The US government controls which companies can buy advanced AI chips — the agency that manages those rules is called the Bureau of Industry and Security (BIS). The rules say Chinese companies can't buy certain Nvidia and AMD chips without a special license. But until now, the rules applied based on where a company is located, not who owns it. That created an obvious workaround: Chinese companies set up subsidiaries in Malaysia, Singapore, or other countries, and those subsidiaries bought the chips freely — legally — because they weren't Chinese addresses. Tencent was one of them. Supply-chain sources cited by Al Jazeera, TrendForce, and Tom's Hardware estimate that hundreds of thousands of Nvidia Blackwell and AMD MI350x chips moved through this gap since at least May 2025. Sunday's guidance closes it: BIS now looks at who owns the company, not where it's incorporated. One hole remains — chip manufacturers like TSMC still aren't required to check whether their customers are shell companies for Chinese buyers. The rules changed at the buyer level, but not yet at the factory level.
Lambda announced at GTC Taipei on June 1 that it is a launch partner for NVIDIA's Vera CPU platform and is leading one of the largest NVIDIA Quantum-X InfiniBand Photonics deployments — a production cluster with 10,000+ GB300 GPUs. Lambda is also among the first adopters of NVIDIA BlueField-4 STX for rack-scale AI storage. Bare metal instances on Vera Rubin NVL72 superclusters are slated for H2 2026. The photonics commitment at 10,000-GPU scale is a genuine infrastructure milestone: Lambda is being named as the neocloud most visibly building at that hardware tier.
Firmus was featured in NVIDIA's AI Cloud ecosystem expansion at GTC Taipei on June 1, highlighted for its Project Southgate AI factory rollout across Australia and the APAC region — deployments in Tasmania, Melbourne, South Australia, and New South Wales — using the company's liquid-cooled HyperCube AI factory design. Firmus raised $505M in April backed by Nvidia and is the named reference deployment partner for sovereign AI infrastructure outside the US and Europe. The GTC spotlight is a validation signal in a market where hyperscaler procurement decisions track visible proof-of-deployment, not just capital raised.
Marvell announced the Teralynx T100 on June 1 — the industry's first 102.4 Tbps switch silicon built on 3nm and purpose-built for AI. The T100 delivers 25% lower power than competing switch ASICs, runs under 1,000W typical, and supports 512-port scale-out radix with SAI/SONiC software. Sampling begins this quarter. The stock closed at an all-time high, building on the May 27 Q1 FY2027 earnings beat: revenue $2.418B (+28% YoY), data center at 76% of mix ($1.83B), non-GAAP EPS $0.80. Marvell raised FY2027 guidance to ~$11.5B (+40% YoY) and FY2028 to ~$16.5B (+45% YoY) — the clearest public articulation yet of where custom ASIC revenue lands at scale.
HPE reported Q2 FY2026 results after market close on June 1: revenue $10.7B (+40% YoY) against a $9.77B estimate; non-GAAP EPS $0.79 vs $0.53 expected — the biggest earnings beat since 2018 per CNBC. AI Systems backlog entering Q3: $5.9B. New AI orders during Q2: $1.8B. CEO Antonio Neri described traditional server bookings as up triple digits and the company backlog as the largest in its history. HPE raised its fiscal 2026 revenue growth outlook to 29–33% (from 17–22%). The stock was up 9.4% during the regular session on pre-earnings momentum, then surged an additional ~32% in after-hours trading on the print — the full-day picture is a stock that nearly doubled from Friday's close through extended trading.
TSMC 3nm pricing: TrendForce reported last week that TSMC is planning a 15% price increase for its 3nm node in H2 2026, with a potential additional 5–10% in 2027. Current 3nm wafer prices run approximately $20,000 each. AI and ASIC demand from Nvidia, AMD, Google, and AWS is running roughly three times available manufacturing capacity. The pricing trajectory signals supply constraint will persist for years — bullish for TSMC margins, a structural squeeze on fabless customers with contracted pricing that doesn't adjust automatically.
SIA–Deloitte AI chip report: The Semiconductor Industry Association released a report on June 1 finding that semiconductors account for more than 95% of a leading AI server rack's content value and over 50% of total data center capex. Cumulative data center infrastructure investment is projected at more than $4T through 2028, of which approximately $2.8T goes to chips. Annual AI data center chip revenue is projected at $1.2T by 2028 — roughly 10x growth in four years. The quantification makes explicit what market pricing has been implying: the AI build-out is structurally a semiconductor spending cycle, and the infrastructure layer is secondary to silicon in economic terms.
Intel CEO Lip-Bu Tan keynote at Computex (1:30 pm GMT+8). First major competitive response to Nvidia's GTC Taipei AI factory announcements.
Broadcom fiscal Q2 2026 earnings after market close. Watch AI ASIC revenue vs. $10.7B guide and updated commentary on hyperscaler custom silicon programs and design-win trajectory.
GTC Taipei continues through June 4-5. Additional partner announcements expected from 80+ Nvidia MGX ecosystem members. CoreWeave, Lambda, and Nscale Vera Rubin deployment specifics likely to surface.