How are leading AI companies translating their corporate growth strategies into tangible real estate decisions in 2026, and what do the signals — from aggressive SF office expansion to "disciplined execution" capex caution to full-building data center leases in the NY metro — tell us about the pace and shape of AI-driven demand?

Three Connect CRE source articles published in March–April 2026 were synthesized:

1. "Anthropic Grows San Francisco Footprint Yet Again" (Connect CRE, Paul Bubny, April 2, 2026) — based on San Francisco Business Times reporting. Package f238fa634b87a45e1f3afbb6.
2. "U.S. Companies Focus on 'Disciplined Execution' in AI Rollout" (Connect CRE, Paul Bubny, March 18, 2026) — based on BMO Business Outlook survey. Package a9da4040b401c991bc96077f.
3. "Cresa Reps Russo Development in Bergen County Data Center Lease" (Connect CRE, Paul Bubny, March 16, 2026) — primary deal report. Package 848c01f943e35e3b8b28f839.

Reviewed AI Infrastructure and Office Demand 2026 and AI Office Demand Engine 2026 to avoid duplicating macro-level SF recovery framing. The Anthropic SF campus cluster and the agglomeration thesis are covered there. This analysis adds the specific lease details for 400 Howard, the BMO corporate AI strategy signal, and the Bergen County / Meadowlands data center deal — then synthesizes the tension between aggressive AI office expansion and disciplined AI capex signals.

Anthropic has now committed to four SF office addresses across three leases in the SoMa neighborhood:

The new lease at 400 Howard: Approximately 100,000 SF in the roughly 370,000-square-foot building, which is one of four buildings in the Foundry Square complex. Owned by the State Teachers Retirement System of Ohio (Ohio STRS). 400 Howard is directly across the street from 300 Howard, which Anthropic leased in its entirety in January 2026.

A company spokesperson characterized the 400 Howard commitment as part of Anthropic's "long-term commitments" as it continues to scale in San Francisco. Notably, 400 Howard will house Anthropic employees before 300 Howard does — suggesting the company is staging its occupancy sequentially across buildings while securing space ahead of headcount.

The "yet again" context: The Connect CRE headline ("Yet Again") reflects that this is Anthropic's third or fourth demonstrable SF expansion event. The 500 and 505 Howard subleases/leases predate the 2026 acceleration; the 300 Howard full-building commitment in January 2026 was the first full-building trophy lease; the 400 Howard commitment follows within approximately three months. The pace of expansion — back-to-back six-figure SF commitments in a single quarter — is the story. These are not single-building relocations; they are parallel cluster acquisitions.

Total disclosed SF footprint: Based on the reported figures, Anthropic's SF commitments span approximately 540,000+ SF across the four addresses (240K + 100K + full building + 100K at minimum). The exact SF at 300 Howard is not disclosed in this source.

Deal economics: The article does not disclose asking rent, deal terms, or lease duration at 400 Howard. The BMO and broker context do not add pricing data.

For macro context on why this pattern matters for the SF office recovery thesis, see AI Infrastructure and Office Demand 2026.

BMO's March 2026 BMO Business Outlook survey covers U.S. companies broadly — not just AI firms — and finds business leaders are shifting from "wait-and-see" toward what BMO's head of Commercial Bank U.S., Tony Sciarrino, characterizes as "disciplined execution."

What "disciplined execution" means in this context:

• Companies are moving beyond pilots and experimentation toward practical deployment of AI and automation
• The focus is on streamlining operations, improving productivity, and redeploying resources toward higher-value opportunities
• Capital allocation discipline is the frame: the winners "won't be the companies that take the most risk — they'll be the ones that allocate capital well, protect margins, and put technology to work in ways that measurably improve productivity"
• AI-driven business investment is cited as one of several meaningful supports for the U.S. economy in 2026
• Risks remain elevated around trade policy, inflation dynamics, and geopolitics

What this survey does NOT say:

The BMO article does not name specific companies pausing or cutting AI capex. It does not report data on office or data center demand. It describes a general disposition shift among U.S. commercial banking clients — a broad survey signal, not a transaction signal. There is no data in this source about companies canceling leases, deferring data center commitments, or pulling back from AI infrastructure investment.

CRE interpretation: "Disciplined execution" as BMO defines it is not a pullback signal — it is a maturation signal. The shift from pilot to deployment means AI-related expenses are moving from experimental to budgeted. For CRE, this could represent a more durable demand base: companies committing to AI-enabled operations are making multi-year infrastructure decisions (data center contracts, office headcount tied to AI teams) rather than exploratory one-off pilots.

The primary risk flag in the BMO survey — trade policy, inflation, geopolitics — applies to general capex cycles, not specifically to AI office demand. Anthropic's concurrent expansion suggests that at least the frontier AI research tier is not in a cautious posture.

The transaction: Cresa represented Russo Development in a 249,640 SF full-building data center lease at 600 Commerce Blvd., Carlstadt, NJ, to a confidential tenant. The lease was announced March 2026.

