Data center networking revenue in Q1 FY27 reached $14.8B, up 199% year-over-year and 35% sequentially — outpacing data center compute, which grew 77% YoY to $60.4B. Total revenue of $81.6B beat consensus by $2.4B; Q2 is guided at $91.0B ±2%. Nvidia added $80B to its buyback authorization and raised the quarterly dividend from $0.01 to $0.25, a 25-fold increase that reflects management's confidence in the durability of cash generation at this scale.
Networking — InfiniBand, Spectrum-X Ethernet, NVLink — is now running at roughly a $59B annualized revenue pace and growing nearly three times faster than the GPU business it supports. The binding constraint in large-scale AI clusters is increasingly how GPUs connect. For Arista Networks, Credo, and Astera Labs, the number is both a validation of the interconnect opportunity and a competitive warning around Nvidia's growing dominance in networking.
Anthropic will pay xAI $1.25B per month for all 300MW of compute capacity at xAI's Colossus 1 data center in Memphis, with the deal running through May 2029 and a total commitment exceeding $40B. Disclosed via a SpaceX SEC filing on May 20; Anthropic is also expanding into Colossus 2 on GB200 capacity starting June. The contract includes a 90-day exit clause for either party. xAI is monetizing spare Colossus capacity — Grok utilization has declined — by signing Anthropic as its anchor compute tenant, alongside its other new partner, Cursor. For Anthropic, the deal diversifies its supply stack beyond Google TPUs and Amazon Trainium, amidst a compute crunch for the company.(TechCrunch, Axios — May 20, 2026)
Nebius CRO Marc Boroditsky confirmed on the Q1 earnings call: "We just raised prices again in the latest quarter, and we are still selling out across all chip types at the higher prices" — with four or more customers competing for every GPU the company brings online. The margin tells the same story: AI cloud adjusted EBITDA expanded from 24% in Q4 to 45% in Q1, entirely consistent with a business raising price into inelastic demand. That combination — price up, utilization at 100%, margin expanding — is the clean signal of genuine pricing power rather than volume-led revenue growth. The risk is the Google and Blackstone $25B TPU cloud JV, which raised the competitive question when it was announced May 18 and contributed to Nebius's 3% decline on May 20. Watch contract duration in Q2 as the forward signal: if customers start shortening commitments ahead of the 2027 TPU capacity window, it will appear there before it shows up in revenue.
Six months after the U.S. Commerce Department cleared ten Chinese firms — Alibaba, ByteDance, Tencent among them — to purchase up to 75,000 H200 units each, not a single chip has shipped to China. The bottleneck is not Washington; it is Beijing. China's central government has not authorized its own companies to buy, a deliberate choice to protect domestic chip manufacturers from a flood of imported AI hardware. Jensen Huang confirmed zero China data center revenue on the May 20 earnings call and framed the question back at Beijing: "The Chinese government has to decide how much of their local market do they want to protect." The ~$50B annually China AI market sits locked by Chinese industrial policy. (Guru Focus, May 20, 2026)
Nvidia's Vera Rubin NVL72 rack — shipping in the second half of 2026 — requires 130–140 kW of cooling per rack. By 2027, the Vera Rubin Ultra platform is projected to push single-rack power to 200–300 kW. Liquid cooling capacity has already drawn level with air cooling globally; by end of 2026 it is expected to double air cooling capacity, and the NVL72 rack will be 100% liquid-cooled with no air cooling components at all. The adoption wave is also spreading beyond GPUs: CPUs, HBM, and network cards in next-gen racks are now entering the liquid cooling envelope, creating demand across the full cold-plate supply chain. Nvidia named four cold plate suppliers when it standardized Vera Rubin cooling at GTC. (Digitimes, May 19, 2026)
The Compute 100 is led by Brian Schechter and Gaby Lorenzi, compute-focused investors at Primary. Primary is a pre-seed and seed-stage venture firm that backs founders building across markets including compute, industrials, healthcare and vertical AI. With $1.6B AUM and a 60+ person operating team, Primary delivers unparalleled support to teams across recruiting, finance, GTM, and brand. Primary’s compute portfolio includes Etched, The Biological Computing Company, Haiqu, and Atero (acquired by Crusoe).