Sector 7 · 19 sub-themes
Compute & Infrastructure
CoreWeave's backlog surging from $104B to $129B in just weeks — while reporting 112% revenue growth to $2.58B — is the clearest single signal of how violently AI compute demand is outpacing supply. Nebius reinforces this: 454% YoY revenue growth, with new capacity contracts now priced at $40–50M per megawatt, up from a $12M baseline earlier this year. The money chasing supply is staggering — Nvidia has corralled Apollo, BlackRock, Blackstone, Brookfield, Goldman, and KKR into financing platforms targeting $500B in AI infrastructure capital, with Jensen Huang backstopping up to $125B personally. Intel raised $20B in equity — its first such raise in decades — after attracting $100B in demand, as Bank of America revised its 2030 server CPU market forecast up to $210B. Applied Materials posted record revenue of $9.12B, up 25% YoY, with Q4 guidance pointing even higher. Meanwhile, supply-side constraints are hardening: Texas grid operators have paused new data center connections as requests hit 474 GW across 1,800+ projects, over 500 local governments have enacted data center bans, and a developer has sued one town for $300M. Tesla and SpaceX are trying to sidestep chip dependency entirely with the $16.8B Terafab in Texas. Microsoft is moving similarly, targeting 1M+ Maia 300 chips for 2027. The infrastructure race is real, constrained, and expensive — and the backlash is only beginning.
CoreWeaveNebiusNvidiaIntelApplied MaterialsTeslaSpaceXMicrosoftAnthropicOpenAIMistral AIJensen HuangAI compute infrastructure demanddata center capacity constraintsGPU cloud revenue growthAI chip manufacturing investmentenergy grid pressure from data centersAI infrastructure financingcustom AI silicon developmentsemiconductor market expansiondata center regulatory backlashKV cache NAND demand
7.1
SanDisk projects KV cache will drive 35% of AI data center NAND demand by 2030
- At its Investor Day, SanDisk projected that persistent KV cache workloads will account for 35% of AI data center NAND demand by 2030, framing flash storage as a major beneficiary of LLM inference infrastructure growth as longer context windows and larger batch sizes make cached inference increasingly attractive. [3][5][7]
- SanDisk's HBF (High Bandwidth Flash) architecture is positioned to deliver HBM-like read bandwidth at NAND economics, with 8–16x more capacity than HBM, targeting memory-intensive AI inference workloads where KV cache is offloaded from HBM to NAND-tier storage. [1][2]
- Goldman Sachs coverage of SanDisk noted that the SanDisk/Kioxia BiCS roadmap remains the core NAND scaling engine, while HBF and KV cache represent upside optionality on top of an already strong core NAND roadmap. [4]
- Analysts at the Future of Memory and Storage conference highlighted persistent KV cache as an emerging major new source of NAND demand, reinforcing SanDisk's positioning of NAND as moving from pure storage into the broader AI memory stack. [6]
SanDiskKioxiaGoldman SachsHBFBiCSHBMNANDKV cachepersistent KV cache storageAI inference infrastructureNAND demand growthhigh bandwidth flashLLM memory stackAI data center storageflash storage economicscontext window scaling
7.2
CoreWeave Q2 2026 Earnings: $2.58B Revenue, $104B Backlog
- CoreWeave reported Q2 2026 revenue of $2.575B, up approximately 112% year-over-year from $1.212B in Q2 2025, beating the consensus estimate of approximately $2.55B; H1 2026 revenue totaled $4.65B, also up 112% YoY. [3][5][7]
- CoreWeave's revenue backlog stood at $104.2B at the end of Q2 2026, growing further to $129.2B by August 11, reflecting sustained contracted demand for AI compute infrastructure from clients including OpenAI, Microsoft, and Meta. [1][3][9]
- CoreWeave guided toward an annualized recurring revenue (ARR) target of $18B–$19B by end of 2026, implying an approximately $10.4B ARR run-rate based on Q2 results. [1]
- Adjusted EBITDA margin hit what analysts described as an "inflection point," with Rosenblatt maintaining a $250 price target and noting CoreWeave delivered upside to both revenues and operating income this quarter. [2][4]
- Despite the strong revenue growth, CoreWeave posted a GAAP operating loss of $49M (–2% margin), a GAAP EPS of –$1.14 (beating the –$1.17 estimate), and negative free cash flow of –$5.7B, reflecting the heavy capital expenditure of hyperscale AI infrastructure buildout. [5][6][8]
- CoreWeave's stock rose approximately 14% following the earnings release despite the deeply negative free cash flow and ongoing net losses. [8][9]
CoreWeaveOpenAIMicrosoftMetaRosenblattAI compute infrastructurecloud GPU revenue growthrevenue backlogannualized recurring revenuehyperscale capital expenditureEBITDA margin inflectionearnings beatfree cash flow burn
7.3
OpenAI previews GPT-5.6 Sol "Ultrafast mode" at up to 14x speed