Key deal facts:

• Address: 600 Commerce Blvd., Carlstadt, NJ
• Submarket: Meadowlands (Bergen County)
• SF: 249,640 SF — full building
• Tenant: Confidential (identity not disclosed)
• Developer/Landlord: Russo Development
• Broker: Cresa — Michael Morris, Sumner Putnam, Kensey Gawne, and Matt Deutsch
• Building history: Purpose-built in 2008 for one of the world's largest banks; continually upgraded since
• MW capacity: Not disclosed in source
• Draw factors per Cresa: Bergen County's proximity to New York City and the building's "immediate, meaningful compute capacity"

The Meadowlands submarket context: The article describes the Meadowlands as "among the most well-established data center markets in the country." Carlstadt sits within the broader Meadowlands submarket — a well-connected NJ enclave approximately 10 miles from Midtown Manhattan with strong fiber density given its history serving major financial institutions. The building's origin as a purpose-built bank data center (2008) and subsequent upgrades reflect the market's heritage as NY metro financial-sector compute overflow.

Cresa's track record with Russo: Michael Morris, head of Cresa's Data Center Capital Markets & Advisory practice, has completed "nearly 1.0 million square feet of transactions in the New Jersey area with Russo Development over the past several years" — indicating this is an ongoing advisory relationship and that Russo is an active NJ data center developer.

What is missing: MW capacity is not disclosed, making it impossible to assess power density or compute tier. The confidential tenant's identity prevents sector analysis — whether this is a hyperscaler, a financial services firm, or an AI operator affects how the deal reads as a demand signal.

These three sources, read together, reveal a two-speed pattern in AI corporate real estate:

Speed 1 — Frontier AI lab expansion (no caution): Anthropic is the clearest case. The company is not pausing. It signed a full-building SF lease in January 2026 and added another 100,000 SF three months later. The leasing pace is faster than its headcount (it is pre-leasing into future headcount) and the company characterizes these as "long-term commitments." At the frontier AI research tier — companies building foundation models and competing for elite ML researchers — the demand signal is aggressive and accelerating.

Speed 2 — Corporate AI adopters (disciplined but moving): The BMO survey describes a much larger population of U.S. companies adopting AI for operational productivity. This tier is not pulling back; it is graduating from pilots to budgeted deployment. But "disciplined execution" implies a slower, more deliberate real estate footprint change than what frontier labs are doing. This cohort's CRE implication is not campus-scale SF leasing — it is incremental IT infrastructure contracts and modest efficiency-driven office footprint adjustments.

How these coexist: The tension between Anthropic's aggressive leasing and the BMO caution signal is not a contradiction — it is segmentation. Frontier AI labs and enterprise AI adopters have fundamentally different growth profiles:

• Frontier labs are in an arms race for talent and compute; their real estate decisions reflect existential competitive pressure to secure capacity before rivals do
• Enterprise adopters are making ROI-gated decisions; they will expand when returns are demonstrated, not in anticipation of them

The Bergen County data center deal sits between these two speeds. A confidential tenant leasing a 249,640 SF full building in the Meadowlands is making a significant enterprise-scale compute commitment — but the confidential nature and the absence of MW data prevent definitive sector attribution.

The underlying CRE investment thesis: For office, the investable signal concentrates on the frontier AI tier — Anthropic, OpenAI, and their direct competitors — in the named SF/NYC/Boston agglomeration nodes. For data centers, both tiers (labs and enterprise adopters) represent demand, and the Meadowlands deal is evidence that NY metro overflow markets with legacy financial-sector infrastructure remain viable absorption venues.

• 400 Howard lease terms not disclosed: No asking rent, lease duration, or deal economics. The SF trophy office rent benchmark gap (flagged in AI Infrastructure and Office Demand 2026) remains unfilled.
• 300 Howard total SF not reported: The source confirms it is a full-building lease but does not state the building's total SF, making it impossible to compute Anthropic's precise total committed SF.
• Bergen County tenant identity unknown: Confidential tenant status prevents sector attribution. If financial services, this is a legacy demand continuation; if AI/hyperscale, it represents the same demand wave as the SF office story in data center form.
• Bergen County MW capacity not reported: Power capacity is the key metric for data center underwriting; its absence limits deal comparability to other Meadowlands transactions.
• BMO survey methodology not detailed: Sample size, industry composition, and geographic distribution of the BMO Business Outlook respondents are not described in the Connect CRE article. The survey covers BMO's commercial banking client base, which skews mid-market U.S. companies.
• No SF rent comps: This source batch does not add to the SF trophy rent picture. Qualitative expansion evidence is strong; quantitative pricing confirmation is still absent.

• Connect CRE / San Francisco Business Times: "Anthropic Grows San Francisco Footprint Yet Again," Paul Bubny, April 2, 2026. URL: https://www.connectcre.com/stories/anthropic-grows-san-francisco-footprint-yet-again/ — source note at wiki/sources/source-anthropic-400-howard-sf-office-expansion.md.
• Connect CRE / BMO Commercial Bank: "U.S. Companies Focus on 'Disciplined Execution' in AI Rollout," Paul Bubny, March 18, 2026. URL: https://www.connectcre.com/stories/u-s-companies-focus-on-disciplined-execution-in-ai-rollout/
• Connect CRE: "Cresa Reps Russo Development in Bergen County Data Center Lease," Paul Bubny, March 16, 2026. URL: https://www.connectcre.com/stories/cresa-reps-russo-development-in-bergen-county-data-center-lease/

• Analyses Hub
• AI Infrastructure and Office Demand 2026
• AI Office Demand Engine 2026
• National Digital Infrastructure Capital Deployment 2026
• Denver AI Infrastructure Cluster 2026
• Office Bifurcation
• Digital Infrastructure Real Estate
• San Francisco
• United States