エピソード

  • Apple's $30B Broadcom Deal, H200 China Exemptions & Private Credit's $700B Data Centre Bet

    Apple's $30B Broadcom Deal, H200 China Exemptions & Private Credit's $700B Data Centre Bet
    2026/07/09
    (00:00:00) Apple's $30B Broadcom Deal, H200 China Exemptions & Private Credit's $700B Data Centre Bet
    (00:00:29) Broadcom's Five-Customer ASIC Platform
    (00:01:27) US Onshoring: Scale and Execution Risk
    (00:02:06) China's H200 Exemption Window
    (00:02:39) Private Credit's $700B Infrastructure Bet
    (00:03:17) TSMC Chokepoint and Stranded Asset Risk
    (00:03:48) Key Watchpoints

    Apple has committed thirty billion dollars to Broadcom for custom RF and ASIC chips through 2031 — making it the fifth hyperscaler-tier customer on Broadcom's platform, alongside Google, Meta, ByteDance, and OpenAI. This episode examines what that consolidation means for Broadcom's competitive position as a genuine structural alternative to Nvidia for companies determined to own their silicon rather than rent it.

    Broadcom is investing one-and-a-half billion dollars in its Fort Collins facility to support a target of fifteen billion domestically manufactured chips. The commitment is real, but the execution risks — labour availability, yield rates, and US cost structure — remain unresolved. Apple's deal is also the most concrete single capex expression of the Trump administration's domestic chip manufacturing push yet.

    Elsewhere, China is permitting limited Nvidia H200 purchases for its top AI firms through a case-by-case regulatory exemption process. That signals a shift from blanket export controls to managed access — a channel, not a reopened market.

    On the capital side, Apollo, Blackstone, KKR, and Ares have all launched data centre financing platforms in the last thirty days. Private credit is targeting roughly seven hundred billion dollars of the estimated one-point-seven-five trillion needed in global data centre capex through 2028. But US relative value is already compressing as banks enter aggressively — Europe and power-and-cooling infrastructure are the less saturated plays.

    Underpinning all of it: TSMC remains the production chokepoint for advanced AI silicon, and no amount of private capital resolves a geopolitical disruption to Taiwan. Debt investors are already shortening loan terms to manage GPU obsolescence risk. The ten-year infrastructure thesis is under active revision.

    This episode includes AI-generated content.
    続きを読む 一部表示
    4 分
  • Export Controls, DeepSeek Silicon & SK Hynix's $28B Nasdaq IPO

    Export Controls, DeepSeek Silicon & SK Hynix's $28B Nasdaq IPO
    2026/07/08
    (00:00:00) Export Controls, DeepSeek Silicon & SK Hynix's $28B Nasdaq IPO
    (00:00:52) DeepSeek Enters Chip Design Race
    (00:01:51) Huawei's Guaranteed Demand Windfall
    (00:02:30) SK Hynix Nasdaq IPO Signals HBM Conviction
    (00:03:08) Power Grid Bottleneck Crystallizes
    (00:03:50) Intel Retreats, Valuation Test Ahead
    (00:04:35) Key Watchpoints Going Forward

    The US Bureau of Industry and Security has tightened export control guidance, shifting compliance burden directly onto TSMC and other foundries — and the consequences are cascading across the entire AI hardware supply chain. Today's briefing covers six interconnected stories that define where the industry is heading.

    DeepSeek, the Chinese AI startup known for inference efficiency, is now actively developing custom inference ASICs and recruiting chip engineers — a move that signals structural, not just tactical, independence from Nvidia and Huawei's Ascend line. Meanwhile, those same export controls are handing Huawei a guaranteed demand windfall, as Alibaba, ByteDance, and Tencent are pushed into Ascend and Zhenwu 810E orders with no US alternative available.

    On the memory side, SK Hynix has moved to active Nasdaq IPO marketing, targeting around $28 billion in American Depositary Receipts with roughly $7 billion in pre-commitments from Baillie Gifford, Coatue, and Situational Awareness Partners. Proceeds are earmarked for ASML EUV equipment — a clear bet on sustained HBM demand for AI accelerators. Pricing finalizes around July 10th.

    Intel's stock pulled back sharply from a record $142 despite 22% data center revenue growth, as valuation rotation hits the semiconductor sector ahead of its July 23rd earnings. And NTT Data's analysis confirms that power grid constraints — not capital or chip availability — are now the binding limit on data centre expansion in key regions.

