<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>The Silicon Wire</title>
    <link>https://siliconwire.com</link>
    <description>Intelligence across the full AI supply chain. From rare earth extraction to frontier model deployment.</description>
    <language>en</language>
    <lastBuildDate>Thu, 12 Mar 2026 02:24:03 GMT</lastBuildDate>
    <atom:link href="https://siliconwire.com/feed.xml" rel="self" type="application/rss+xml"/>
    
    <item>
      <title><![CDATA[Standard Kernel Raises $20M to Bet That AI Can Write Its Own GPU Code]]></title>
      <link>https://siliconwire.com/article/standard-kernel-seed-round-ai-kernel-generation</link>
      <guid isPermaLink="true">https://siliconwire.com/article/standard-kernel-seed-round-ai-kernel-generation</guid>
      <description><![CDATA[Standard Kernel has raised a $20 million seed round led by Jump Capital to build an autonomous kernel generation platform — software that uses AI to write the low-level GPU code that AI itself runs on. The company claims 80% to 4x performance gains over NVIDIA's cuDNN on H100 workloads. If that holds at scale, the implications reach far beyond one startup.]]></description>
      <pubDate>Thu, 12 Mar 2026 02:15:00 GMT</pubDate>
      <category>Software & Models</category>
    </item>
    <item>
      <title><![CDATA[War in Iran, the $119 Oil Spike, and What It Actually Means for AI Infrastructure]]></title>
      <link>https://siliconwire.com/article/iran-war-oil-shock-ai-infrastructure-energy-crisis</link>
      <guid isPermaLink="true">https://siliconwire.com/article/iran-war-oil-shock-ai-infrastructure-energy-crisis</guid>
      <description><![CDATA[The US-Israeli strikes on Iran sent crude to $119. Prices retreated to $85. The conventional wisdom — that expensive oil threatens AI by raising electricity costs — is mostly wrong. Only 0.6% of US electricity comes from petroleum. The real threat is to new capacity: construction diesel, petroleum-derived materials, supply chain delays, and an $870 billion debt stack sensitive to every basis point of inflation-driven rate increases.]]></description>
      <pubDate>Tue, 10 Mar 2026 08:00:00 GMT</pubDate>
      <category>Infrastructure</category>
    </item>
    <item>
      <title><![CDATA[TSMC's $56 Billion Year: Inside the Largest Semiconductor Capex Budget in History]]></title>
      <link>https://siliconwire.com/article/tsmc-56-billion-capex-2026</link>
      <guid isPermaLink="true">https://siliconwire.com/article/tsmc-56-billion-capex-2026</guid>
      <description><![CDATA[TSMC will spend up to $56 billion in 2026 — more than any semiconductor company has ever invested in a single year. We break down where the money is going, from 2nm node buildout to a fourfold expansion of advanced packaging capacity, and what it means for chip pricing and customer allocation.]]></description>
      <pubDate>Mon, 09 Mar 2026 12:00:00 GMT</pubDate>
      <category>Materials & Fab</category>
    </item>
    <item>
      <title><![CDATA[NVIDIA GTC 2026: Vera Rubin in Production, Feynman on the Horizon]]></title>
      <link>https://siliconwire.com/article/nvidia-gtc-2026-vera-rubin-feynman</link>
      <guid isPermaLink="true">https://siliconwire.com/article/nvidia-gtc-2026-vera-rubin-feynman</guid>
      <description><![CDATA[NVIDIA's Vera Rubin platform — 336 billion transistors, 288GB of HBM4, 50 petaflops per GPU — is already in production and shipping to hyperscalers. But the real reveal at GTC 2026 may be Feynman: the first 1.6nm AI chip with silicon photonics, targeting 2028.]]></description>
      <pubDate>Mon, 09 Mar 2026 11:00:00 GMT</pubDate>
