TECfusions, a Florida-based data center operator running one of the largest AMD-based AI training clusters in North America, will go public through a $4 billion SPAC merger with Apex Treasury. The deal values the operator at a moment when hyperscale training infrastructure trades at premium multiples and alternative GPU architectures begin drawing serious capital.

The combination with New York-listed Apex Treasury marks the first major data center SPAC in fourteen months, following the sector's $180 billion in private deployment since early 2023. TECfusions operates facilities optimized for AMD's MI300 series accelerators, positioning the company outside NVIDIA's 92% training-cluster market share but inside the emerging cost-arbitrage tier that Meta, Stability AI, and two sovereign funds have quietly adopted for non-frontier workloads. The AMD architecture delivers 60-70% of NVIDIA H100 performance at 40% lower all-in costs per FLOP, a spread that matters at petascale.

The timing aligns with three structural shifts allocators have tracked since Q3 2024. First, inference workloads now represent 41% of hyperscale AI compute demand, up from 18% a year prior, favoring diversified silicon estates over single-vendor concentration. Second, power constraints in Northern Virginia, Phoenix, and Dallas have pushed new capacity into secondary metros where TECfusions holds 2.1 gigawatts of contracted power across five sites. Third, the AMD MI350 roadmap, due late 2025, closes the architectural gap to 85% H100-equivalent throughput, enough to pull training workloads out of NVIDIA's ecosystem if supply tightness persists into 2026.

The SPAC structure carries execution risk. Apex Treasury raised $350 million in its 2022 IPO, requiring $3.65 billion in PIPE financing or debt to close at the stated valuation. The last data center SPAC, Switch's $7.1 billion combination in March 2023, required three deadline extensions and a 19% valuation cut. TECfusions will need to demonstrate committed revenue from anchor tenants—typically three-to-five-year lease agreements with sovereign AI labs or hyperscalers—to secure institutional PIPE commitments at $10 per share.

Operators should watch for the S-4 filing within 30 days, which will disclose customer concentration, power-purchase agreements, and the AMD supply contract duration. If TECfusions has locked multi-year MI300 allocation from AMD's foundry partner TSMC, the deal prices attractively against CoreWeave's last private round at 11.2x forward revenue. If not, the valuation assumes growth that depends on a supply chain the company does not control. The shareholder vote will likely occur in Q2 2025, with trading under the new ticker shortly after.

Two of the five largest sovereign wealth funds now run AI training pilots on non-NVIDIA architecture. TECfusions goes public as that pilot budget converts to production scale.