    The through-line: export controls intended to slow China's AI hardware build-out are instead locking in Huawei's position, accelerating domestic chip programs, and offloading compliance risk onto foundries. The gap between policy intent and real-world outcome is widening fast.

    This episode includes AI-generated content.
    続きを読む 一部表示
    6 分
  • Kyber Delay, Micron's Long-Term AI Contracts & Singapore GPU Diversion | Ep 1

    Kyber Delay, Micron's Long-Term AI Contracts & Singapore GPU Diversion | Ep 1
    2026/07/07
    (00:00:00) Kyber Delay, Micron's Long-Term AI Contracts & Singapore GPU Diversion | Ep 1
    (00:01:10) NVL576 and Cluster Timeline Risk
    (00:01:53) Micron Locks AI Memory Supply
    (00:02:53) Singapore Chip Diversion Prosecution
    (00:03:41) Intel Valuation Gap

    Nvidia's most ambitious rack-scale system, the Kyber NVL144, has been pushed back to 2028 — not because of software or yield issues, but because the midplane PCB at its core cannot yet be reliably manufactured at scale. That physical constraint has direct consequences for the NVL576 cluster system, which depends on Kyber racks linking together optically. For hyperscalers planning billion-dollar infrastructure bets around that architecture, this is more than a product schedule story.

    Micron, meanwhile, is moving decisively in the other direction. The company has signed 16 long-term AI memory supply contracts running through 2030, complete with price floors and ceilings. Hyperscalers are reserving high-bandwidth memory years in advance — a signal that HBM is now treated as mission-critical infrastructure, not a commodity. The old boom-and-bust memory cycle may not apply here. The risk to watch: whether South Korea's planned capacity expansion skews toward commodity DRAM rather than AI-optimised HBM4.

    On the export control front, Singapore's prosecution of Aperia Group executives is escalating. Executives now face money laundering charges, with bail raised to 1.2 million dollars, in a case alleging 55 million dollars in proceeds from diverting Nvidia GPU servers through Dell and Super Micro into China. Enforcement is becoming personal — individuals face prison time, not just corporate fines.

    Finally, Intel's market cap sits at 528 billion dollars against foundry revenue of just 174 million dollars last quarter, with analysts projecting 21% downside. The gap between Intel's valuation and its foundry reality is the sharpest test of whether any bull case holds.

    This episode includes AI-generated content.
    続きを読む 一部表示
    5 分
  • SK Hynix NASDAQ, Samsung's Record Q2 & HBM Supply Crunch

    SK Hynix NASDAQ, Samsung's Record Q2 & HBM Supply Crunch
    2026/07/06
    (00:00:00) SK Hynix NASDAQ, Samsung's Record Q2 & HBM Supply Crunch
    (00:01:03) Samsung Record Q2 Profit
    (00:01:55) HBM CapEx Lag Problem
    (00:02:32) Qualcomm's Hyperscaler CPU Deals
    (00:03:08) Consumer Memory Price Squeeze
    (00:03:41) Geopolitical Risk and Watchpoints

    SK Hynix is pursuing a NASDAQ listing — and it's one of the most strategically loaded capital markets moves in the semiconductor industry this year. This episode unpacks why the world's leading HBM supplier is repositioning its centre of gravity toward US customers and regulators, and what that signals for the AI memory market.

    Samsung, meanwhile, has guided for a record Q2 operating profit of 86 trillion won — an eighteen-fold year-on-year jump driven by HBM3 and HBM3E demand and a structurally undersupplied DRAM and NAND market. Samsung, SK Hynix, and Micron have all crossed the one-trillion-dollar market cap threshold this year. That's not a rally; it's a structural repricing of what memory is worth in an AI-driven world.

    But the supply outlook is tighter than the headlines suggest. Samsung and SK Hynix have committed a combined 3.2 trillion won in HBM capex through 2040 — yet new fabs take three to five years to ramp. The shortage is effectively locked in through late 2027, regardless of capital deployed today.

    Elsewhere, Qualcomm has disclosed three major hyperscaler CPU deals targeting 15 billion dollars in revenue by 2029, directly threatening Intel's grip on data centre CPU supply. And for consumers, DRAM and NAND tightening driven by AI infrastructure demand is pushing 32GB DDR5 kits toward 110 dollars by mid-2026 — colliding directly with PC refresh cycles tied to upcoming Windows upgrades.

    The episode closes with the geopolitical watchpoints: US export controls, CXMT and YMTC capacity builds in China, and the metrics that will define whether the HBM supercycle sustains through the supply ramp.