      <category>Chips & Architecture</category>
    </item>
    <item>
      <title><![CDATA[DeepSeek V3.2: How a Chinese Lab Matched Frontier Performance Under Export Controls]]></title>
      <link>https://siliconwire.com/article/deepseek-v3-2-frontier-under-export-controls</link>
      <guid isPermaLink="true">https://siliconwire.com/article/deepseek-v3-2-frontier-under-export-controls</guid>
      <description><![CDATA[DeepSeek's V3.2 — 685 billion parameters, 37 billion active per token — achieves gold at the IMO and matches GPT-5 on key benchmarks, all trained on export-restricted hardware. Its FP8 training framework and MoE innovations prove that chip restrictions may force innovation rather than prevent it. And V4, optimized for Huawei Ascend, signals something bigger.]]></description>
      <pubDate>Mon, 09 Mar 2026 10:00:00 GMT</pubDate>
      <category>Software & Models</category>
    </item>
    <item>
      <title><![CDATA[The HBM4 Race: SK Hynix, Samsung, and Micron Battle for NVIDIA's Memory Orders]]></title>
      <link>https://siliconwire.com/article/hbm4-race-sk-hynix-samsung-micron</link>
      <guid isPermaLink="true">https://siliconwire.com/article/hbm4-race-sk-hynix-samsung-micron</guid>
      <description><![CDATA[With NVIDIA's Vera Rubin demanding 288GB of HBM4 per GPU — 576 stacks per rack — the three memory giants are in an all-out production war. SK Hynix leads with ~70% of NVIDIA's allocation, Samsung is shipping its turnkey solution, and Micron is ramping 15,000 wafers per month.]]></description>
      <pubDate>Mon, 09 Mar 2026 09:00:00 GMT</pubDate>
      <category>Chips & Architecture</category>
    </item>
    <item>
      <title><![CDATA[The $690 Billion AI Capex Sprint: Where the Hyperscaler Money Is Going]]></title>
      <link>https://siliconwire.com/article/690-billion-ai-capex-sprint-2026</link>
      <guid isPermaLink="true">https://siliconwire.com/article/690-billion-ai-capex-sprint-2026</guid>
      <description><![CDATA[Amazon, Alphabet, Microsoft, Meta, and Oracle will collectively spend $660–690 billion on capital expenditure in 2026 — nearly doubling 2025 levels. We break down who is spending what, where it's going, and why the power grid has become the binding constraint.]]></description>
      <pubDate>Mon, 09 Mar 2026 08:00:00 GMT</pubDate>
      <category>Infrastructure</category>
    </item>
    <item>
      <title><![CDATA[TSMC Arizona Turns Profitable: What the First US Advanced Fab Means for Reshoring]]></title>
      <link>https://siliconwire.com/article/tsmc-arizona-turns-profitable-reshoring</link>
      <guid isPermaLink="true">https://siliconwire.com/article/tsmc-arizona-turns-profitable-reshoring</guid>
      <description><![CDATA[TSMC's Arizona subsidiary posted NT$16.1 billion in profit in 2025 — its first profitable year — just one year after Fab 21 entered volume production. With N4 yields exceeding Taiwan benchmarks, NVIDIA Blackwell GPUs rolling off the line, and a third fab breaking ground on 2nm, the most consequential test of US semiconductor reshoring is passing.]]></description>
      <pubDate>Mon, 09 Mar 2026 08:00:00 GMT</pubDate>
      <category>Materials & Fab</category>
    </item>
    <item>
      <title><![CDATA[The Liquid Cooling Inflection: Why Air Can't Cool a 1,200-Watt GPU]]></title>
      <link>https://siliconwire.com/article/liquid-cooling-inflection-1200-watt-gpu</link>
      <guid isPermaLink="true">https://siliconwire.com/article/liquid-cooling-inflection-1200-watt-gpu</guid>