- OpenAI announced "Ultrafast mode" for GPT-5.6 Sol, claiming speeds up to 14x faster than standard (equivalent to a 1,300% speed increase), with access launching initially in the OpenAI API to a select group of customers in limited preview. [6][7][8][9]
- GPT-5.6 Sol is part of a broader limited-preview release that also includes models named Terra and Luna, with access restricted to limited preview partners per U.S. government arrangements. [3][5]
- OpenAI applied stronger cybersecurity safeguards and restricted early access to GPT-5.6 Sol specifically, with the model reportedly outperforming Claude Opus on early benchmark testing. [1][2][4]
OpenAIGPT-5.6 SolTerraLunaClaude Opuslarge language model releaseultrafast inference modelimited preview accessAPI speed optimizationcybersecurity safeguardsbenchmark performance comparisonU.S. government AI arrangements
7.4
Mistral AI targets 1 gigawatt of European AI compute by 2030
- Mistral AI announced a three-part expansion of its infrastructure business: regional inference endpoints letting customers choose whether workloads run in Europe or the US, a new "Priority Tier" with an uptime guarantee for mission-critical deployments, and a coalition of European enterprises making multi-year compute commitments to underwrite 200 megawatts of infrastructure across Europe by end of 2027, scaling to a full gigawatt by end of 2030. [1][4]
- The company is positioning itself as critical infrastructure rather than a model provider, offering enterprises and governments assured capacity, regional data control, and contractual reliability — a strategic shift described by co-founder and CTO Timothée Lacroix as "strengthening one part of this infrastructure." [1][3]
- In a move noted as potentially controversial among sovereignty advocates, Mistral also announced it will begin hosting third-party open models on its platform, starting with GLM-5.2 from Z.ai, the Chinese AI lab formerly known as Zhipu. [1][3][4]
- The gigawatt infrastructure plan is described as a project valued in the tens of billions of dollars, aimed at building a European alternative to US hyperscaler AI infrastructure. [2]
Mistral AITimothée LacroixGLM-5.2Z.aiZhipuEuropean AI infrastructureregional inference endpointssovereign computePriority Tier uptime guaranteethird-party model hostinghyperscaler alternativemulti-year compute commitmentsdata residency control
7.5
Nvidia Promotes Water-Free Cooling; L&T Wins India AI Factory Deal
- Nvidia has proposed a new data center design it claims largely solves AI's water consumption problem, centered on liquid cooling technology that eliminates the need for evaporative water cooling systems traditionally used in data centers. [1][3][4]
- TechCrunch noted that while Nvidia's new cooling approach reduces water use within data centers themselves, this does not address the broader water consumption associated with AI infrastructure, such as water used in electricity generation upstream. [2]
- Larsen & Toubro (L&T) has secured a major contract to build India's largest single-cluster AI Factory, located in Chennai, with the facility set to be based on Nvidia B300 infrastructure. [5][6]
NvidiaTechCrunchLarsen & ToubroL&TB300ChennaiAI data center coolingliquid cooling technologywater consumption reductionevaporative cooling alternativesAI infrastructure energyAI factory constructionIndia AI development
7.6
Anthropic Signs $9.1B, 20-Year AI Data Center Deal With Riot Platforms
- Anthropic has signed a $9.1 billion, 20-year agreement with Riot Platforms, a Bitcoin mining company that has been expanding into AI data center hosting, to secure AI computing capacity. [1][2][3]
- The deal covers 191 MW of capacity at Riot Platforms' Rockdale, Texas campus, converting the company's existing mining infrastructure into AI compute infrastructure for Anthropic. [3][4]
AnthropicRiot PlatformsRockdaleAI computing infrastructuredata center capacity dealBitcoin mining to AI conversionAnthropic expansioncloud compute hostingTexas AI infrastructure
7.7
Tesla and SpaceX announce $16.8B "Terafab" chip factory in Texas
- Tesla and SpaceX have jointly broken ground on "Terafab," a planned semiconductor fabrication facility in Grimes County, Texas, with an initial investment of $16.8 billion from both companies combined. The facility is planned to span over 100 million square feet, making it one of the largest chip manufacturing complexes ever announced. [1][2][4][5][6]
- The Terafab project is intended to produce AI chips to help meet the internal compute demands of Tesla and SpaceX, representing an unusual cross-company capital deployment into domestic semiconductor manufacturing by two of Elon Musk's companies. [4][6]
- The project has drawn comparisons to large-scale megaprojects in scope, with Blaze Media reporting it is transforming Texas and drawing comparisons to the world's tallest buildings in terms of ambition and scale. [3]
