• SEP 27, 2026 — The industry keeps skipping the step in between

  Crusoe scraps its $1.25B jet-turbine deal with Boom Supersonic

  Crusoe walked away from 29 gas turbines it ordered as a fast fix for AI power demand, also stepping back from a planned gigawatt-scale Wyoming campus in favor of smaller, more flexible sites.

  Unite.AI

• SEP 27, 2026 — The industry keeps skipping the step in between

  NANO Nuclear, IP3, and Cybernetic Intelligence team up on nuclear-powered AI compute

  The non-binding framework pairs NANO's reactor technology with IP3's project financing and Cybernetic Intelligence's compute-sovereignty push, the latest entrant in AI's crowded rush toward dedicated nuclear power.

  GlobeNewswire

• SEP 26, 2026 — The frontier race outruns its own referees

  Musk plans to double Colossus 2's Nvidia chip count by year-end

  xAI's Memphis-area cluster is set to roughly double in size as Musk races to keep pace with rivals' compute buildouts.

  Bloomberg

• SEP 26, 2026 — The frontier race outruns its own referees

  AI's nuclear power bet is running into cost and permitting reality

  Nuclear startups have raised $4.6B this year on AI demand, but most new reactors are still years from actually switching on.

  Bloomberg

• SEP 25, 2026 — The week agents stopped waiting for permission

  Alibaba unveils Zhenwu V900, its most powerful AI chip yet

  The 216GB accelerator clusters up to 500,000 units and is aimed at training a planned 10-trillion-parameter Qwen model.

  Tom's Hardware

• SEP 25, 2026 — The week agents stopped waiting for permission

  Google will launch AI chips into orbit next week

  A test satellite carrying four Trillium TPUs launches Oct. 1 on a SpaceX Falcon 9 — the first real hardware test of Google's orbital datacenter bet.

  Data Center Dynamics

• SEP 24, 2026 — When the humans stepped out of the loop

  IBM cuts quantum error-correction overhead 63-fold

  A hybrid error-cancellation and detection scheme, validated on a 27-qubit processor, sharply cuts the sampling cost of mitigating quantum errors.

  The Quantum Insider

• SEP 23, 2026 — The frontier just got a price war

  IonQ to install first on-prem QPU inside Nvidia's quantum research center

  An IonQ Superion 256 system will link directly to an Nvidia GB200 NVL72 cluster via NVQLink, pairing quantum and classical AI compute on one rack.

  The Quantum Insider

• SEP 22, 2026 — Safety talk gets loud as the market gets louder

  Qualcomm unveils dual Snapdragon 8 Elite Gen 6 chips built for on-device agents

  The 2nm flagship and Pro variant lean on a beefed-up Hexagon NPU aimed at keeping AI agents running locally instead of in the cloud.

  Android Authority

• SEP 22, 2026 — Safety talk gets loud as the market gets louder

  D-Wave, Rigetti, and Quantinuum finalize $300M in CHIPS Act awards

  Commerce takes a minority equity stake in each firm as a condition of the $100M-apiece grants for domestic quantum hardware R&D.

  HPCwire

• SEP 22, 2026 — Safety talk gets loud as the market gets louder

  TSMC and Samsung scramble for talent as the US fab labor shortage deepens

  TSMC says it's recruiting from a dozen universities to help fill roughly 6,000 open roles at its Arizona fabs.

  CNBC

• SEP 21, 2026 — Trump orders full speed as a slowdown pledge becomes a lawsuit

  $68 billion in new US data centers disrupted by local opposition

  Data Center Watch counted 45 projects blocked or delayed by pushback in one quarter alone, with 843 opposition groups now active across 49 states.

  Bloomberg

• SEP 21, 2026 — Trump orders full speed as a slowdown pledge becomes a lawsuit

  China's CXMT starts mass production of fifth-generation DRAM

  CXMT's G5 platform hits an 11.95nm half-pitch and 50% more dies per wafer, pushing its global DRAM share past 10% amid the ongoing memory shortage.

  TechNode

• SEP 20, 2026 — AI kept going past where it was told to stop

  Google says power, not chip supply, is now the ceiling on scaling TPUs

  Speaking at SEMICON Taiwan, Google described linking ever-larger TPU fleets while conceding it can't deploy hardware as fast as it wants — the bottleneck is electricity, not silicon.

  DigiTimes

• SEP 19, 2026 — The week AI stopped staying inside the sandbox

  Jensen Huang: Nvidia will sell twice as many chips next year

  Speaking at a UK AI summit alongside King Charles III, Huang pushed back on slowdown talk with a bold unit-volume forecast, though Nvidia doesn't disclose the base number being doubled.

  CNBC

• SEP 19, 2026 — The week AI stopped staying inside the sandbox

  SK Hynix in talks to make memory chips on US soil with Intel

  Options on the table include leasing part of Intel's stalled Ohio fab or a joint venture with cloud hyperscalers to secure domestic HBM and DRAM supply.

  Quartz

• SEP 19, 2026 — The week AI stopped staying inside the sandbox

  GlobalFoundries and Marvell expand their bet on optical chips for AI data centers

  The pair are adding silicon-germanium manufacturing capacity in Vermont as copper interconnects run out of headroom for AI datacenter bandwidth.

  GlobeNewswire

• SEP 18, 2026 — Claude is now building its own successor

  Axelera launches Europa, a data-center inference chip built on 16 RISC-V cores

  The European chip startup's second-generation AIPU claims 629 TOPS at 45 watts, with a sales pipeline north of $1.5 billion and Dell and Supermicro as launch partners.

  Axelera AI

• SEP 18, 2026 — Claude is now building its own successor

  Nebius hikes AI cloud prices again — its second increase in three months

  H100 rentals jump to $4.50 an hour and B300s to $9.50, with Nebius and CoreWeave both citing GPU demand that's outrunning supply.

  Silicon UK

• SEP 17, 2026 — Amodei's pacing plea splits the industry wide open

  Google and Nvidia launch a coalition to make data centers flex around the grid

  Twenty companies, including Anthropic and utilities National Grid and AES, will coordinate data-center demand to ease strain during peak load.

  Axios

• SEP 16, 2026 — The industry found the cooperation Congress couldn't

  Cornelis Networks raises $205M to challenge Nvidia's InfiniBand

  Its new Active Compute Fabric processes data mid-transit instead of just moving it, aimed at GPU clusters that waste cycles waiting on data.

  SiliconANGLE

• SEP 16, 2026 — The industry found the cooperation Congress couldn't

  Nvidia's unreleased Rubin NVL72 leaks a 30x efficiency gain over Blackwell

  Independent agentic-coding benchmarks show Rubin hitting far higher AI-factory throughput per megawatt than GB300 on real coding-agent workloads.

  SemiAnalysis

• SEP 15, 2026 — The slowdown call gets an answer from everyone but AI

  Fortaegis raises $50M to sell 'unclonable' silicon security keys

  The Dutch startup's chips generate encryption keys from silicon's own physical quirks instead of storing them, aimed at AI infrastructure, defense, and satellites.

  The Quantum Insider

• SEP 14, 2026 — No one agreed on who sets the rules

  Nvidia expands its AI infrastructure partnership in Australia toward 2GW of capacity by 2027

  Nvidia is teaming with Australian data-center operators to build land, power, and shell capacity for multi-generation AI-factory deployments.

  GlobeNewswire

• SEP 14, 2026 — No one agreed on who sets the rules

  Samsung and SK Hynix's memory inventories fall below 10 days of supply as HBM4 eats DRAM capacity

  HBM4 production consumes roughly triple the wafer capacity of ordinary DRAM; analysts now warn the resulting shortage could be the worst on record, lasting past 2030.

  Tech Times

• SEP 13, 2026 — The industry asks for guardrails it keeps outrunning

  Intel's inference GPU slips toward a 2027 launch

  Customer sampling of Crescent Island starts this quarter, but the commercial launch now looks likely to slide past 2026 with Nvidia still holding roughly 80% of the market.

  Tech Insider

• SEP 12, 2026 — The risk stopped being hypothetical

  TSMC's August revenue jumps 53% as AI chip demand outstrips supply

  Record monthly revenue of $16.3 billion, with advanced-node capacity fully booked on AI server chip orders.

  CNBC

• SEP 12, 2026 — The risk stopped being hypothetical

  Ayar Labs raises another $150M for optical chip-to-chip links, hits $650M this year

  A secondary sale priced the interconnect startup above $5B, up a third since March, as it races to solve GPU-cluster bandwidth limits.

  SiliconANGLE

• SEP 11, 2026 — The rulebook tries to catch the roadmap

  Positron AI raises $875M to bring its memory-heavy inference chip to market

  At a $5B valuation, Positron bets that commodity LPDDR5X memory beats scarce HBM for serving huge models cheaply; its Asimov chip tapes out on TSMC N3P by year-end.

  PR Newswire

• SEP 11, 2026 — The rulebook tries to catch the roadmap

  Microsoft plans to triple data-center capacity to 38GW by 2032

  The expansion, up from about 12GW today, is as much an admission as a plan: compute shortages have already forced Microsoft to turn away cloud and AI business.