      <description><![CDATA[NVIDIA's Blackwell GPUs consume 1,000 watts each. Vera Rubin will hit 1,800 watts. A single NVL72 rack draws 120 kilowatts. Air cooling physically cannot remove that much heat from that small a space. The liquid cooling market is projected to grow from $6 billion to $16 billion by 2030, and a wave of billion-dollar acquisitions — Vertiv, Schneider Electric, Eaton, Daikin — is reshaping the supply chain.]]></description>
      <pubDate>Mon, 09 Mar 2026 08:00:00 GMT</pubDate>
      <category>Infrastructure</category>
    </item>
    <item>
      <title><![CDATA[The EU Chips Act at Two Years: €69 Billion Mobilized, Strategic Gaps Remain]]></title>
      <link>https://siliconwire.com/article/eu-chips-act-two-years-69-billion</link>
      <guid isPermaLink="true">https://siliconwire.com/article/eu-chips-act-two-years-69-billion</guid>
      <description><![CDATA[Two years after entering force, the EU Chips Act has catalyzed €69 billion in public and private semiconductor investment. But the European Court of Auditors projects the EU will reach only 11.7% global market share by 2030 — far short of the 20% target. Intel Magdeburg is cancelled. The STMicroelectronics/GlobalFoundries French fab is shelved. And Chips Act 2.0 is already in the works.]]></description>
      <pubDate>Mon, 09 Mar 2026 08:00:00 GMT</pubDate>
      <category>Policy & Capital</category>
    </item>
    <item>
      <title><![CDATA[The MoE Revolution: How Mixture-of-Experts Became the Dominant Frontier Architecture]]></title>
      <link>https://siliconwire.com/article/moe-revolution-dominant-frontier-architecture</link>
      <guid isPermaLink="true">https://siliconwire.com/article/moe-revolution-dominant-frontier-architecture</guid>
      <description><![CDATA[Every major frontier model released in the past year uses Mixture-of-Experts. DeepSeek V3.2: 685B parameters, 37B active. Llama 4 Behemoth: 2 trillion total, 288B active. Gemini, Mixtral, and reportedly GPT-4 — all MoE. NVIDIA says Blackwell runs MoE 10x faster at 1/10th the token cost. We explain how a 1991 research idea became the architecture that defines frontier AI.]]></description>
      <pubDate>Mon, 09 Mar 2026 08:00:00 GMT</pubDate>
      <category>Software & Models</category>
    </item>
    <item>
      <title><![CDATA[ASML Enters Advanced Packaging: The $50 Billion Opportunity Beyond EUV]]></title>
      <link>https://siliconwire.com/article/asml-enters-advanced-packaging-50-billion-opportunity</link>
      <guid isPermaLink="true">https://siliconwire.com/article/asml-enters-advanced-packaging-50-billion-opportunity</guid>
      <description><![CDATA[ASML shipped its first advanced packaging lithography system — the TWINSCAN XT:260 — in late 2025, marking the company's strategic expansion beyond EUV. With the advanced packaging market surpassing $50 billion annually and TSMC's CoWoS capacity constraining every major AI chip program, the lithography monopolist is opening a new competitive front.]]></description>
      <pubDate>Mon, 09 Mar 2026 08:00:00 GMT</pubDate>
      <category>Materials & Fab</category>
    </item>
    <item>
      <title><![CDATA[Japan's $26 Billion Data Center Paradox: Hyperscaler Demand Meets Power Constraints]]></title>
      <link>https://siliconwire.com/article/japan-26-billion-data-center-paradox</link>
      <guid isPermaLink="true">https://siliconwire.com/article/japan-26-billion-data-center-paradox</guid>
      <description><![CDATA[AWS, Microsoft, Oracle, and Google have committed $26 billion to data center infrastructure in Japan. Simultaneously, TSMC and Rapidus are building advanced fabs that will consume gigawatts of power. The problem: Japan's grid can't deliver both.]]></description>