TeslaSpaceXTerafabGrimes CountyTexasElon MuskBlaze Mediasemiconductor fabricationAI chip manufacturingdomestic chip productionTesla SpaceX joint ventureTexas megaprojectcompute infrastructure investmentvertical integration
7.8
Intel raises $20B equity offering amid surging CPU demand forecasts
- Intel raised $20 billion in an upsized equity share sale — increased from an initial $15 billion target — after attracting more than $100 billion in investor demand, marking the company's first such offering since 1971 (or 1988, per varying sourcing). The proceeds are intended to fund new manufacturing facilities and advanced packaging capabilities. [2][3][5][6]
- Bank of America raised its 2030 server CPU market forecast to over $210 billion, up from a prior estimate of $170 billion, representing nearly 5x the estimated 2025 market size, citing AI agent workloads as a key demand driver for server CPUs. [1]
- Bank of America analyst Vivek Arya named AMD the firm's top CPU pick, citing the company's "dual leadership" across both frequency and core count. [4]
IntelBank of AmericaAMDVivek Aryasemiconductor equity offeringserver CPU market forecastAI agent workloadschip manufacturing investmentadvanced packagingCPU market growthdata center infrastructure
7.9
CME and Silicon Data launch AI compute futures; Nvidia debuts new cooling tech
- CME Group and Silicon Data (backed by DRW) are launching the world's first AI compute futures contracts on October 5, based on NVIDIA H100 and B200 GPU rental price indices; each contract is sized at 730 GPU-hours (approximately one GPU-month) and will be cash-settled. Silicon Data explicitly confirmed the partnership with CME Group on the launch. [6][7][8]
- Nvidia unveiled a warm water closed-loop cooling system designed to eliminate on-site data center water use entirely, addressing one of AI infrastructure's major resource challenges. [2][5]
- TechCrunch and CryptoRank note that while Nvidia's new cooling tech removes water consumption at the data center level, it does not address AI's total water footprint across the broader supply chain and power generation, meaning the overall water problem remains partially unsolved. [3][4]
- Nvidia is offering startup customers the option to exchange compute power for a revenue share arrangement, providing an alternative path to GPU access for early-stage AI companies that lack capital. [1]
CME GroupSilicon DataDRWNvidiaTechCrunchCryptoRankH100B200AI compute futures contractsGPU rental price indicescash-settled derivativesdata center water coolingAI infrastructure resource consumptioncompute access for startupsrevenue share for GPU accessAI supply chain water footprint
7.10
Microsoft Maia 300 chip targets September unveil amid TSMC capacity push
- Reuters, citing The Information, reports that Microsoft could unveil its next-generation Maia 300 AI chip as early as September, with the company already eyeing capacity for more than 300,000 Maia 300 chips for 2027 delivery and planning a sharp production ramp; Microsoft is in active talks with TSMC to secure manufacturing capacity, though component constraints and ongoing negotiations could temper those ambitions. [5][6][7][8]
- Microsoft is courting major cloud customers such as Anthropic for the Maia 300, and is reportedly aiming to secure capacity for over 1 million chips, representing an aggressive strategic effort to reduce financial reliance on Nvidia GPU hardware by scaling its in-house custom silicon program. [9][10][5]
- Microsoft's current Maia 200, unveiled in early 2026 and built on TSMC's 3nm process, is an inference-focused chip packing more than 140 billion transistors, 216GB of HBM3e memory, and support for lower-precision FP4 and FP8 computations, with a design emphasis on performance per dollar for large-scale AI workloads. [5]
- Microsoft launched a new AI division called "Frontier Company" staffed with 6,000 engineers and backed by $2.5 billion, designed to embed AI engineers directly inside enterprise customers to accelerate AI adoption and deployment at scale. [1][2][3][4]
MicrosoftMaia 300Maia 200TSMCNvidiaAnthropicFrontier CompanyReutersThe Informationcustom AI chip developmentin-house silicon strategycloud infrastructure investmentreducing Nvidia dependencyenterprise AI adoptionAI chip manufacturing capacityHBM3e memoryTSMC manufacturing partnership
7.11
Nvidia Partners Six Financial Giants to Raise $500B for AI Infrastructure
- Nvidia signed memorandums of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to establish independent financing platforms targeting more than $500 billion in third-party capital for AI infrastructure; the initiative is intended to broaden access to Nvidia-based infrastructure among frontier AI developers, enterprises, governments, and cloud providers. [3][4]
