The Compute 100 — Thursday June 18, 2026

The Federal Energy Regulatory Commission issued guidance June 18 fast-tracking grid interconnection for AI data centers — the first federal mechanism to directly address the interconnection queue that has been the binding constraint on hyperscale buildout. Facilities drawing more than 20 megawatts can access interconnection studies compressed to as little as 60 days by accepting curtailment agreements, committing to reduce load during peak grid stress, and funding 100% of their own network upgrade costs rather than spreading them to ratepayers. FERC also directed PJM, the largest U.S. grid operator, to develop three new transmission service options enabling data centers to co-locate with on-site generation and bypass congested transmission entirely.

PJM's expedited interconnection track, effective July 31 through 2027, handles up to 10 projects annually at 250 MW or above and targets grid agreements within 10 months of submission; the Southwest Power Pool will run a parallel 90-day fast-track for tech hubs paired with generation. A Camus Energy analysis found flexible 500 MW data centers on the new pathway could reach full operation in roughly two years — three to five years faster than traditional interconnection timelines.

PsiQuantum broke ground Wednesday at Moreton Bay Central in Queensland — its second active construction site after Chicago — on a facility anchored by a Linde Engineering cryoplant scheduled for H2 2027 delivery, the gating item for first operational demonstrations. The A$940 million joint investment from Australia's federal and Queensland state governments funds a system built on the company's silicon photonics approach, manufacturing qubits on GlobalFoundries' standard process nodes. The 2025 Nature paper documents 99.98% qubit state preparation fidelity and 99.22% two-qubit fusion fidelity on commercial fab infrastructure.

Architect Labs raised a $24 million seed round led by Kindred Ventures, with TQ Ventures, Race Capital, and Together Fund participating and Google DeepMind Chief Scientist Jeff Dean among the angels. The Palo Alto startup — founded by Ebrahim Hussain and Aaditya Subedi — is building an AI-driven platform to compress custom ASIC design timelines from the current two-year, hundreds-of-millions-of-dollars baseline to weeks, targeting the custom silicon design services market currently dominated by Broadcom and Marvell.

Bernstein published a structural call arguing agentic AI is shifting server CPU-to-GPU deployment ratios from roughly 1:4 toward 1:1, expanding the server CPU total addressable market to $223 billion by 2030 — with ARM's power-efficient architecture named the primary beneficiary. Mizuho lifted its target to $500 on the same day; ARM advanced 5.7% in a session the Nasdaq fell 1.3%, with the SOXX outperforming on hardware-versus-software rotation.

Four banks raised price targets on June 17 — Citi's Atif Malik moved to $710 from $550, the session's highest call — citing NAND equipment demand and a structural DRAM supply-demand gap driven by AI compute requirements through 2028. Applied Materials also confirmed a $500 million expansion of its Singapore manufacturing and R&D footprint, more than doubling advanced cleanroom capacity.

SpaceX is preparing a bond sale of at least $20 billion — its first investment-grade dollar offering — to refinance the bridge loan taken on in February when the company acquired xAI, with proceeds earmarked for AI infrastructure: data centers, compute, and power. Bank of America, Citigroup, JPMorgan, Goldman Sachs, and Morgan Stanley are running the deal; investor calls are set to begin as early as next week. The offering comes six days after SpaceX's Nasdaq IPO under ticker SPCX, which closed its debut session at $161 before rallying to a $2.44 trillion implied market cap at today's close.

SandboxAQ — The U.S. Department of Commerce signed a definitive $500 million CHIPS Act R&D agreement with SandboxAQ on June 17, tasking the NVIDIA-backed, Alphabet-spun startup with developing commercial alternatives across four semiconductor supply chain vulnerability categories: PFAS-free process chemicals, advanced catalysts, rare earth-free permanent magnets, and battery chemistries for fab infrastructure. The contract structure is unprecedented in the CHIPS R&D program: Commerce receives a minority, non-voting equity stake plus royalty rights on commercially licensed formulas — the first time the program has taken an ownership position in a recipient. Chinese rare earth export control exposure and the EU's PFAS phase-out timeline both threaten the chemical inputs for advanced semiconductor manufacturing independent of fab location; SandboxAQ's AI-driven materials simulation platform is the mechanism the government is funding to address both simultaneously.

Synopsys launched its Multiphysics Fusion portfolio on June 17, integrating Ansys physics simulation — acquired in 2024 — into EDA design flows for the first time across four production-ready solutions: timing signoff (up to 3x faster, SPICE-accurate multiphysics), design closure (up to 10x faster engineering change orders), 3DIC/chiplet multi-die assembly, and co-packaged optics. At advanced nodes and in multi-die architectures, thermal, electromagnetic, and mechanical effects have become design-closure blockers that conventional EDA — which stops at electrical analysis — cannot resolve; this is the industry's first unified platform to address all four in a single environment. Validated by Cisco, MediaTek, and Samsung Foundry, it directly targets AI accelerator teams where co-packaged optics integration is now a first-order design constraint and the current multi-tool signoff process creates schedule risk.