    This episode includes AI-generated content.
    続きを読む 一部表示
    5 分
  • Micron's Hiroshima Gamble, TSMC Pricing Power & AMD's CPU Surge

    Micron's Hiroshima Gamble, TSMC Pricing Power & AMD's CPU Surge
    2026/07/05
    (00:00:00) Micron's Hiroshima Gamble, TSMC Pricing Power & AMD's CPU Surge
    (00:00:47) HBM Sold Out Through 2028
    (00:01:36) Samsung SK Hynix Already Ahead
    (00:02:09) TSMC Pricing Power Test
    (00:02:55) AMD CPU Surge and Memory Crisis
    (00:03:41) Laptop GPU Now Leads Desktop
    (00:04:13) Key Watchpoints Ahead

    Micron has broken ground on a $9.3 billion factory expansion in Hiroshima, backed by roughly 500 billion yen in Japanese government subsidies — a signal that HBM undersupply is a structural condition, not a short-term glitch. With orders already locked through 2027 and into 2028 via multi-year take-or-pay contracts, Barclays has raised its price target on Micron to $2,000, citing a decade-long supply deficit. The open question is whether AI data-centre spending stays elevated long enough to meet the 2028 production ramp.

    On the foundry side, TSMC raised chip prices by up to 10% and committed $20 billion to its Arizona expansion, sending its stock up nearly 4% and prompting a Goldman Sachs price target upgrade. TSMC's pricing power rests on being irreplaceable for leading-edge logic — but Samsung's growing interest in inference ASIC production and customers already absorbing surging memory costs will eventually sharpen the search for alternatives.

    In consumer hardware, Gartner forecasts a 130% surge in DRAM and SSD prices by end-2026, with DDR5 kit prices already up roughly 400% since 2025. AMD has capitalised: its X3D cache architecture reduces dependence on expensive external memory, helping it reach nearly 46% CPU share among Steam gamers — up 14 percentage points in 18 months. In a related first, the Nvidia RTX 4060 Laptop GPU now leads the Steam hardware survey, overtaking the desktop RTX 3060 as rising memory costs erode the appeal of premium DIY builds.

    The throughline: scarcity is the defining commercial force across AI accelerators, leading-edge fabs, and consumer memory — and it is actively redirecting capital and purchasing behaviour.

    This episode includes AI-generated content.
    続きを読む 一部表示
    5 分
  • Memory Prices Triple, AMD Gains Ground & Samsung's 2nm Foundry Gamble

    Memory Prices Triple, AMD Gains Ground & Samsung's 2nm Foundry Gamble
    2026/07/04
    (00:00:00) Memory Prices Triple, AMD Gains Ground & Samsung's 2nm Foundry Gamble
    (00:00:57) TSMC Saturation Drives Chip Diversification
    (00:01:45) Memory Crisis Inflating AI Infrastructure Costs
    (00:02:32) AMD Narrows Intel CPU Gap
    (00:03:03) Microsoft Frontier Company Enterprise Race
    (00:03:34) What To Watch Next

    The AI hardware industry is facing a converging set of pressures that are reshaping costs, supply chains, and competitive dynamics across the stack. In today's episode, we unpack six major developments that every investor and engineer tracking the semiconductor space needs to understand.

    The headline story is Anthropic's early-stage partnership with Samsung on a two-nanometer custom chip — a move that signals real strain on TSMC's capacity. TSMC remains the backbone of AI silicon, but when a major lab formally initiates a design program with a less-proven foundry just to escape the waitlist, it marks a threshold moment. Samsung's 2nm yield rates and the South Korean government's $518 billion chip factory plan are now central to the foundry competition story.

    Running in parallel is a memory crisis that is already producing visible price action. AMD has raised GPU and GDDR kit prices by 10% effective July 2026 — the second hike in six months. GDDR6 spot prices have tripled from $2.50 to $7.50 per gigabyte since autumn 2025. Micron and SanDisk have roughly tripled in market value. The infrastructure cost story is no longer hypothetical.

    On the processor side, AMD has reached 45.99% of the Steam hardware survey, gaining nearly 14 percentage points since January. Intel holds 54% but the gap is narrowing at the fastest pace in 18 months, driven by AMD's X3D cache architecture.

    Finally, Microsoft has launched Frontier Company — a $2.5 billion unit with 6,000 engineers embedding directly inside enterprise clients alongside Accenture, PwC, and KPMG — a direct challenge to AWS for production AI deployment.

    Two metrics to watch: Samsung's 2nm yield data and whether GDDR supply eases in H2 2026.