      <pubDate>Mon, 09 Mar 2026 08:00:00 GMT</pubDate>
      <category>Infrastructure</category>
    </item>
    <item>
      <title><![CDATA[Meta's Inference Fleet Transformation: 35% Custom Silicon by Year-End]]></title>
      <link>https://siliconwire.com/article/meta-inference-fleet-transformation-35-percent-custom-silicon</link>
      <guid isPermaLink="true">https://siliconwire.com/article/meta-inference-fleet-transformation-35-percent-custom-silicon</guid>
      <description><![CDATA[Meta is executing the most aggressive custom silicon transition in tech history. MTIA v2 is deployed across 16 data center regions. MTIA v3 (Iris) entered broad deployment in February 2026. The target: 35% of inference on custom chips by year-end, with a 44% TCO reduction vs. GPUs.]]></description>
      <pubDate>Mon, 09 Mar 2026 08:00:00 GMT</pubDate>
      <category>Software & Models</category>
    </item>
    <item>
      <title><![CDATA[Cerebras Targets Q2 IPO: Inside the $23 Billion Challenger to NVIDIA]]></title>
      <link>https://siliconwire.com/article/cerebras-targets-q2-ipo-23-billion-challenger</link>
      <guid isPermaLink="true">https://siliconwire.com/article/cerebras-targets-q2-ipo-23-billion-challenger</guid>
      <description><![CDATA[Cerebras Systems has filed confidentially for an IPO targeting Q2 2026, with Morgan Stanley as lead underwriter and a $23 billion valuation. The company claims 21x faster inference than Blackwell, has a $10 billion OpenAI deal, and has resolved the G42 controversy.]]></description>
      <pubDate>Mon, 09 Mar 2026 08:00:00 GMT</pubDate>
      <category>Policy & Capital</category>
    </item>
    <item>
      <title><![CDATA[Japan's Semiconductor Revival: TSMC Expands AI Chip Production to Third Advanced Base]]></title>
      <link>https://siliconwire.com/article/japan-semiconductor-revival-tsmc-3nm-expansion</link>
      <guid isPermaLink="true">https://siliconwire.com/article/japan-semiconductor-revival-tsmc-3nm-expansion</guid>
      <description><![CDATA[TSMC has upgraded its second Kumamoto fab to produce 3nm AI chips — elevating Japan to its third advanced manufacturing base alongside Taiwan and Arizona.]]></description>
      <pubDate>Mon, 09 Mar 2026 08:00:00 GMT</pubDate>
      <category>Materials & Fab</category>
    </item>
    <item>
      <title><![CDATA[Alphabet Loses TPU Production Slots as NVIDIA Locks Up CoWoS Capacity]]></title>
      <link>https://siliconwire.com/article/alphabet-loses-tpu-production-slots-cowos</link>
      <guid isPermaLink="true">https://siliconwire.com/article/alphabet-loses-tpu-production-slots-cowos</guid>
      <description><![CDATA[Google has cut its 2026 TPU production target from four million to three million units after NVIDIA secured over 50% of TSMC's CoWoS advanced packaging capacity through 2027.]]></description>
      <pubDate>Mon, 09 Mar 2026 08:00:00 GMT</pubDate>
      <category>Chips & Architecture</category>
    </item>
    <item>
      <title><![CDATA[The 16-Hi HBM4 Push: NVIDIA's Demand Is Rewriting Memory Roadmaps]]></title>
      <link>https://siliconwire.com/article/16-hi-hbm4-nvidia-rewriting-memory-roadmaps</link>
      <guid isPermaLink="true">https://siliconwire.com/article/16-hi-hbm4-nvidia-rewriting-memory-roadmaps</guid>
      <description><![CDATA[NVIDIA has requested all three major memory suppliers deliver 16-layer HBM4 by Q4 2026. SK Hynix debuted a 48GB, 2+ TB/s module at CES 2026. The demand is compressing roadmaps designed for 2027 into 2026 deliverables.]]></description>
      <pubDate>Mon, 09 Mar 2026 08:00:00 GMT</pubDate>