  Data Center Dynamics

• SEP 11, 2026 — The rulebook tries to catch the roadmap

  DOJ probes whether Nvidia structured its Groq deal to dodge antitrust review

  The $20B 'non-exclusive license' that also came with Groq's founder and top executives is starting to look, to the Justice Department, a lot like an acquisition.

  Bloomberg

• SEP 8, 2026 — The bill for agent autonomy comes due

  A blacklisted Chinese server maker rebranded and kept buying Nvidia's best chips

  Inspur's Aivres subsidiary shipped $5.6B in tech to Southeast Asia, including $3B of Nvidia Blackwell servers later used by ByteDance and Alibaba, per the New York Times.

  Asia Times

• SEP 8, 2026 — The bill for agent autonomy comes due

  Nvidia's custom NVHBM memory promises 30% more bandwidth for AI chips

  Moving the memory controller into HBM's base die frees up die space for compute; Amazon's Annapurna Labs is the first NVLink Fusion partner building with it.

  Tom's Hardware

• SEP 7, 2026 — The AI IPO queue gets its second member

  Wiz finds live attacks chaining an MCP server RCE bug against LiteLLM and agent frameworks

  A 90-day honeypot sweep caught attackers combining a LiteLLM command-injection flaw with an auth bypass for unauthenticated remote code execution.

  Wiz

• SEP 7, 2026 — The AI IPO queue gets its second member

  Nscale, the compute firm behind Anthropic's $45B deal, seeks $3.5B in pre-IPO funding

  About $2B of the raise is said to come from Nvidia, as Nscale races to build out the capacity it already promised to Anthropic.

  TechCrunch

• SEP 6, 2026 — Hiding in plain sight, and nobody looked

  Commerce Secretary confirms chip tariffs are coming: build here, or pay

  Howard Lutnick said a new round of semiconductor tariffs is coming, with an exemption for chipmakers like Nvidia that manufacture inside the United States.

  TipRanks

• SEP 5, 2026 — Smarter than us, harder to watch

  Crusoe triples its valuation to $30 billion in new funding round

  The AI-datacenter builder raised over $3 billion, tripling its valuation in under a year on demand from OpenAI and Meta.

  Bloomberg

• SEP 5, 2026 — Smarter than us, harder to watch

  Gimlet Labs raises $300M to build a chip-agnostic inference cloud

  The Series B, backed by Arm and Microsoft, values Gimlet at $3B for software that splits inference across any silicon.

  SiliconANGLE

• SEP 5, 2026 — Smarter than us, harder to watch

  ChatGPT, Claude, and Grok all went down within hours of each other

  A routing error tied to Azure knocked out three rival chatbots simultaneously Thursday; all recovered by early afternoon.

  Decrypt

• SEP 3, 2026 — Nvidia buys the Hub, everyone else ships

  Dell books a record $60.9B in AI server orders, backlog hits $95B

  Dell's Q2 AI server revenue hit $16.4B and the company raised its full-year guidance by $25B — even as it flags continuing DRAM and NAND supply constraints.

  Slashdot

• SEP 3, 2026 — Nvidia buys the Hub, everyone else ships

  Memory prices keep climbing as AI demand outruns supply

  DRAM contract prices are projected up 13-18% and NAND 10-15% quarter over quarter in Q3, even as consumer affordability starts to buckle.

  Tom's Hardware

• SEP 1, 2026 — A Pentagon platform with a Claude-shaped hole

  Amazon triples its Nvidia GPU order as AWS demand outpaces plans

  Amazon is now ordering Blackwell Ultra, Rubin, and Rubin Ultra chips for 2027-2028 delivery, months after committing to over a million Nvidia GPUs for AWS.

  TechCrunch

• SEP 1, 2026 — A Pentagon platform with a Claude-shaped hole

  Samsung puts a processor inside the memory chip

  Samsung's LPDDR5X-PIM adds compute logic to every memory bank, tripling AI inference throughput on Llama 3.1 with no PCB redesign needed.

  Tom's Hardware

• SEP 1, 2026 — A Pentagon platform with a Claude-shaped hole

  Meta unveils its first in-house AI training chip

  MTIA 300 builds networking directly into the silicon, cutting communication time on large recommendation-model training runs nearly 4x versus a comparable GPU cluster.

  Engineering at Meta

• SEP 1, 2026 — A Pentagon platform with a Claude-shaped hole

  Google quietly pulled a security feature from the Pixel 11

  The Tensor G6's dropped ARM Memory Tagging Extension is a hardware-level cut GrapheneOS says it can't patch around, putting its Pixel 11 support in doubt.

  Privacy Guides

• AUG 31, 2026 — The checks keep getting bigger, the trust keeps thinning

  Samsung hikes foundry prices as AI demand overflows from TSMC

  Samsung raised advanced-node contract prices up to 15% as customers TSMC can't fit spill over onto its 4nm and 5nm lines.

  Tom's Hardware

• AUG 31, 2026 — The checks keep getting bigger, the trust keeps thinning

  TSMC completes its 1.6nm A16 process, eyes Q4 mass production

  The angstrom-class node promises 8-10% more speed at the same power and is aimed at AI and HPC silicon, not phones.

  Seoul Economic Daily

• AUG 30, 2026 — Who's actually in charge of the machine

  Intel's Xeon 7 "Diamond Rapids" slips to 2027, grows to 256 cores

  Intel detailed the delayed server chip at Hot Chips: up to 256 P-cores, 1.28GB of cache, and a jump to 1.6 TB/s memory bandwidth.

  Tom's Hardware

• AUG 29, 2026 — The week AI got serious about worst cases

  Meta lays out a four-chip MTIA roadmap through 2027

  MTIA 300 is already training ranking models; 400, 450, and 500 follow on a six-month cadence, each aimed at getting Meta off commodity GPUs.

  Tom's Hardware

• AUG 29, 2026 — The week AI got serious about worst cases

  Server DDR5 kits are up nearly 5x since Q3 2025 as HBM eats wafer capacity

  Manufacturers are shifting capacity to AI-bound HBM and data-center DDR5, leaving consumer memory structurally short into at least 2027.

  Tech Times

• AUG 28, 2026 — Anthropic beats the Pentagon, then goes hands-on

  Intel details Crescent Island, its inference-focused datacenter GPU

  Unveiled at Hot Chips 2026: 32 Xe3P cores, up to 480GB of LPDDR5X memory, and a 350W design aimed squarely at agentic AI inference rather than training.

  Tom's Hardware

• AUG 28, 2026 — Anthropic beats the Pentagon, then goes hands-on

  AWS and Nvidia to deploy 2 million more GPUs through 2028

  The expanded deal adds Blackwell Ultra, Rubin, and Rubin Ultra GPUs to AWS's infrastructure and includes 100,000 GPUs earmarked for U.S. government AI factories.

  Nvidia Newsroom

• AUG 28, 2026 — Anthropic beats the Pentagon, then goes hands-on

  Micron lags peers despite Nvidia's $279B memory commitment

  Profit-taking hit memory stocks even as Nvidia's disclosed HBM commitments grew, a sign the AI-memory trade is getting crowded and jumpy.

  24/7 Wall St.

• AUG 28, 2026 — Anthropic beats the Pentagon, then goes hands-on

  StarkWare demonstrates a quantum-resistant Bitcoin transaction

  A researcher spent a 10,000-satoshi output on mainnet using hash-based one-time signatures, showing Bitcoin can gain quantum resistance without a fork — for now, at a steep transaction cost.

  The Quantum Insider

• AUG 27, 2026 — Nvidia cashes in as the receipts pile up

  Nvidia lays out its 88-core Vera CPU at Hot Chips

  A monolithic 3nm die, 1.5TB of LPDDR5X, and a design tuned for single-thread IPC over core count — built, Nvidia says, for reinforcement learning and agentic workloads.

  Tom's Hardware

• AUG 26, 2026 — Nvidia's throne wobbles, but doesn't fall

  Cisco expands its Nvidia-based AI factory with Supermicro racks

  Cisco is adding Supermicro's liquid-cooled rack-scale systems to its Nvidia-compliant AI infrastructure architecture for neoclouds and sovereign clouds, available in October.

  Cisco

• AUG 26, 2026 — Nvidia's throne wobbles, but doesn't fall

  TSMC pushes a new packaging format to challenge Samsung

  TSMC is rallying Taiwan's equipment suppliers around a 310x310mm panel-level packaging format, aiming to catch Samsung's early lead in AI chip packaging.

  DigiTimes

• AUG 26, 2026 — Nvidia's throne wobbles, but doesn't fall

  Xiaomi's new phone chip matches Apple's cores, beats them multithreaded

  Xiaomi's Xring O3, with 44MB of cache and an all-high-core design, roughly ties Apple's single-core speed and crushes it on multithreaded benchmarks.

  Notebookcheck

• AUG 25, 2026 — Hugging Face's very bad, very lucrative week

  Intel outlines Diamond Rapids, Crescent Island and Wildcat Lake at Hot Chips 2026

  A 256-core Xeon, a 480GB inference GPU and an edge SoC — Intel's answer to a conference now built around agentic-AI silicon.

  Intel Newsroom

• AUG 22, 2026 — The industry perfects the almost-acquisition

  DDR5 memory prices near 5x their year-ago level in Germany

  AI datacenters and HBM production are absorbing so much wafer capacity that ordinary DRAM keeps spiking — China saw a 14% jump in a single week.

