Thursday, October 1, 2026

Micron printed $54.2 billion fiscal fourth-quarter revenue (+379% year over year), 87% gross margin, and a $61.5 billion first-quarter guide versus $57 billion consensus (Sep 30) — the catalysts flagged yesterday. With 75%-plus of fiscal-2027 output committed against $32 billion in customer deposits, analysts attribute the muted stock reaction to cycle-durability skepticism; near-term margin guidance bears watching.

Google DeepMind launched Gemini 4 Argon on Sep 30 at $4/$20 per million tokens with a 1-million-token output ceiling; Artificial Analysis rates it level with GPT-6 Astra at roughly 60% of the per-task cost. Google says Argon agents freed 300+ TiB in its own data centers; with GPT-6.1 Sol already at $2/$10, frontier per-task price deflation bears watching.

Tencent agreed Oct 1 to lease 100,000 AI chips from Oracle across Asian data centers — a five-year, ~$7 billion deal with roughly 30% paid upfront. SpaceX's AI unit separately discussed leasing Microsoft capacity amid billions in monthly lab commitments; whether upfront-heavy, multi-year chip rental becomes the standard compute-supply structure bears watching.

Amazon signed a 20-year, 690-megawatt contract with Constellation Energy's Calvert Cliffs nuclear plant (Oct 1), with over $3 billion planned investment and ~190 megawatts of added capacity expected by 2030–2032. Grid hardware remains the chokepoint — transformer bushings exceed 130-week lead times, generators up to 100 weeks — making interconnection timing the watch variable.

Digitimes projects 25.95 million cloud AI accelerators shipping in 2026 (+64.5%), with ASICs passing GPUs in unit volume for the first time; Nvidia keeps top share at 35.1% (Oct 1). HPE's $1.2 billion liquid-cooled AMD Helios rack order from Vultr (Sep 30) reinforces the broadening; whether unit crossover translates into value share is the open question.

• Micron reported FY26 Q4 revenue of $54.23B (+379% YoY), adjusted EPS of $33.42, and 87% gross margin, all above consensus; Q1 FY27 revenue guided to $61.5B, well above the $57B estimate. (Sep 30)?
• Management said calendar 2027 and 2028 supply-demand conditions will be tighter than 2026, with over 75% of FY27 output already committed and 26 strategic customer agreements covering 35%+ of revenue through 2030, backed by $32B in customer cash deposits. (Sep 30)?
• Analysts noted the muted stock reaction reflects lingering market skepticism about cycle durability and near-term gross-margin guidance, even as the structural shift toward long-term agreements and HBM co-design deepens. (Sep 30)?
• Korea's September semiconductor exports hit a record $60.3B, up 263% year-over-year, with AI memory identified as the primary driver. (Oct 1)?

• Google DeepMind launched Gemini 4 Argon on Sep 30, pricing it at $4/$20 per 1M input/output tokens standard (discounted to $2/$10 for trusted testers), with a 1M-token output limit—up from 64K—targeting coding, enterprise, and cybersecurity. (Sep 30)?
• Artificial Analysis benchmarks put Gemini 4 Argon level with GPT-6 Astra on its Intelligence Index at roughly 60% of Astra's cost per task; the model reached #1 on the Text Arena leaderboard with 1,525 points. (Sep 30)?
• Google said Argon agents already freed over 300 TiB of memory inside its own data centers by analyzing profiling data and applying optimizations autonomously. (Sep 30)?
• OpenAI simultaneously released GPT-6.1 Sol, priced at $2/$10 with ~30% lower cost per task than GPT-6 Sol, positioning it as near-Astra intelligence at a fraction of Astra's price. (Oct 1)?

• HPE secured a $1.2B AI rack order from cloud firm Vultr using AMD Helios systems with MI455X GPUs, EPYC CPUs, Pensando networking, and liquid cooling; HPE also raised its 2026–2029 networking revenue growth outlook to high-teens CAGR annually. (Sep 30)?
• J.P. Morgan estimated Amazon's Trainium 3 liquid-cooled racks will reach ~$86,000 in cooling content per rack, exceeding Nvidia Vera Rubin racks at ~$66,250, driven by aggressive thermal layouts. (Oct 1)?
• Digitimes projected global cloud AI accelerator shipments to reach 25.95M units in 2026 (+64.5% YoY), with ASICs surpassing GPUs in unit volume for the first time; Nvidia retains the top share at 35.1% with 9.1M units. (Oct 1)?
• Data center component lead times remain extreme—pad-mounted transformers at up to 115 weeks and generators up to 100 weeks—creating bottlenecks that slow AI infrastructure build-out independent of GPU or memory availability. (Oct 1)?