      <category>Chips & Architecture</category>
    </item>
    <item>
      <title><![CDATA[Microsoft's $80 Billion Problem: When AI Demand Outpaces Infrastructure]]></title>
      <link>https://siliconwire.com/article/microsoft-80-billion-problem-azure-power-constraints</link>
      <guid isPermaLink="true">https://siliconwire.com/article/microsoft-80-billion-problem-azure-power-constraints</guid>
      <description><![CDATA[Microsoft has disclosed an $80 billion backlog of Azure AI orders it cannot fulfill — not because of demand weakness, but because it lacks the electricity to power the GPUs already in inventory.]]></description>
      <pubDate>Mon, 09 Mar 2026 08:00:00 GMT</pubDate>
      <category>Infrastructure</category>
    </item>
    <item>
      <title><![CDATA[DeepSeek V4: The First Trillion-Parameter Open-Weight Model]]></title>
      <link>https://siliconwire.com/article/deepseek-v4-trillion-parameter-open-weight</link>
      <guid isPermaLink="true">https://siliconwire.com/article/deepseek-v4-trillion-parameter-open-weight</guid>
      <description><![CDATA[DeepSeek has released V4 — approximately one trillion total parameters with 32 billion active per token, a million-token context window, and native multimodal capabilities. Optimized for Huawei Ascend and open-sourced under Apache 2.0.]]></description>
      <pubDate>Mon, 09 Mar 2026 08:00:00 GMT</pubDate>
      <category>Software & Models</category>
    </item>
    <item>
      <title><![CDATA[NVIDIA's Blackwell NVL72: 10x MoE Performance Changes the Inference Math]]></title>
      <link>https://siliconwire.com/article/nvidia-blackwell-nvl72-10x-moe-inference</link>
      <guid isPermaLink="true">https://siliconwire.com/article/nvidia-blackwell-nvl72-10x-moe-inference</guid>
      <description><![CDATA[NVIDIA's GB200 NVL72 delivers 10x faster inference for MoE models at one-tenth the cost per token. Vera Rubin promises another 5x when it ships in H2 2026.]]></description>
      <pubDate>Mon, 09 Mar 2026 08:00:00 GMT</pubDate>
      <category>Software & Models</category>
    </item>
    <item>
      <title><![CDATA[The 25% AI Chip Tariff: How Trump's Section 232 Proclamation Reshapes Semiconductor Trade]]></title>
      <link>https://siliconwire.com/article/25-percent-ai-chip-tariff-section-232</link>
      <guid isPermaLink="true">https://siliconwire.com/article/25-percent-ai-chip-tariff-section-232</guid>
      <description><![CDATA[President Trump imposed a 25% tariff on advanced AI chips not destined for the US supply chain, effective January 15, 2026. Broad exemptions protect domestic data center buildout while penalizing international chip flows.]]></description>
      <pubDate>Mon, 09 Mar 2026 08:00:00 GMT</pubDate>
      <category>Policy & Capital</category>
    </item>
    <item>
      <title><![CDATA[Nine EU Nations Form Semiconductor Coalition as Chips Act 2.0 Looms]]></title>
      <link>https://siliconwire.com/article/nine-eu-nations-semiconductor-coalition</link>
      <guid isPermaLink="true">https://siliconwire.com/article/nine-eu-nations-semiconductor-coalition</guid>
      <description><![CDATA[Austria, Belgium, Finland, France, Germany, Italy, Poland, Spain, and the Netherlands have formed a Semiconductor Coalition pushing for €200 billion in investment by 2035.]]></description>
      <pubDate>Mon, 09 Mar 2026 08:00:00 GMT</pubDate>
      <category>Policy & Capital</category>
    </item>
    <item>
      <title><![CDATA[Weekly Briefing #1: The Week NVIDIA Went to GTC and the World Followed]]></title>
      <link>https://siliconwire.com/article/weekly-briefing-001-gtc-week</link>