  TrendForce

• AUG 22, 2026 — The industry perfects the almost-acquisition

  AWS makes AgentCore Payments generally available, letting AI agents pay in stablecoin

  Built with Coinbase and Stripe, the service lets Bedrock agents autonomously discover and pay for APIs and content, with spending limits enforced at the infrastructure layer.

  AWS

• AUG 21, 2026 — The week AI agents started acting on their own

  Nvidia denies report it's building a China-specific AI chip for year-end shipment

  The Information reported small-batch shipments of a Groq-derived inference chip were already ordered by Chinese customers; Nvidia says no such product is on its roadmap.

  Reuters

• AUG 21, 2026 — The week AI agents started acting on their own

  QpiAI opens an 8-inch quantum chip foundry in Bengaluru

  Billed as Asia's largest quantum foundry, the facility can fabricate 128-qubit superconducting processors today and is targeting 10,000-qubit chips by 2027.

  The Quantum Insider

• AUG 20, 2026 — The AI boom prices itself

  H200s resume flowing to China, at 13% of the legal cap

  ByteDance and Tencent have each received about 10,000 of a licensed 75,000 H200 units; Beijing's own review process is now the bottleneck, not Washington's.

  Tech Times

• AUG 20, 2026 — The AI boom prices itself

  Samsung raises foundry prices up to 15% as AI orders overflow

  4nm and 5nm wafer prices are climbing as chip demand outgrows even TSMC's capacity, turning around a foundry business that's lost money since 2022.

  Tom's Hardware

• AUG 19, 2026 — OpenAI hits the brakes; the money doesn't

  Nvidia's H200 chips reach China in small batches — but Beijing is discouraging buyers

  ByteDance and Tencent have each taken roughly 10,000 approved H200s, but Chinese officials want firms to hold off in favor of domestic chips, muddying a deal both governments technically cleared.

  Business Standard

• AUG 17, 2026 — The $67 billion weekend

  Agentic AI is quietly reviving demand for server CPUs, not just GPUs

  Tool calls and orchestration generate far more CPU load than a single query, argues a new analysis — the next shortage might not be GPUs.

  IEEE Spectrum

• AUG 17, 2026 — The $67 billion weekend

  Claude services go down worldwide for about 40 minutes

  An authentication issue Sunday night cascaded into degraded performance across Claude.ai, the API, Claude Code, and Claude Cowork before a fix landed.

  BleepingComputer

• AUG 16, 2026 — Two IPOs, one tangled balance sheet

  Six grid operators face a Monday deadline to justify their data-center rules to FERC

  FERC gave the operators 60 days to rewrite or defend interconnection tariffs its preliminary finding called unjust and unreasonable for handling AI-driven load growth; the clock runs out tomorrow.

  Tech Insider

• AUG 15, 2026 — The bill for the AI boom comes due

  Sony and TSMC to invest $6.4B in a joint chip plant in Japan

  The Kumamoto joint venture will mass-produce next-generation image sensors for robots and cars starting as early as 2029.

  Bloomberg

• AUG 15, 2026 — The bill for the AI boom comes due

  South Korea launches a $3.5B fund for chip materials and equipment makers

  The 5 trillion won fund steers government money toward chip suppliers rather than giants like Samsung and SK Hynix.

  Reuters via Yahoo Finance

• AUG 15, 2026 — The bill for the AI boom comes due

  L&T to build India's largest Nvidia AI factory for Together AI

  A roughly ₹15,000-crore order will put 10,000 Nvidia B300 GPUs to work at a gigawatt-scale Chennai campus.

  Business Standard

• AUG 14, 2026 — The AI productivity math still doesn't add up

  AMD's data center revenue more than doubles on Instinct MI350 demand

  Data center revenue hit $6.7B, up 107% year over year, as AMD posts a record $11.5B quarter and projects the segment to double again in 2027.

  CNBC

• AUG 13, 2026 — A billion users, an autonomous hack, and phantom power

  CoreWeave and Lambda both tap leveraged loans to fund GPU buildouts

  CoreWeave closed a $2.6B term loan and Lambda is selling a $917M loan of its own, part of a growing wave of 'neocloud' debt financing for Nvidia GPUs.

  HPCwire

• AUG 13, 2026 — A billion users, an autonomous hack, and phantom power

  Nvidia commits $5B to power-infrastructure firms Lancium and Firmus

  Separate from its $500B Wall Street financing platform, Nvidia is now directly funding the power buildout underneath AI data centers, not just the chips inside them.

  Yahoo Finance

• AUG 12, 2026 — The industry starts building AI's permanent fixtures

  Quantinuum and Oracle will deploy the Helios quantum computer inside Oracle Cloud

  The multi-year deal puts Quantinuum's 98-qubit trapped-ion system inside a US OCI data center for hybrid quantum-AI workloads, with a preview service coming in the following months.

  HPCwire

• AUG 11, 2026 — The tests fail, the money moves, the robots ship

  TSMC's July revenue jumps 45% on unrelenting AI chip demand

  The world's largest contract chipmaker posted roughly $14.5 billion for the month, putting it ahead of its own 40%-growth guidance for the year.

  CNBC

• AUG 10, 2026 — The free model and the billion-dollar backlash

  SK hynix commits $39B to two new fabs to feed AI memory demand

  The board approved 54 trillion won for a Yongin DRAM plant and a Cheongju NAND plant, staged through 2031 to keep pace with AI's appetite for memory.

  The Korea Herald

• AUG 10, 2026 — The free model and the billion-dollar backlash

  AMD calls report of MI455X delays 'BS,' says Helios ships on time

  AMD disputes a report claiming its next AI accelerator slips to 2027, insisting rack-scale Helios systems remain on track for this year.

  Tom's Hardware

• AUG 10, 2026 — The free model and the billion-dollar backlash

  Consumer GPU prices keep climbing as AI eats the memory supply

  An RTX 5090 now lists near $4,900 as memory chips — over 80% of a GPU's bill of materials — get diverted to AI datacenters.

  Tom's Hardware

• AUG 9, 2026 — Nobody's sure what their own AI can do

  IonQ wins expanded DARPA deal to mass-produce optical atomic clocks

  A $28M extension funds 125 Evergreen-05 clocks for government use, with $30M more available through DARPA's It's About Time program.

  The Quantum Insider

• AUG 8, 2026 — AI learns to write genomes, and to outlast a shutdown

  Lumilens emerges from stealth with $700M+ to fix AI datacenter bottlenecks

  The optical-interconnect startup, now valued at $5.5 billion, is already shipping co-packaged optics into hyperscale data centers under a multibillion-dollar customer deal.

  Business Wire

• AUG 7, 2026 — Google's brain drain meets AI's accountability gap

  Commerce Department offers $874M to seven chipmakers for equity stakes

  The CHIPS Act money goes out in exchange for non-controlling equity, with GlobalFoundries eligible for up to $300M toward photonics and AI compute research.

  GV Wire

• AUG 6, 2026 — The scaffolding cracks before the models do

  Anthropic confirms it's building an in-house chip team for Claude

  The lab is hiring silicon engineers at up to $485,000 to co-design custom chips alongside its models, while insisting it isn't dropping Nvidia, AMD, AWS, or Google.

  TechCrunch

• AUG 6, 2026 — The scaffolding cracks before the models do

  Google's $15B India data center runs into water and wildlife opposition

  Activists in Visakhapatnam are suing over the Adani-built campus's proximity to a wildlife sanctuary and its draw on local water supplies.

  BusinessWorld

• AUG 5, 2026 — The week everyone argued about who's driving

  SpaceX's AI compute segment revenue jumps 247% in its first earnings report since going public

  Total revenue rose 92% to $7.8 billion, though capex nearly doubled quarter over quarter and knocked the stock down after hours.

  Fortune

• AUG 4, 2026 — A billion users, a nuclear bet, and a price war

  OLIX raises $312M to build a photonic alternative to Nvidia

  The UK chip startup's Series B, backed by Arm, funds an inference accelerator that swaps HBM for optical interconnects — Britain's biggest semiconductor bet yet.

  Tech Times

• AUG 3, 2026 — AI aces the math, but everything downstream is creaking

  The global memory shortage reaches the MacBook Air

  AI datacenters' appetite for DRAM has pushed MacBook Air ship dates into late August and September, even as Apple raises prices.

  TechCrunch

• AUG 3, 2026 — AI aces the math, but everything downstream is creaking

  Hyperscalers plan ~$5T on off-grid gas power for data centers through 2030

  Enverus projects 62GW of new US data-center capacity by 2030, with grid delays pushing roughly 40% of it onto private, behind-the-meter gas plants.

  The Center Square

• AUG 2, 2026 — Robots got bodies while regulators got teeth

  UN renews pressure on AI firms to run data centers on clean power

  The UN repeated its call for every AI data center to run on renewables by 2030, even as gas and coal keep outpacing clean-energy growth.

  Axios

• AUG 1, 2026 — The build-out gets bigger, the frontier gets cheaper

  Eliyan raises $145M, becomes a unicorn betting on AI chip interconnects

  The Series C, led by Seligman Ventures with Cisco and Lumentum joining in, targets the wiring bottleneck between AI chip clusters, not the chips themselves.