- Nvidia CEO Jensen Huang stated the company has the option to backstop up to $125 billion — 25% of potential deals — under the arrangement; Nvidia said the platforms would "create dedicated pools of capital at significant scale at attractive rates" for its customers, though no financial terms, individual firm commitments, or deployment timetable were disclosed. [4]
- The financing platforms are designed to create longer-duration, usage-linked investment opportunities for large asset managers and private capital firms, as surging demand for AI computing capacity draws institutional investors while governments, companies, and startups race to build out data centers. [4]
- Separately, Nvidia became the top vendor by revenue in data center Ethernet switching for the first time, marking an expansion of Nvidia's dominance in AI beyond chips into networking infrastructure. [1]
- Nvidia is also offering startup customers the option to swap compute power for a revenue share arrangement, providing an alternative financing model for early-stage AI companies. [2]
NvidiaApolloBlackRockBlackstoneBrookfieldGoldman SachsKKRJensen HuangAI infrastructure financingthird-party capital platformsdata center investmentNvidia partnershipsinstitutional investors in AIcompute financing modelsdata center Ethernet switchingrevenue share for startupsprivate capital for AIAI infrastructure buildout
7.12
Applied Materials and SMIC post record results driven by AI semiconductor demand
- Applied Materials reported record Q3 FY2026 revenue of $9.12 billion, up 25% year over year, beating analyst expectations of $8.99 billion; the semiconductor systems segment alone contributed $7.04 billion, up from $5.56 billion a year earlier, and the company generated record cash from operations of $3.04 billion. [5]
- Applied Materials earned $2.54 billion in net income for Q3 FY2026, or $3.17 per diluted share (adjusted EPS of $3.50 vs. analyst estimate of $3.40), and guided Q4 FY2026 revenue at $10.25–$10.75 billion, with CEO Gary Dickerson stating the company is "raising Semiconductor Systems revenue expectations for calendar 2026" and expects to "grow faster than the market this year." [5][6]
- AI-driven demand across logic, HBM (High Bandwidth Memory), and advanced packaging segments was cited as the key growth driver on the Applied Materials earnings call. [1][3]
- SMIC posted Q2 revenue of $3.006 billion — the first time it has exceeded $3 billion in a quarter — up 36.1% year over year and 20% quarter over quarter, while gross margin rose to 25.3% from 20.1% the prior quarter. [2][4]
- SMIC's profit attributable to shareholders reached $479.2 million, nearly tripling from $132.5 million a year earlier, while net profit came in at $164.6 million; both revenue and profit beat average analyst estimates. [4][7]
- SMIC Co-CEO Zhao Haijun attributed volume gains to accelerating demand for chips used in AI applications outside traditional CPU workloads; the company shipped 2.9 million 8-inch-equivalent wafer units, a 14% gain over the prior quarter, while per-wafer prices climbed 5.7% following price negotiations concluded with customers in Q1. [4]
Applied MaterialsSMICGary DickersonZhao HaijunHBMHigh Bandwidth Memorysemiconductor revenue recordsAI-driven chip demandadvanced packaging growthHBM memory expansionearnings beatswafer pricing increasessemiconductor systems segmentlogic chip demandquarterly financial resultsfoundry capacity utilization
7.13
US data center backlash grows amid power, water, and local governance conflicts
- Texas grid operators have temporarily paused new data-center grid connections as power requests reached 474 GW across more than 1,800 projects, overwhelming the state's grid capacity. [1]
- More than 500 local governments have enacted active data center bans, with Texas and New York among states where resistance to AI infrastructure is accelerating. [7]
- A data center developer has filed a $300 million lawsuit against a US town after it banned data centers, with the developer arguing the ban is unconstitutional. [2]
- According to a June Reuters poll cited by TIME, just one-third of Americans support data center development in their communities, with opposition described as broadly bipartisan and spanning the country. [3]
- AI data centers are emerging as a midterm political issue, with bipartisan voter backlash driven by concerns over rising electricity bills and water consumption. [4]
- A frontier AI data center — a legal category coined by the Biden administration — consumes as much electricity as a mid-sized city and uses millions of gallons of water to cool its processors; these are described as among the most resource-intensive structures ever built. The US holds nearly half of the roughly 12,000 data centers on Earth, with the highest concentration in Northern Virginia. [6]
- Only one AI company discloses its data centers' indirect water usage, according to a report on data centers' true water consumption highlighted by observers tracking corporate transparency on the issue. [5]