      <guid isPermaLink="true">https://siliconwire.com/article/weekly-briefing-001-gtc-week</guid>
      <description><![CDATA[Our inaugural weekly briefing synthesizes the biggest stories across the AI supply chain: GTC 2026 previews, the HBM4 race, TSMC's record capex, Cerebras's IPO, the EU Semiconductor Coalition, and the liquid cooling inflection.]]></description>
      <pubDate>Mon, 09 Mar 2026 08:00:00 GMT</pubDate>
      <category>Infrastructure</category>
    </item>
    <item>
      <title><![CDATA[Stargate's $500 Billion Bet: Inside the Largest AI Infrastructure Project in History]]></title>
      <link>https://siliconwire.com/article/stargate-500-billion-bet</link>
      <guid isPermaLink="true">https://siliconwire.com/article/stargate-500-billion-bet</guid>
      <description><![CDATA[The Stargate joint venture — OpenAI, SoftBank, Oracle, and MGX — has committed over $400 billion across six US sites with nearly 7 GW of planned capacity. But months of partner disputes over control nearly derailed the project before it began.]]></description>
      <pubDate>Mon, 09 Mar 2026 07:00:00 GMT</pubDate>
      <category>Infrastructure</category>
    </item>
    <item>
      <title><![CDATA[Trump's AI Chip Policy: Rescinding Biden's Diffusion Rule, Easing China Restrictions]]></title>
      <link>https://siliconwire.com/article/trump-ai-chip-policy-diffusion-rule-export-controls</link>
      <guid isPermaLink="true">https://siliconwire.com/article/trump-ai-chip-policy-diffusion-rule-export-controls</guid>
      <description><![CDATA[In fourteen months, the Trump administration rescinded Biden's AI Diffusion Rule, shifted H200 export reviews from "presumption of denial" to case-by-case, and imposed a 25% tariff on chip exports to China. Congress is pushing back with the GAIN AI Act and AI Overwatch Act.]]></description>
      <pubDate>Mon, 09 Mar 2026 07:00:00 GMT</pubDate>
      <category>Policy & Capital</category>
    </item>
    <item>
      <title><![CDATA[ASML's High-NA EUV Rollout: Intel First, Samsung Second, TSMC Waiting]]></title>
      <link>https://siliconwire.com/article/asml-high-na-euv-rollout-intel-samsung-tsmc</link>
      <guid isPermaLink="true">https://siliconwire.com/article/asml-high-na-euv-rollout-intel-samsung-tsmc</guid>
      <description><![CDATA[ASML's $380 million TWINSCAN EXE:5200B is the most expensive machine tool ever built. Intel completed acceptance testing in December 2025. Samsung is deploying High-NA for 2nm. SK Hynix is using it for DRAM. TSMC says it doesn't need it until 2029. We map who has which machines and why TSMC's decision to skip High-NA reshapes the competitive landscape.]]></description>
      <pubDate>Mon, 09 Mar 2026 06:00:00 GMT</pubDate>
      <category>Materials & Fab</category>
    </item>
    <item>
      <title><![CDATA[Meta's Custom Silicon Roadmap: From MTIA to Full Inference Stack]]></title>
      <link>https://siliconwire.com/article/meta-custom-silicon-roadmap-mtia-inference</link>
      <guid isPermaLink="true">https://siliconwire.com/article/meta-custom-silicon-roadmap-mtia-inference</guid>
      <description><![CDATA[Meta is building the most aggressive custom silicon program outside of Google. MTIA v2 delivers 44% lower TCO than GPUs for inference. Iris (v3) is in broad deployment on TSMC 3nm with HBM3E. The goal: 35% of Meta's inference fleet on custom silicon by year-end 2026 — while still buying millions of NVIDIA GPUs for training.]]></description>
      <pubDate>Mon, 09 Mar 2026 05:00:00 GMT</pubDate>
      <category>Chips & Architecture</category>
    </item>
  </channel>
</rss>