  SiliconANGLE

• AUG 1, 2026 — The build-out gets bigger, the frontier gets cheaper

  China weighs its own export controls on AI models and chips

  Beijing is reportedly considering rules to keep Chinese AI models and chip designs from being acquired or copied by foreign firms.

  The Next Web

• JUL 31, 2026 — The buildout outruns the guardrails

  AMD's Advancing AI 2026: 6th-gen EPYC, Instinct MI400, and the Helios rack

  AMD detailed 256-core 'Venice' EPYC CPUs on TSMC's 2nm node, the Instinct MI400 GPU family, and a rack-scale system aimed squarely at agentic workloads.

  AMD

• JUL 31, 2026 — The buildout outruns the guardrails

  Z.AI switches on a 1-gigawatt data center built entirely on Chinese chips

  The US-blacklisted lab now runs clusters of 10,000-plus domestic chips with zero Nvidia silicon — a milestone for China's chip self-sufficiency push.

  Bloomberg

• JUL 31, 2026 — The buildout outruns the guardrails

  Nvidia turns coding agents loose on its own chip engineering

  Nvidia is using AI agents inside its EDA workflows to speed up the design of the GPUs those same agents ultimately run on.

  The Next Platform

• JUL 30, 2026 — Washington becomes AI's banker and its bouncer

  A fallen power line briefly knocked 3GW of Virginia data centers off the grid

  A single downed transmission line triggered a voltage disturbance felt from DC to Chicago as data centers switched to backup power simultaneously.

  TechCrunch

• JUL 29, 2026 — The week AI got good at finding cracks — including its own

  Taiwan detains an Nvidia employee in a chip-smuggling probe

  Prosecutors allege forged paperwork routed roughly 50 Super Micro AI servers to China — apparently the first time any government has moved against an Nvidia employee personally.

  Bloomberg

• JUL 29, 2026 — The week AI got good at finding cracks — including its own

  AMD locks up to 2.5GW of data-center capacity from Core Scientific in a $14B deal

  A 15-year lease gives AMD reserved capacity across five US sites as Core Scientific completes its pivot from bitcoin mining to AI hosting.

  Tech Times

• JUL 29, 2026 — The week AI got good at finding cracks — including its own

  Google is reportedly baking Gemini's architecture directly into a new chip

  "Frozen v2" would hardwire part of Gemini into silicon for 6-10x more tokens per watt than current TPUs, targeting a 2028 debut amid Google's own capacity crunch.

  Tom's Hardware

• JUL 28, 2026 — The AI trade's assumptions cracked all at once

  CXMT surges 466% in record Shanghai debut, becomes China's priciest listed company

  The DRAM maker's $8.6B IPO — Asia's largest of the year — closed with a market cap near $489B, ahead of ICBC.

  CNBC

• JUL 28, 2026 — The AI trade's assumptions cracked all at once

  Samsung and SK Hynix post their worst rout in decades as Kospi sinks 10.8%

  Samsung fell 13.4% and SK Hynix 14.7% on China's chip progress and Nvidia financing jitters, the Kospi's worst day since March.

  Reuters

• JUL 28, 2026 — The AI trade's assumptions cracked all at once

  Meta and BlackRock form a $14B joint venture for a 1GW Texas data center

  BlackRock-managed funds take 80% of the El Paso campus, with Meta as sole tenant when it comes online in 2028.

  Bloomberg

• JUL 28, 2026 — The AI trade's assumptions cracked all at once

  America's biggest grid operator warns data centers could face involuntary outages

  PJM plans an emergency auction for up to 15GW of new generation after a nearly 7GW shortfall, with data centers picking up the cost.

  Bloomberg

• JUL 26, 2026 — Open weights become a loyalty test

  The global chip race has split into four incompatible national strategies

  Even after $265B in US fab investment, advanced packaging capacity still lags Taiwan's — a gap the analysis says no single country is close to closing.

  Tech Times

• JUL 26, 2026 — Open weights become a loyalty test

  Samsung to break ground on a new HBM packaging plant in October

  The $884M Onyang expansion targets high-bandwidth memory mass production by 2029, chasing AI-driven demand that's already reshaping DRAM pricing.

  DigiTimes

• JUL 25, 2026 — Bug-hunting becomes an AI product category

  AMD and Cerebras pair Helios racks with the Wafer-Scale Engine for 5x faster inference

  The disaggregated inference system, aimed at coding copilots and live agents, is set to reach Cerebras Cloud in the second half of the year.

  Tom's Hardware

• JUL 25, 2026 — Bug-hunting becomes an AI product category

  Nvidia and 44 Japanese firms plan a Rubin-powered national AI factory for 2028

  SoftBank, Sony, and Honda's Noetra consortium, backed by $6.2B in government funding, will run 27,500 Rubin GPUs dedicated to physical AI in Osaka.

  Nikkei Asia

• JUL 24, 2026 — AI's capex race outruns its safety margins

  Nvidia signs $1.5B chip-packaging deal with Amkor to expand US capacity

  A prepayment funds expanded Arizona packaging and test lines for Nvidia's AI chips; Amkor shares jumped 17% on the news.

  Bloomberg

• JUL 24, 2026 — AI's capex race outruns its safety margins

  Nvidia GPUs are headed to the moon, aboard Lunar Outpost's next rover

  Jetson chips will run the rover's lidar and compress sensor data for downlink — likely the first GPU to operate on the lunar surface.

  TechCrunch

• JUL 24, 2026 — AI's capex race outruns its safety margins

  US data centers on track to eat a fifth of the country's electricity by 2035

  BloombergNEF nearly doubled its forecast from seven months ago — nearly 200GW of data center capacity, most of it in the US.

  TechCrunch

• JUL 23, 2026 — The week nothing stayed inside its box

  General Compute secures a $400M loan backed by inference chips, not GPUs

  Upper90's loan uses SambaNova SN50 inference ASICs as collateral instead of Nvidia H100s, a first that signals money moving from training toward cheap inference.

  TechCrunch

• JUL 23, 2026 — The week nothing stayed inside its box

  Meta reportedly in talks to lease $10B more compute from Anthropic's side of the deal

  The still-unconfirmed deal would make Meta a fourth major cloud provider, on top of the $45B, three-year compute deal Anthropic already signed with SpaceX.

  The Motley Fool

• JUL 21, 2026 — The sandbox stopped holding

  Google is reportedly developing a chip that bakes Gemini's architecture into silicon

  'Frozen v2' would hardwire Gemini's model structure into hardware for a projected 6-10x efficiency gain over current TPUs, with deployment eyed for 2028.

  SiliconANGLE

• JUL 21, 2026 — The sandbox stopped holding

  Microsoft commits to AMD's Helios rack-scale AI platform on Azure

  Azure becomes the first hyperscaler to deploy Helios racks pairing MI455X accelerators with Epyc Venice CPUs, a real production alternative to Nvidia's rack systems.

  CNBC

• JUL 21, 2026 — The sandbox stopped holding

  Jensen Huang's week in Tokyo previews Nvidia's next supply-chain partners

  Huang spent last week courting Japanese officials and robotics makers as Nvidia deepens ties beyond its usual chip customers.

  DigiTimes

• JUL 20, 2026 — Every gate in AI got tested at once

  Huawei puts its 8,192-chip Atlas 950 SuperPoD on the floor at WAIC

  The physical debut, claiming 6.7x Nvidia's NVL144, is China's clearest proof yet that it can cluster AI compute at scale without US chips.

  Seoul Economic Daily

• JUL 20, 2026 — Every gate in AI got tested at once

  SK Group's chairman warns the AI memory shortage could turn geopolitical

  Chey Tae-won says customers want 60-100% more AI memory in 2027 than this year, and calls today's 'abnormal' DRAM prices an invitation to retaliation.

  The Korea Herald

• JUL 19, 2026 — The week AI got physical

  TSMC adds $100B more to its Arizona buildout after a record quarter

  The pledge funds at least four more 2nm-and-below fabs, pushing total Arizona investment to $265B, after Q2 profit jumped 77% year over year.

  Tom's Hardware

• JUL 19, 2026 — The week AI got physical

  Meta expands its Louisiana data center to 5GW and roughly $50B in spend

  The Richland Parish campus, originally pitched at $10B, now needs ten new gas plants from Entergy Louisiana just to power it.

  CNBC

• JUL 18, 2026 — The AI money story splits in two

  Nvidia ships a 'trivial' number of H200 chips to China under new licenses

  US officials confirmed limited, licensed H200 shipments to select Chinese buyers even as Nvidia trims its approved Asian buyer list to stay compliant.

  Bloomberg

• JUL 17, 2026 — The week everyone claimed to be open

  Japan commits $6.2B to a national, Nvidia-built 'physical AI' factory

  The Noetra consortium — SoftBank, Sony, NEC, and Honda — will train open robotics foundation models on 27,500 Rubin GPUs at 140 megawatts, sharing the resulting weights domestically.

  Tom's Hardware

• JUL 17, 2026 — The week everyone claimed to be open

  NetApp acquires DataPelago to embed GPU processing at the storage layer

  DataPelago's Nucleus engine processes AI data where it's stored instead of moving it to separate compute clusters first, claiming up to 80% lower cost and 10x speed.

