Dell's Q1 FY2027 AI server revenue reached $16.1 billion — up 757% year over year — against a consensus total revenue estimate of $34.95 billion; the company delivered $43.84 billion in total revenue and raised its full-year AI server guidance from $50 billion to $60 billion, sending shares up 30% in after-hours trading.
This wraps an incredible week for public markets, with Micron and SK Hynix joining the $1T club, Marvell custom chip revenue soaring, and continued momentum behind Nvidia, TSMC, and compute incumbents.
In the private markets, this week's story centered around Anthropic's Series H, compute trading markets, and Huawei's Tau scaling law, declaring a potential path beyond Moore's Law.
Next week, we're watching announcements at Computex, Broadcom earnings, and startup activity closely.
CEO Park Sung-hyun keynoted Seoul Forum 2026 on May 28, projecting the global AI inference chip market will grow from $39 billion in 2024 to $475 billion by 2030 and claiming Rebellions' NPUs deliver 5–7x price-efficiency versus GPU-based inference systems. The day before, Rebellions signed an MOU with KB Financial Group — South Korea's largest financial holding company — for NPU inference infrastructure deployment in exchange for financial services and funding support. The company carries ~$950 million in cumulative investment at a $2.5 billion valuation; strategic investors include Samsung Electronics, SK Hynix, ARM, and Marvell. As the first direct recipient of South Korea's National Growth Fund, Rebellions stacks sovereign capital, chaebol strategic alignment, and a named bank as the first customer — a go-to-market architecture distinct from Western inference chip peers.
TechCrunch today profiled General Compute, the first inference neocloud deploying SambaNova's SN40 and SN50 dataflow chips at commercial scale. The company raised a $15M seed at $60M post-money valuation from FUSE VC and went live on May 22 with $300 million of SN50 chips on order. Its cloud is designed so AI coding agents — not humans — complete the onboarding flow autonomously. Claimed performance: SN50 generates 600–700 tokens per second versus ~250 for GPUs, air-cooled and installable in standard data centers. General Compute's $300 million in on-order chips is the a real read on SN50 commercial deployment velocity.
SiMa.ai and Emerson announced a partnership embedding SiMa.ai's MLSoC into Emerson's industrial PCs for on-premise Physical AI inference in factories and remote field environments — no cloud connectivity required. Emerson's distribution across oil and gas, power generation, and manufacturing provides SiMa.ai with an enterprise anchor with built-in reach into heavy industry — the critical go-to-market step that purpose-built edge AI chips need to displace general-purpose silicon in industrial settings.
Wiwynn — a major AI server ODM and Foxconn subsidiary — announced on May 28 it will demonstrate a co-packaged optics rack built around Ayar Labs' TeraPHY optical engines at Computex 2026 (June 2–5, Taipei). The demonstration integrates an AI ASIC in a 2.5D advanced package within an optically connected rack targeting 1,024-GPU-scale deployments. Ayar Labs carries both Nvidia and AMD as strategic investors from its $500M Series E at $3.75 billion valuation — the only active CPO company with backing from both GPU incumbents. The Computex demonstration puts a production-form-factor reference design in front of hyperscaler architects for post-copper scale-up interconnects; first volume products target 2028.
Arm hit an all-time high of $326.12, driven by Snowflake's $6 billion Amazon cloud spending commitment explicitly naming Graviton CPUs — a direct royalty read-through. Every CPU deployed for AI inference (Graviton, Apple Silicon, Qualcomm Oryon, Nvidia Grace) pays Arm a per-chip royalty; the agentic AI shift from GPU-only to CPU+GPU hybrid architectures is the primary structural tailwind for Arm's licensing revenue, and hyperscaler infrastructure commitments are now beginning to quantify it.
Supermicro announced collaboration with Taiwanese authorities to seize 50 Nvidia-equipped servers being illicitly diverted to China, resulting in three arrests. The market read this as proactive compliance leadership — reducing the legal overhang from co-founder Yih-Shyan Liaw's prior $2.5 billion smuggling indictment and signaling active cooperation with export enforcement, which is critical for maintaining Nvidia GPU allocation access.
Co-Packaged Optics: Manufacturability Moves Commercial
AuthenX announced on May 28 its Detachable 2D Fiber Array Unit — a field-serviceable CPO packaging module using Meta-Lens nanostructures for sub-micron beam alignment — targeting a first demonstration at Computex June 2–5. The detachable architecture addresses CPO's primary production barrier: existing modules require permanent fiber attach during assembly, making field repair impossible at hyperscale. AuthenX is the second company in 2026 to announce a detachable FAU after Lightmatter's vClick Optics in March — arriving alongside the Ayar Labs/Wiwynn rack demonstration and GlobalFoundries' OCI MSA-compliant SCALE CPO platform. The CPO manufacturability problem is being attacked from multiple commercial angles simultaneously; the packaging bottleneck timeline is compressing.
Dell opens for its first regular-session trading after Thursday's earnings beat and guidance raise — AI server revenue $16.1B in Q1, full-year guidance raised from $50B to $60B, shares +30% in after-hours trading. The first open is the sustained confirmation signal for the Q2 capex read-through.
Computex 2026, Taipei. Intel CEO Lip-Bu Tan presents the full 18A product stack (Panther Lake / Arc G3 already shipping; Nova Lake desktop and Clearwater Forest Xeon to be detailed). Qualcomm CEO Cristiano Amon and Marvell CEO Matt Murphy keynoting. Ayar Labs and Wiwynn CPO rack demonstration. Watch for Intel foundry customer disclosures beyond the confirmed Apple partnership — Amazon is in active discussions per prior reporting.