    This episode includes AI-generated content.
    続きを読む 一部表示
    4 分
  • Anthropic's 2nm ASIC, Intel's Price Hike & Meta's GPU Rental Play

    Anthropic's 2nm ASIC, Intel's Price Hike & Meta's GPU Rental Play
    2026/07/03
    (00:00:00) Anthropic's 2nm ASIC, Intel's Price Hike & Meta's GPU Rental Play
    (00:00:53) Inference ASIC Race Accelerates
    (00:02:06) Nvidia Export Control Credibility Erodes
    (00:02:43) Intel Platform Economics vs AMD AM5
    (00:03:33) Meta GPU Monetization Strategy

    The inference ASIC race has moved from speculation to confirmed action. Anthropic is in active talks with Samsung to build a custom two-nanometer inference accelerator on Samsung's SF2P process — a move triggered in part by OpenAI's Jalapeño chip unveiling in late June. With custom silicon delivering roughly 50% cost savings over general-purpose GPUs for language model serving, the unit economics are too significant to ignore. Samsung, meanwhile, freed engineering capacity and 2nm yield learnings after cancelling an OpenAI chip project in June, making it a credible — if unproven — foundry partner for Anthropic.

    Nvidia's policy standing is under pressure after the company argued to the Trump administration that Huawei could satisfy global AI chip demand if export controls held. Policy experts pushed back hard: Huawei holds around 6% of global AI chip sales and trails Nvidia significantly on performance. The credibility of Nvidia's semiconductor policy analysis on Capitol Hill is now openly questioned.

    On the consumer CPU front, Intel raised prices on Arrow Lake Refresh parts by 15–17% shortly after launch — on a socket with no upgrade path. AMD's AM5 platform offers a multi-generation roadmap at the same price tier, and AMD has already surged to nearly 46% CPU share on Steam, up 14 points since January, driven by X3D cache dominance in gaming.

    Finally, Meta is monetising its H100 and B200 GPU clusters by offering compute rentals to external customers, putting it in direct competition with AWS, Azure, and Google Cloud. The risk: a pricing war with hyperscalers that Meta may not be able to win.

    A YesWee production.

    This episode includes AI-generated content.
    続きを読む 一部表示
    5 分
  • Intel 90% Yield, Meta's GPU Cloud & Nvidia China Curbs

    Intel 90% Yield, Meta's GPU Cloud & Nvidia China Curbs
    2026/07/02
    (00:00:00) Intel 90% Yield, Meta's GPU Cloud & Nvidia China Curbs
    (00:01:27) Intel 18A-P Yield Beats Forecast
    (00:02:32) Meta's GPU Cloud Enters the Market
    (00:03:29) ASML Raises Guidance on AI Capex
    (00:04:14) US Export Controls Tighten on Nvidia China Sales
    (00:04:54) Intel Santa Clara Mask Hub — The Hidden Constraint
    (00:05:32) Key Watchpoints Through Q3

    Intel's foundry turnaround gets its clearest proof point yet: the 18A-P process variant entered risk production on June 16th with first-month yields tracking at 90% — two quarters ahead of internal forecasts — delivering a 9% speed improvement or 18% power reduction over the standard 18A node. It's the credible technical milestone Intel's foundry narrative desperately needed, even if volume production and confirmed customer contracts are still months away.

    Meta is moving aggressively into the AI cloud market, deploying over 600,000 H100-equivalent GPUs into a dedicated cloud service targeting private beta in August 2026 and general availability in November. Because the capital is already sunk in Meta's own AI infrastructure, the company can price against Azure and AWS in ways pure-play cloud providers cannot match without compressing margins. Enterprise compliance depth remains the open question.

    Samsung's 1.4nm delay to 2029 puts the company roughly one year behind TSMC's A14 node. The date matters less than what a foundry customer can do in that gap — lock in yield baselines, sign exclusives, and embed process choices into two-generation chip architectures. Samsung's counter-bet is on yield quality and DTCO rather than timing.

    ASML raised revenue guidance on the back of AI capex demand, with the stock up over 23% in a month, while new US export restrictions on Nvidia's most capable AI accelerators into China prompted Beijing to respond with a sweeping outbound investment and service-export framework effective July 1st. Intel also broke ground on a 107,000-square-foot mask manufacturing hub in Santa Clara, shoring up a quiet but critical constraint in domestic foundry scaling.

    All of this sets up a loaded Q3 watchlist: Intel's July 23rd earnings, Meta's August beta, and where TSMC's A14 customer roster stands by year-end.

    This episode includes AI-generated content.
    続きを読む 一部表示
    7 分