  Nasdaq

• JUL 17, 2026 — The week everyone claimed to be open

  A heat dome pushes East Coast data centers onto diesel backup power

  A DOE emergency order let grid operator PJM force any facility drawing 50-plus megawatts onto onsite diesel generators within 15 minutes — the agency's third such order to PJM this year.

  Tech Times

• JUL 16, 2026 — The next bet isn't the model anymore

  US official tells Congress H200 shipments to China remain 'trivial' despite a Blackwell loophole

  Chinese firms ordered 2 million H200s against Nvidia's 700K in stock, while a Malaysia/Singapore/UAE subsidiary loophole let Blackwell chips reach China for nearly a year before May's fix.

  Tech Times

• JUL 16, 2026 — The next bet isn't the model anymore

  Blackstone's QTS triples a data-center loan to $3.25B as AI financing floods the loan market

  QTS scrapped a parallel bond sale to upsize the loan instead, with Moody's rating it at the lowest investment-grade rung as financing costs for AI buildouts creep up.

  Bloomberg

• JUL 15, 2026 — The guardrails show up all at once

  SK Hynix CEO warns the memory shortage peaks in 2027 and could last until 2030

  AI-driven HBM demand is crowding out conventional DRAM production, with the crunch expected to outlast the current buildout by years.

  Tom's Hardware

• JUL 15, 2026 — The guardrails show up all at once

  Google ramps up TPU pitch to Nvidia-backed 'neocloud' GPU renters

  Google is courting neoclouds like Nscale with steadier performance and simpler networking claims, as Nvidia counters with incentives to keep them on GPUs.

  DigiTimes

• JUL 15, 2026 — The guardrails show up all at once

  Samsung Foundry reportedly lands Anthropic as a custom AI-chip customer

  Talks that began in early July have advanced to a reported manufacturing agreement on Samsung's 2nm process, though neither side has confirmed terms.

  Cryptonomist

• JUL 14, 2026 — The bill comes due for moving fast

  SK Hynix starts shipping HBM4 to Nvidia ahead of schedule

  12-layer HBM4 shipments began at the end of June, with output ramping as memory packaging becomes as much a bottleneck as GPU supply.

  DigiTimes

• JUL 13, 2026 — Nobody's checking anybody's homework

  Cloudflare opens waitlist for a gateway that lets publishers bill AI crawlers directly

  Monetization Gateway builds on the x402 payment standard, giving sites per-route pricing for bot traffic instead of a blanket block-or-allow toggle.

  Tech Times

• JUL 12, 2026 — The week scale ran into friction

  RTX 5090 street prices top $4,300 as AI datacenters absorb the world's GDDR7 supply

  VRAM now makes up over 80% of some cards' bill of materials as memory makers redirect GDDR7 output toward AI accelerators, more than doubling the flagship GPU's launch price.

  Tech Times

• JUL 12, 2026 — The week scale ran into friction

  Chipmakers shed over $1 trillion in value on doubts about AI capex

  Intel fell 21% and the broader semiconductor index gave back ground as Wall Street questioned whether record AI infrastructure spending can keep paying for itself.

  Forbes

• JUL 11, 2026 — This week, everybody crossed a line

  Micron pours first concrete on its New York fab, raises US investment target to $250B

  Construction on the Clay, NY megafab is running ahead of schedule as Micron aims for 40% of its DRAM made domestically.

  Micron

• JUL 11, 2026 — This week, everybody crossed a line

  China's CXMT sets subscription date for a $4.3B Shanghai IPO

  China's top DRAM maker, revenue up 719% year over year on the AI memory boom, confirms pricing for the country's biggest listing since 2022.

  Bloomberg

• JUL 11, 2026 — This week, everybody crossed a line

  Gas plants are outracing renewables to power the AI data-center boom

  Natural-gas buildout is outpacing wind and solar construction for new data centers, prompting state-level pushback in New York, Michigan, Oregon, and Minnesota.

  Associated Press

• JUL 10, 2026 — The week the frontier outran its guardrails

  Meta breaks ground on its first Canadian data center, a $13B CAD Alberta site

  The 1GW, closed-loop liquid-cooled facility near Edmonton is Meta's 33rd data center globally and its first outside the US and Europe footprint pattern.

  CTV News

• JUL 10, 2026 — The week the frontier outran its guardrails

  AMD hikes Radeon GPU and memory bundle prices about 10%

  GDDR6 spot prices have roughly tripled as AI datacenter demand for memory crowds out consumer GPU supply, and board partners are passing the cost on.

  TrendForce

• JUL 9, 2026 — Every lab picked a new frontier this week

  Apple expands its Broadcom deal to build billions more US-made chips

  A new multiyear commitment worth over $30B funds more than 15 billion chips at Broadcom's Fort Collins, Colorado plant.

  Apple

• JUL 9, 2026 — Every lab picked a new frontier this week

  SK Hynix prices its $28B Nasdaq IPO after 7x oversubscription

  Demand for the memory maker's US listing ran seven times the available shares, funding new fabs ahead of Friday's trading debut.

  The Motley Fool

• JUL 9, 2026 — Every lab picked a new frontier this week

  Report: power shortages could delay nearly half of planned US AI data centers

  A Gartner-sourced analysis finds a roughly 7GW capacity gap and multi-year waits for transformers threatening planned US AI data center builds.

  Help Net Security

• JUL 9, 2026 — Every lab picked a new frontier this week

  Microsoft-backed Aalo Atomics reaches its first nuclear criticality

  The microreactor startup's test reactor achieved a self-sustaining chain reaction, aiming to pair 50-megawatt reactor pods directly with AI data centers.

  BusinessWire

• JUL 8, 2026 — Three labs just failed their own safety report card

  China's Biren raises $892M to scale up domestic GPU production

  The Hong Kong-listed chipmaker is selling new shares to fund mass production of its next-gen GPU, chasing the gap Nvidia's export ban left at home.

  South China Morning Post

• JUL 8, 2026 — Three labs just failed their own safety report card

  SoftBank launches SB Neo, a US GPU-rental business anchored at a former uranium plant

  The neocloud venture is built around SB Energy's Portsmouth, Ohio campus, part of a plan SoftBank says could reach 10GW and $500B in investment by 2030.

  Winbuzzer

• JUL 8, 2026 — Three labs just failed their own safety report card

  Virginia's first-of-its-kind data center electricity tax takes effect

  The $0.011-per-kWh consumption tax, expected to raise about $600M a year, is the first US state levy tied directly to data centers' power draw.

  Office of the Governor of Virginia

• JUL 8, 2026 — Three labs just failed their own safety report card

  Samsung Foundry reportedly wins the contract to build Meta's next AI chip on its 2nm process

  The roughly $7.3B deal for Meta's third-generation MTIA accelerator reportedly has Samsung involved in chip design, not just fabrication.

  SamMobile

• JUL 8, 2026 — Three labs just failed their own safety report card

  Edge-AI chipmaker Syntiant files for a Nasdaq IPO

  The Intel- and Microsoft-backed maker of ultra-low-power AI chips for earbuds and cars is seeking a roughly $300M listing under ticker SYTN.

  Benzinga

• JUL 7, 2026 — The week AI oversight got teeth

  DeepSeek is reportedly developing its own AI inference chip

  The Chinese lab is a year into talks with foundries and chip-design partners to cut its reliance on Nvidia and Huawei silicon for serving its models.

  U.S. News

• JUL 7, 2026 — The week AI oversight got teeth

  A heatwave is pushing grid operators to force AI data centers onto backup power

  PJM, the largest US grid operator, wants authority to order data centers onto generators within 15 minutes of an emergency signal — a preview of summers ahead.

  Al Jazeera

• JUL 5, 2026 — The bill for agentic AI starts arriving

  Samsung, SK Hynix and Micron sued over alleged 700% DRAM price-fixing

  A federal class action says the trio throttled legacy DDR3/DDR4 output under cover of the AI-driven shift to high-bandwidth memory.

  Tech Times

• JUL 4, 2026 — The record streak nobody wants credit for

  AI factories are creating winners and losers in the power-equipment market

  Next-gen AI data centers moving from 30-40kW racks toward megawatt scale are forcing Schneider Electric and Siemens to redesign products for a market now above $200B a year.

  Bloomberg

• JUL 4, 2026 — The record streak nobody wants credit for

  South Korea unveils a $200B AI and robotics push for its southeast

  Samsung, SK, Hyundai, and Hanwha commit 312 trillion won to AI data centers and humanoid-robot manufacturing in the Yeongnam region, the third leg of a trillion-dollar national plan.

  Seoul Economic Daily

• JUL 3, 2026 — The chip market flinched, the money didn't

  National Grid to invest $1.75B in Joulent to power AI data centers

  National Grid Ventures is taking a 35% stake in Joulent to fund gigawatt-scale gas power for AI data centers, including a 2.67GW Texas facility feeding a Microsoft campus.

  citybiz

• JUL 3, 2026 — The chip market flinched, the money didn't

  Neocloud stocks CoreWeave and Nebius crater on Meta compute-resale fears

  CoreWeave fell 14% and Nebius 17% after a report that Meta — their biggest customer, with $21B and $27B in committed spend — may soon compete with them directly.