• Shin-Etsu Chemical disclosed that a new wafer plant would cost more than twice its previous investment, underscoring why suppliers are reluctant to add capacity even as fab orders surge. (event date unknown)?
• Taiwanese substrate maker Unimicron acquired Bridgestone's Hsinchu factory for NT$10B to expand ABF substrate capacity as AI board demand pushes Q4 pricing higher. (Oct 1)?

• Samsung signaled that HBM will account for nearly 30% of industry DRAM capacity in 2027, benefiting conventional DRAM makers as capacity is converted; SK Hynix clarified no concrete plans exist yet for Solidigm's external financing despite IPO speculation. (Sep 29 / Oct 1)?
• At ECOC 2026, Marvell demonstrated 2nm chips for 1.6T coherent and O-band coherent-lite, reinforcing that AI optics is not solely a co-packaged optics story; silicon photonics foundry capacity—spanning TSMC, GlobalFoundries, Tower, STMicro and UMC—is emerging as the next bottleneck. (Sep 29 / Oct 1)?
• Experts noted HBM de-specification (shifting to more 8-layer stacks rather than higher layer counts) is a supply-side response, and that packaging thermal management at the hybrid-bonding layer is now a central integration constraint per ECTC 2026 papers. (Oct 1)?

• Tencent agreed to lease 100,000 AI chips from Oracle across Asian data centers in a five-year deal worth ~$7B, with ~30% paid upfront; SpaceX's AI unit separately held talks with Microsoft about leasing compute capacity, generating billions in monthly commitments from labs including Anthropic and Google. (Oct 1)?
• Amazon signed a 20-year, 690MW power contract with Constellation Energy's Calvert Cliffs nuclear plant in Maryland, with $3B+ in planned investment and ~190MW of additional capacity expected by 2030–2032. (Oct 1)?
• Ameren Missouri filed a 20-year resource plan adding 10.6 GW of gas capacity, while NTIA's BEAD broadband program directed 35% of 3.79M U.S. locations to satellite and fixed wireless over fiber. (Sep 30)?
• Transformer bushings for grid upgrades carry lead times exceeding 130 weeks at 345 kV, and no 765 kV units are yet made in the U.S., creating multi-year bottlenecks for data center power interconnection. (Oct 1)?

• OpenAI launched 'Dots,' always-on personal AI agents with their own compute, accessible via ChatGPT Pro; DoorDash announced iMessage-based food ordering via its AI assistant; and Robinhood unveiled AI agents for around-the-clock trading alongside expanded weekend market hours. (Sep 30 / Oct 1)?
• Meta's Muse enterprise platform surpassed 3M weekly users, with revenue quintupling, as the company opened its Muse API, Muse Code, and Muse agent to businesses; Vercel hit $600M annualized revenue (+148% YoY) as coding agents drove hosting demand. (Oct 1)?
• Synopsys and OpenAI signed a multi-year deal to build GPT-Synopsys, a specialized EDA model for chip design covering PPA optimization, timing closure and verification, with shared revenue. (Oct 1)?
• The FTC opened a probe into frontier AI labs including Anthropic and OpenAI, the first formal regulatory scrutiny of the sector's major players. (Sep 30)?

• Figure AI raised over $1B in Series C funding led by Parkway Venture Capital with participation from Nvidia, Intel Capital, Qualcomm Ventures and others; NEURA Robotics raised $1.4B and Apptronik $520M in September. (Sep 2026)?
• The International Federation of Robotics counted roughly 7,000 full-size humanoids sold in 2025 for commercial use, with most applications still specialized and many requiring teleoperation, tempering hype about autonomous deployment. (Sep 30 event date per IFR release)?
• China shipped over 19,000 humanoid robots in H1 2026, capturing ~78% of global volume per IDC data, cementing its lead in unit production even as autonomy gaps persist. (Oct 1)?
• Google DeepMind CEO Koray Kavukcuoglu described an Android-style open platform robotics strategy—licensing foundation models to many hardware partners—rather than building a single flagship humanoid. (Oct 1)?

• DeepSeek and Huawei are collaborating to develop programming infrastructure for Huawei's Ascend AI chips, deepening China's effort to reduce dependence on Nvidia. (Oct 1)?
• Chinese technology hardware stocks suffered their worst quarterly performance of the AI rally through end-September, despite underlying AI buildout momentum, as export restriction fears and valuation concerns weighed. (Sep 30)?
• ByteDance occupies over 20% of China's delivered data center capacity, making it the country's largest data center tenant as China holds the world's second-largest total DC capacity. (Oct 1)?
• AI concentration is crowding out broader startup investment in China and globally; 2025 saw the first decline in both deal count and venture capital investment since 2018. (Oct 1)?