post-rebuild. Kirkland may end up running a Thomson Reuters CoCounsel relationship plus direct Anthropic API access, rather than a Freshfields-style co-build. The economic case for direct co-build is weaker when the litigation-research vendor is already on the same model layer.

Probability read: Kirkland is structurally fit but more likely to pursue a CoCounsel + direct Anthropic dual-track than a single-vendor co-build. The next 12 months will reveal whether the dual-track structure crystallizes publicly.

The five firms that don't fit, and why that matters

Five categories of firms structurally don't fit the Freshfields template, even at large scale:

- AmLaw 50 firms with under $2B revenue. Skadden ($3.5B), Sidley Austin ($2.7B), Davis Polk ($2.3B) all fit revenue. Cravath ($1.0B revenue but high revenue-per-lawyer) doesn't clear the threshold; their procurement model is structurally different. - US-only firms without global footprint. Wachtell, Cravath, and Sullivan Cromwell have the prestige and revenue per lawyer but lack the geographic scope co-build deployment requires. Co-build economics break down without 20+ offices to amortize the engineering investment across. - Firms with deep Microsoft Copilot or vertical-vendor commitments. Some AmLaw 100 firms have publicly bet on Microsoft Copilot for Microsoft 365 plus a vertical legal vendor. Co-build with Anthropic would replace or run alongside both, which most procurement boards would resist absent a forcing function. - Firms with internal-build preference. Some BigLaw firms have publicly committed to building rather than buying: internal AI engineering teams writing custom tooling on raw API access. Co-build adds coordination overhead these firms structurally avoid. - Firms in regulatory ambiguity zones. Firms with major practice exposure in jurisdictions where AI legal use is regulatorily unsettled (some EU member states, certain Asian jurisdictions) face deployment friction co-build amplifies.

The second-order point: most BigLaw firms (including most AmLaw 100 firms) won't be co-build candidates. The procurement conversation for those firms is enterprise vendor selection (Harvey, Spellbook, CoCounsel) plus internal tooling on top. The co-build vs buy comparison walks through which procurement track fits which firm size.

The Bottom Line: The structural read: Latham Watkins is the most probable next named co-build deal based on revenue, footprint, AI deployment posture, and lack of vendor incumbency. Allen Overy Shearman fits structurally but has Harvey incumbency that complicates a clean co-build pivot. DLA Piper fits scale but the verein structure adds operational complexity. Kirkland Ellis is more likely to run a CoCounsel-plus-direct-Anthropic dual-track than a single-vendor co-build. The next 6-12 months will produce 1-3 more named deals; this list is the structural-fit short list, not a rumor list. Watch for procurement signals (engineering hires, RFP language shifts, public AI committee statements) more than vendor announcements.