  Tech Times

• JUL 3, 2026 — The chip market flinched, the money didn't

  Anthropic is discussing a new custom chip with Samsung

  Early-stage talks would use Samsung's 2nm process to build Anthropic's first custom accelerator, following OpenAI's Broadcom-built inference chip.

  TechCrunch

• JUL 1, 2026 — Loosen the model, tighten the rules

  PJM Capacity Auction for 2028/2029 Delivery Year Opens

  The largest US grid operator opened bidding for 2028/29 capacity with its price collar expired for the first time — data centers are the main driver of the demand growth pushing prices up.

  PJM Inside Lines

• JUL 1, 2026 — Loosen the model, tighten the rules

  Netris raises $15M Series A from a16z to help AI neoclouds go live faster

  Netris automates networking inside GPU-cloud datacenters, claims 800% ARR growth, and just landed a16z partner Guido Appenzeller on its board.

  TechCrunch

• JUL 1, 2026 — Loosen the model, tighten the rules

  Overland AI Is the First Ground Autonomy Company to Prime a Production Contract for the U.S. Military

  Overland AI won a $20M Marine Corps contract to produce its ULTRA autonomous ground vehicle — the first time a ground-autonomy company has primed a US military production deal.

  DefenseScoop

• JUN 30, 2026 — Out of the lab

  Qualcomm acquires Modular to strengthen data center AI push

  Qualcomm is paying roughly $4B in stock for Modular, the software company co-founded by Chris Lattner that lets AI workloads run across CPUs, GPUs, NPUs, and custom ASICs without rewriting for each — a software layer Qualcomm has conspicuously lacked.

  CNBC

• JUN 29, 2026 — Proof of work

  Upscale AI raises $190M at $2B for AI-native networking

  Nvidia and Salesforce Ventures joined the Series A-1 for the Santa Clara networking startup building AI-native switch silicon targeting ultra-low-latency training and inference interconnects at scale.

  SiliconANGLE

• JUN 28, 2026 — The vacuum fills

  Qualcomm said to be circling AI chip startup Tenstorrent in $10B RISC-V power play

  Qualcomm is in early talks to acquire Jim Keller's RISC-V AI chip startup at $8-10B — a 3x jump from Tenstorrent's $3.2B valuation a year ago and a direct challenge to Nvidia's software lock-in.

  The Register

• JUN 27, 2026 — Access by appointment only

  SK Hynix plans $29.4B Nasdaq ADR listing for July 10

  The world's top HBM supplier — 60% market share, stock up 850% in 12 months, market cap above $1T — is issuing Nasdaq ADRs at $166 a share; HSBC applied a 20% premium in its first target price.

  CNBC

• JUN 23, 2026 — The receipts came in

  TSMC's first CoPoS packaging tools enter validation

  TSMC's glass-panel successor to CoWoS — built to stop AI GPUs from outgrowing round wafers — hits its first demo-tool validation milestone, with mass production still aimed at 2H28.

  DigiTimes

• JUN 22, 2026 — The week the U.S. took the keys

  FERC fast-tracks data centre grid connections

  Six regional grid operators have 60 days to defend or rewrite the rules for connecting AI-scale loads to the transmission system; bring-your-own-power and demand-response provisions land hard on hyperscalers.

  The Next Web

• JUN 21, 2026 — The grip keeps slipping

  AMD and Rackspace sign 30MW AI compute deal for regulated industries

  The phased deployment of Instinct GPUs and EPYC CPUs targets healthcare and other regulated workloads; Rackspace stock jumped 22% on the news.

  AMD

• JUN 21, 2026 — The grip keeps slipping

  Vultr picks HPE and NVIDIA for its next AI datacenter buildout

  The world's largest privately held hyperscaler is standardizing on NVIDIA GB300 NVL72 racks and Spectrum-X networking; terms and timeline stayed unannounced.

  Business Wire

• JUN 20, 2026 — The control layer underneath the hype

  FERC orders grid operators to fast-track AI datacenters

  FERC issued unanimous show-cause orders to all six US grid operators demanding faster interconnection studies for AI datacenters and other large power loads.

  Engineering News-Record

• JUN 20, 2026 — The control layer underneath the hype

  NVIDIA closes its largest-ever bond sale at $25B

  NVIDIA's first bond sale since 2021 drew $85B in orders, three times oversubscribed, funding more datacenter buildout with borrowed money instead of cash.

  Tech Times

• JUN 19, 2026 — Who's allowed to decide what AI can do

  Samsung Foundry Picks Up Nvidia, Tesla Orders as TSMC Stays Maxed Out

  Samsung is picking up sub-5nm orders from Nvidia, Tesla, and reportedly Qualcomm as TSMC's capacity stays maxed out — including a $16.5B Tesla deal to fab its AI6 chip.

  TechSpot

• JUN 19, 2026 — Who's allowed to decide what AI can do

  Memory Makers Are Sold Out Through Year-End as HBM4 Shortage Deepens

  SK Hynix and Micron are sold out of memory through year-end; Samsung's HBM4 ramp is being called the industry's 'lifeboat' as AI demand keeps outpacing every memory maker's wafer allocation.

  AI CERTs News

• JUN 18, 2026 — Every layer of the stack creaked this week

  Intel's 18A-P chip process enters risk production on schedule

  The node promises 9% more performance or 18% less power than its predecessor and will power next-gen Xeon 'Diamond Rapids' chips, a rare on-time win for Intel's foundry turnaround.

  TweakTown

• JUN 18, 2026 — Every layer of the stack creaked this week

  SandboxAQ wins a $500M CHIPS Act award for AI materials discovery

  The award funds AI-driven discovery of PFAS-free chemicals, rare-earth-free magnets, and new catalysts for chipmaking — in exchange the government takes a minority equity stake in the company.

  NIST / U.S. Commerce Department

• JUN 18, 2026 — Every layer of the stack creaked this week

  Taiwan weighs making AI chip smuggling to China a criminal offense

  The proposal would extend export curbs beyond blacklisted firms like Huawei to all Chinese customers and let prosecutors treat smuggled Nvidia-based servers as a crime, not just a compliance issue.

  Tom's Hardware

• JUN 17, 2026 — The jailbreak that wasn't

  Hetzner hikes dedicated-vCPU prices up to 176%, citing AI-driven memory shortages

  Effective June 15, Hetzner's CPX and CCX lines jump 113-176% as DRAM and NVMe procurement costs spike — a small hosting provider passing along the same AI-demand squeeze hitting Intel's Xeon lead times.

  Hetzner

• JUN 16, 2026 — The kill switch worked. Now what?

  CPU requirements for AI workloads are multiplying, driving intensifying shortages and price hikes

  Intel is supply-constrained on Xeon with a six-month lead time; as agentic AI shifts CPU-to-GPU rack ratios from 1:4 toward 1:1, server CPU prices have risen 10–20% since March.

  Tom's Hardware

• JUN 15, 2026 — Fable 5 had a three-day run

  NVIDIA Vera Rubin NVL72 in Full Production; AWS, GCP, Azure, OCI First to Deploy

  The platform delivers 10× more inference throughput per watt at one-tenth the token cost, with initial cloud instances arriving in the second half of 2026.

  NVIDIA

• JUN 15, 2026 — Fable 5 had a three-day run

  Figure AI's BotQ Factory Now Builds One Figure 03 Humanoid per Hour

  A 24× throughput jump achieved in under 120 days, with more than 350 units shipped and line capacity set at 12,000 humanoids per year.

  Figure AI

• JUN 15, 2026 — Fable 5 had a three-day run

  Boston Dynamics Ships First 2026 Atlas Production Run to Hyundai and DeepMind

  The entire 2026 electric Atlas allocation is committed — Hyundai's RMAC and Google DeepMind take delivery this year; new customers come in 2027.

  AI2Work

• JUN 14, 2026 — Ten thousand bugs, one Gemini phishing kit

  SK Telecom and NVIDIA plan gigawatt-scale AI cloud for South Korea

  SK Telecom will build a gigawatt-class AI factory on NVIDIA's DSX platform, with the first facility expected online in 2027 to support Korea's sovereign and enterprise AI services.

  NVIDIA Newsroom

• JUN 14, 2026 — Ten thousand bugs, one Gemini phishing kit

  Server CPUs are now the scarcest chip in AI infrastructure

  Agentic inference workloads are driving Intel to shift Xeon production away from consumer chips; server CPU prices are up 20% since March, with delivery times stretching to 8–12 weeks.

  Tom's Hardware

• JUN 14, 2026 — Ten thousand bugs, one Gemini phishing kit

  Meta plans $115–135B in AI capital spending in 2026

  Meta announced nearly double its prior-year capex commitment as Zuckerberg's superintelligence bet demands infrastructure at a scale that makes earlier hyperscaler spending look modest.

  CNBC

• JUN 13, 2026 — Open weights, open markets

  NVIDIA Kicks Off the Next Generation of AI With Rubin — Six New Chips, One Incredible AI Supercomputer

  Vera Rubin NVL72 is now in full production, with H2 2026 partner availability; AWS, Google Cloud, Microsoft, and Oracle are first in line for cloud deployments.

  NVIDIA

• JUN 13, 2026 — Open weights, open markets

  Amazon CEO Andy Jassy on the rapid growth of Amazon's chips business

  Amazon's custom silicon (Graviton, Trainium, Nitro) crossed a $20B annual revenue run rate with triple-digit YoY growth, backed by $225B in Trainium commitments from Anthropic and OpenAI.

  Amazon

• JUN 12, 2026 — The labs are going public

  NEURA Robotics announces record Series C of up to $1.4 billion

  The German humanoid robotics startup raised up to $1.4B from NVIDIA, Amazon, Tether, and Qualcomm to scale its Physical AI platform to multi-million robots by 2030.

  CNBC

• JUN 12, 2026 — The labs are going public

  Nuclear and natural gas are teaming up to power the AI data center boom

  Blue Energy and GE Vernova are building a 2.5 GW hybrid facility in Texas to feed AI datacenter demand, with gas coming online by 2030 and nuclear by 2032.

  OilPrice

• JUN 11, 2026 — The labs race to Wall Street

  NVIDIA Kicks Off Next Generation of AI With Rubin

  Vera Rubin NVL72—72 GPUs, 3.6 EFLOPS, 1/10th the token cost per watt vs. prior generation—enters partner deployments at AWS, Google Cloud, Microsoft Azure, and OCI in H2 2026.

  NVIDIA

• JUN 11, 2026 — The labs race to Wall Street

  Nvidia picks Unitree for humanoid robot platform as Chinese startup eyes IPO

  NVIDIA designated Chinese startup Unitree as the reference hardware partner for its humanoid robotics developer platform as the company eyes a public listing.

  CNBC

• JUN 11, 2026 — The labs race to Wall Street

  Meta Builds AI Infrastructure With NVIDIA

  Meta signed a multiyear deal to deploy millions of Blackwell and Rubin GPUs across its data centers, backed by a $115–135B 2026 capex budget covering compute, networking, and full-stack codesign.

  NVIDIA

• JUN 10, 2026 — Fable 5, an IPO filing, and Musk's cloud

  NVIDIA Vera Rubin NVL72 enters full production as hyperscalers queue up

  NVIDIA confirmed the Vera Rubin NVL72 rack-scale system is now in production, with AWS, Google Cloud, Microsoft Azure, and Oracle among the first cloud providers to receive shipments in H2 2026.

  NVIDIA Newsroom

• JUN 9, 2026 — Off the benchmark, into production

  NVIDIA Rubin GPU platform now in production; hyperscalers first in line

  AWS, Google Cloud, Azure, and OCI are first to deploy Vera Rubin NVL72 rack systems, which NVIDIA claims deliver 10x inference throughput per watt versus the prior generation.

  NVIDIA

• JUN 9, 2026 — Off the benchmark, into production

  HBM memory emerges as AI infrastructure's next hard bottleneck

  High-bandwidth memory may account for roughly 30% of hyperscaler AI spending in 2026 — up from 8% in 2023 — as the industry runs into a 'GPU Wall' beyond raw accelerator supply.

  Data Center Knowledge

• JUN 9, 2026 — Off the benchmark, into production

  AMD data-center GPU revenue forecast to jump 114% year-over-year to $15B

  The MI400 series is driving AMD's surge as inference-heavy AI workloads scale; analysts project shipments of roughly 258,000 units in 2026 at an average selling price above $30,000.

  S&P Global

• JUN 9, 2026 — Off the benchmark, into production

  NVIDIA and Meta announce multiyear AI infrastructure partnership

  Meta will deploy millions of Blackwell and Rubin GPUs across hyperscale data centers optimized for both training and inference under the multigenerational deal.

  NVIDIA

• JUN 8, 2026 — The recursion arrives, and Apple waves it in

  SoftBank to Invest Up to €75 Billion to Build French Data Centers

  SoftBank committed to 5 gigawatts of AI data center capacity in France by 2031 — the largest single foreign direct investment in French history.

  TechCrunch

• JUN 8, 2026 — The recursion arrives, and Apple waves it in

  Anthropic Will Pay xAI $1.25 Billion Per Month for Compute

  Disclosed in SpaceX's S-1, Anthropic is paying its Grok-rival xAI $1.25B a month through 2029 for GPU capacity — Google separately signed a $920M/month deal with the same provider.

  TechCrunch

• JUN 8, 2026 — The recursion arrives, and Apple waves it in

  NVIDIA Releases Open Source Agent Skills for Physical AI at CVPR 2026

  At CVPR in Denver, NVIDIA released a collection of open-source tools that turn robotics, AV, and industrial digital-twin workflows into agent-executable task primitives.

  NVIDIA Newsroom

• JUN 7, 2026 — The week AI went public — and personal

  NVIDIA Vera Rubin NVL72: 72 GPUs, 260 TB/s Bandwidth, Now in Full Production

  NVIDIA's Rubin-based NVL72 delivers 10x inference throughput per watt versus Blackwell and will reach AWS, Google Cloud, Microsoft, and CoreWeave customers in H2 2026.

  NVIDIA Newsroom

• JUN 7, 2026 — The week AI went public — and personal

  Amazon's Custom Chip Business Hits $20B Annual Run Rate; Jassy Eyes External Sales

  Amazon's Trainium and Inferentia silicon portfolio crossed $20 billion in annualized revenue, and CEO Andy Jassy suggested expanding to third-party sales with a projected $50B ceiling.

  The Next Web

• JUN 6, 2026 — Open weights, open questions

  NVIDIA Vera CPU: 88 Cores, 3.6 TB/s On-Chip Fabric, Built for Agents

  NVIDIA's Arm-based Vera CPU pairs with Rubin GPUs and packs 1.2 TB/s LPDDR5X bandwidth with no chiplet boundaries; Jensen Huang called it 'a CPU for agents.'

  NVIDIA Newsroom

• JUN 6, 2026 — Open weights, open questions

  After the Power Crunch, AI Infrastructure Hits a GPU Wall

  GPU cloud rental pricing is softening as Rubin-class capacity arrives; hyperscalers are shifting competition from accelerator scarcity to full-stack control of networking and cooling.

  Data Center Knowledge

• JUN 5, 2026 — The week Washington found its AI voice

  The Big Five are spending $650B+ on AI infrastructure in 2026

  Microsoft, Amazon, Alphabet, Meta, and Oracle are on pace to collectively spend $650–700B on AI datacenters this year, straining memory, wafer, and accelerator supply chains simultaneously.

  Bloomberg

• JUN 5, 2026 — The week Washington found its AI voice

  AMD MI400 on track to ship 258,000 units this year

  Analysts estimate AMD will ship roughly 258,000 MI400 accelerators in 2026 at an average $31K each — the most credible competitive challenge to NVIDIA's datacenter dominance since the AI build-out began.

  S&P Global

• JUN 4, 2026 — The bill for AI arrives

  NVIDIA Vera Rubin NVL72 enters full production; AWS, Google, Microsoft deploying H2 2026

  The Vera Rubin platform delivers 10x more inference throughput per watt than Blackwell and will reach major cloud providers — and CoreWeave, Lambda, Nebius — in the second half of this year.

  NVIDIA Newsroom

• JUN 4, 2026 — The bill for AI arrives

  NVIDIA RTX Spark Superchip: Arm CPU, Blackwell GPU, 128GB unified memory for consumer laptops

  Announced at Computex, RTX Spark merges Grace CPU and Blackwell GPU via NVLink-C2C into a 1-petaflop, 128GB unified memory chip for Windows laptops shipping fall 2026.

  Tom's Hardware

• JUN 4, 2026 — The bill for AI arrives

  Intel launches 1.8nm Xeon 6+ and previews Crescent Island datacenter GPU at Computex 2026

  Intel's Clearwater Forest Xeon 6+ hits 1.8nm and the company previewed its upcoming Crescent Island AI datacenter GPU as it pushes deeper into the AI infrastructure stack.

  KAD

• JUN 4, 2026 — The bill for AI arrives

  Apollo and Blackstone line up $36 billion chip-debt deal for Anthropic's Google TPUs

  The two firms are syndicating a $36B private credit package — backed by Broadcom — to buy Google TPUs and lease them to Anthropic, likely the largest chip-financing debt transaction in history.

  Private Equity Wire

• JUN 3, 2026 — Washington blinks first at frontier AI

  After the power crunch, AI infrastructure hits a GPU wall

  GPU rental pricing is showing early signs of cooling as Rubin-era supply arrives — possibly the first meaningful signal that compute scarcity may have peaked.

  Data Center Knowledge

• JUN 2, 2026 — AI bills come due

  Meta commits to 'millions' of NVIDIA Blackwell and Rubin GPUs in multiyear deal

  Meta's multiyear, multigenerational deal covers CPUs, networking, and GPU supply alongside a $115B–$135B capex forecast for 2026.

  NVIDIA Newsroom

• JUN 2, 2026 — AI bills come due

  Intel launches 1.8nm Xeon 6+ and AI infrastructure stack at Computex 2026

  Xeon 6+ (Clearwater Forest) is the first datacenter CPU on Intel's 18A node; Intel also previewed the Crescent Island datacenter GPU.

  KAD

• JUN 1, 2026 — The meter drops on AI coding

  NVIDIA Kicks Off the Next Generation of AI With Rubin

  Vera Rubin is in full production; the platform promises 4× fewer GPUs per MoE training run than Blackwell, with cloud deployments beginning H2 2026.

  NVIDIA Newsroom

• JUN 1, 2026 — The meter drops on AI coding

  NVIDIA Unveils Rubin CPX: A New Class of GPU for Massive-Context Inference

  A new Rubin sub-variant targets inference on extremely long contexts, addressing a gap as 1M-token-plus windows go mainstream.

  NVIDIA Newsroom

• MAY 31, 2026 — The week the agent became the story

  NVIDIA Vera Rubin enters full production with AWS, Google Cloud, and Azure among first deployers

  Rubin's 10x inference token cost reduction versus Blackwell starts ramping into hyperscaler data centers in the second half of 2026, with Nebius and CoreWeave also confirmed.

  NVIDIA

• MAY 31, 2026 — The week the agent became the story

  xAI's Colossus 2 is the world's first 1 GW AI data center, housing 500K+ NVIDIA GPUs

  Memphis now hosts the world's largest single AI training cluster after Colossus 1 was leased to Anthropic; Grok models are training on the new site.

  SemiAnalysis

• MAY 31, 2026 — The week the agent became the story

  Meta and NVIDIA announce multiyear partnership spanning millions of Blackwell and Rubin GPUs

  The strategic deal covers on-premises and cloud GPU deployment at scale, cementing NVIDIA's position as the default infrastructure layer for the largest social network on Earth.

  NVIDIA

• MAY 30, 2026 — The labs are buying the scaffolding

  NVIDIA Vera Rubin enters full production

  Rubin-based systems, offering 5x inference and 3.5x training gains over Blackwell, start shipping from cloud partners in the second half of 2026.

  NVIDIA Newsroom

• MAY 30, 2026 — The labs are buying the scaffolding

  AMD MI400 on track for ~258,000 units this year

  S&P Global forecasts AMD ships a quarter-million MI400 series GPUs in 2026, the most credible chip-level challenge to Nvidia's datacenter dominance yet.

  S&P Global

• MAY 30, 2026 — The labs are buying the scaffolding

  Big tech's combined AI capex: $650B+ in 2026

  Microsoft, Amazon, Alphabet, and Meta are collectively spending over $650 billion on AI infrastructure this year; global datacenter capex is forecast to hit $3–4 trillion by 2030.

  Intellectia AI

• MAY 29, 2026 — Faster models, stubbornly flat gains

  Meta is committing $115–135 billion to AI capital expenditure in 2026

  Meta's 2026 AI spending plan nearly doubles last year's figure, reflecting Zuckerberg's bet that aggressive infrastructure investment is the path to closing the gap with OpenAI and Google.

  PYMNTS

• MAY 29, 2026 — Faster models, stubbornly flat gains

  After the power crunch, AI infrastructure hits a GPU wall

  Semiconductor supply expansion is lagging behind record AI infrastructure spending — silicon is now the binding short-term constraint, with power remaining the long-term bottleneck.

  Data Center Knowledge

• MAY 28, 2026 — The battle for AI's implementation layer

  Google unveils eighth-generation TPUs — two dedicated training and inference chips

  Google's TPU 8t delivers 121 FP4 exaflops in a 9,600-chip superpod — a 3× leap over the previous generation — while the TPU 8i is purpose-built for high-speed inference and agent serving.

  Data Center Dynamics

• MAY 28, 2026 — The battle for AI's implementation layer

  NVIDIA kicks off the next generation of AI with Rubin

  Rubin-based GPU products are now in full production, with first deliveries to AWS, Google Cloud, Microsoft, and Oracle expected in H2 2026, extending NVIDIA's datacenter dominance into the next generation.

  NVIDIA Newsroom

• MAY 28, 2026 — The battle for AI's implementation layer

  AI data centers on track to consume 1,050 TWh in 2026

  If data centers were a country, they'd rank fifth globally for energy consumption — and demand is growing at 15% per year, four times faster than all other sectors combined.

  IEA

• MAY 26, 2026 — Capital piles in. The Pope speaks. Agents go ambient.

  NVIDIA Rubin GPU platform enters production

  NVIDIA's next-gen Rubin platform rolls out in H2 2026, with AWS, Google Cloud, Microsoft, and Oracle as the first cloud providers deploying Rubin-based instances.

  NVIDIA Newsroom

• MAY 26, 2026 — Capital piles in. The Pope speaks. Agents go ambient.

  Big Five hyperscalers on track for $700B+ in capex this year

  Microsoft, Amazon, Alphabet, Meta, and Oracle could collectively spend $700B+ on capital expenditure in 2026 — nearly double 2025 levels — with AI data centers accounting for the bulk of it.

  Intellectia AI

• MAY 26, 2026 — Capital piles in. The Pope speaks. Agents go ambient.

  Google TPU 8t delivers 3× compute, ties 960K GPUs across sites

  Google's TPU 8t posts nearly triple the compute performance of its predecessor and, via the Virgo Network, can connect up to 960,000 GPUs across multiple data center sites.

  Google Cloud Blog

• MAY 25, 2026 — Everything's on sale except the company building it

  OpenAI and NVIDIA Announce Strategic Partnership to Deploy 10 Gigawatts of NVIDIA Systems

  OpenAI will deploy at least 10 gigawatts of NVIDIA infrastructure — millions of GPUs — with the first gigawatt of Vera Rubin hardware arriving in H2 2026.

  NVIDIA

• MAY 25, 2026 — Everything's on sale except the company building it

  Innovations from Google I/O '26 on Google Cloud

  Google's eighth-generation TPU 8t delivers nearly 3x higher compute performance than the prior generation, specifically targeting high-throughput training workloads for large models.

  Google Cloud

• MAY 24, 2026 — The proof, the hire, the lecture

  Meta Builds AI Infrastructure With NVIDIA

  Meta and NVIDIA announced a multigenerational partnership deploying millions of Blackwell and Rubin GPUs across on-premises and cloud — the first major hyperscaler to publicly commit to the full Rubin transition.

  NVIDIA

• MAY 23, 2026 — The trillion-dollar week

  OpenAI and NVIDIA Announce Strategic Partnership to Deploy 10 Gigawatts of NVIDIA Systems

  OpenAI and NVIDIA will deploy at least 10 gigawatts of AI data center capacity on the Vera Rubin platform, with NVIDIA investing up to $100 billion progressively as each gigawatt comes online.

  NVIDIA

• MAY 23, 2026 — The trillion-dollar week

  AI infrastructure at Next '26

  Google's eighth-generation TPU ships as two distinct chips for the first time — one for training at nearly 3x prior-gen compute, one optimized for inference — marking the end of the one-chip-fits-all TPU era.

  Google Cloud

• MAY 22, 2026 — Agents everywhere: AI stops waiting to be asked

  NVIDIA Kicks Off the Next Generation of AI With Rubin

  NVIDIA's Rubin GPU platform enters production in H2 2026, with AWS, Google Cloud, Azure, and OCI first in line for Vera Rubin-based cloud instances.

  NVIDIA

• MAY 22, 2026 — Agents everywhere: AI stops waiting to be asked

  AI infrastructure at Next '26

  Google's eighth-generation TPU 8t delivers nearly 3× the compute performance of its predecessor, with a distinct chip design aimed at high-throughput agentic workloads.

  Google Cloud

• MAY 22, 2026 — Agents everywhere: AI stops waiting to be asked

  AMD's next-generation AI chips set to power 2026 data center growth

  Data-center GPU revenue is forecast to surge 114% year-on-year to $15 billion in 2026, driven largely by AMD's MI400 series expansion.

  S&P Global

• MAY 21, 2026 — The AI compute bill comes due

  Cerebras Raises $5.55 Billion in AI Chip IPO

  Cerebras priced at $185 and closed at $311 on day one — the largest tech IPO since Uber in 2019 — while carrying a notable risk: two UAE entities account for 86% of its revenue.

  Bloomberg

• MAY 21, 2026 — The AI compute bill comes due

  After the Power Crunch, AI Infrastructure Hits a GPU Wall

  Hyperscalers are planning $700B+ in 2026 capex, but semiconductor supply chains cannot expand at the same pace, creating a practical ceiling on deployment timelines.

  Data Center Knowledge

• MAY 20, 2026 — Frontier models cheat, find zero-days, get deployed anyway

  NVIDIA Kicks Off the Next Generation of AI With Rubin — Six New Chips, One Incredible AI Supercomputer

  The Vera Rubin NVL72 rack—72 Rubin GPUs and 36 Vera CPUs on a unified NVLink memory space—is in full production, with AWS, Google Cloud, Microsoft, and OCI deploying instances in H2 2026.

  NVIDIA Newsroom

• MAY 20, 2026 — Frontier models cheat, find zero-days, get deployed anyway

  OpenAI and NVIDIA Announce Strategic Partnership to Deploy 10 Gigawatts of NVIDIA Systems

  OpenAI will deploy at least 10 GW of NVIDIA Vera Rubin infrastructure across multiple years; the first gigawatt arrives this year, enough to power a mid-sized city.

  NVIDIA Newsroom

← All topics