TexasNew YorkNorthern VirginiaBiden administrationReutersTIMEdata center energy consumptionAI infrastructure oppositionTexas grid capacitywater usage transparencybipartisan political backlashfrontier AI data centerslocal government banselectricity demand surgedata center lawsuitsmidterm political issue
7.14
Nebius Vineland AI data center hit with two stop-construction orders
- Vineland, NJ city officials issued two stop-construction orders at the Nebius/DataOne AI data center site, citing installation of an LNG tank and Bloom Energy fuel cells without city approval; Hunterbrook, which disclosed a short position in Bloom Energy ($BE) at the time of reporting, detailed both violations. [1][2][3][4][5][6]
- A Vineland Planning Board continuation public hearing is scheduled for August 17, 2026 at 6:30 PM EDT (with a pre-meeting at 6:00 PM) to address the stop-construction orders at the Nebius ($NBIS) site. [7]
VinelandNebiusDataOneBloom EnergyHunterbrookAI data center constructionstop-construction ordersLNG tank installationBloom Energy fuel cellsregulatory violationsshort position disclosureplanning board hearing
Sources
- [1]hntrbrkmedia — NEW: Vineland City officials issued a STOP CONSTRUCTION ORDER on Aug 10 at the Nebius/DataOne data center in Vineland, NJ, citing installati★twitter
- [2]RJCcapital — *Hunterbook, which disclosed a short position in $BE at the time of reporting, details Vineland data center was issued two stop-construction★twitter
- [3]wakeupnj — Vineland, NJ just hit the AI data center with stop-work orders Developers started installing an LNG tank and fuel cells without city approvatwitter
- [4]Sam_Badawi — Hunterbrook reports Vineland, NJ issued a stop-work order for the $NBIS Nebius/DataOne AI data center over unapproved $BE Bloom Energy infratwitter
- [5]PolymarketMoney — BREAKING: Hunterbrook says Nebius Vineland data center is hit with a second stop-work order over permit issues, this time halting Bloom enertwitter
- [6]tenet_research — $NBIS - faces two stop-construction orders at its Vineland AI data-center site, involving LNG and Bloom fuel-cell installations allegedly latwitter
- [7]jiwanpark1014 — $NBIS Event: Monday, August 17, 2026 — Vineland Planning Board continuation hearing Time: 6:30 PM EDT (public hearing) Pre-meeting: 6:00 PM twitter
7.15
Databricks Launches Smart Routing in Unity AI Gateway for Cost Optimization
- Databricks introduced Smart Routing in its Unity AI Gateway, a feature that automatically matches each coding task to the most cost-effective model without sacrificing quality, targeting reduction in AI spending for coding agents. [1][2]
- Omnigent released v0.9.0, which integrates Databricks AI Gateway smart routing alongside a refreshed web UI and expanded sandbox and deployment controls; the release was shaped by community contributions via issue reports, pull requests, and Discord feature discussions. [3][4]
DatabricksUnity AI GatewaySmart RoutingOmnigentDiscordAI gateway smart routingcost optimization for coding agentsmodel selection automationAI spending reductioncoding agent infrastructurecommunity-driven developmentsandbox and deployment controls
7.16
Wintermute plans $1 billion HFT and AI data center buildout over five years
- Crypto market maker Wintermute plans to invest approximately $1 billion over five years in high-frequency trading technology and AI data center infrastructure, with the spending to be financed through retained earnings. CEO Evgeny Gaevoy stated the firm intends to train quantitative models on large volumes of market data and expand computing, storage, and networking capacity to compete with established electronic trading firms in traditional financial markets. [1][2][3][4][5][6][7]
- Wintermute is targeting a major revenue shift: non-crypto markets (equities, commodities, foreign exchange, and prediction markets) currently generate about 10% of revenue, but the firm aims for non-crypto activities to exceed half of total company revenue by the end of 2027. [7]
- Gaevoy acknowledged the competitive challenge of entering traditional markets, noting that rival firms "have spent decades optimizing their technology and infrastructure for these markets," making substantial infrastructure investment a prerequisite. [7]
WintermuteEvgeny Gaevoyhigh-frequency tradingAI data center infrastructurequantitative model trainingcrypto market makertraditional financial markets expansionnon-crypto revenue diversificationelectronic trading competitionmarket data analytics
7.17
Nebius Group Q2 FY2026 AI Cloud Earnings Show Explosive Revenue Growth
- Nebius Group reported Q2 FY2026 revenue of $582.3M, up 454% year-over-year and 46% quarter-over-quarter, beating analyst expectations of $573.9M, while adjusted EBITDA reached $236.2M compared to -$21M in the same period the prior year. [6][7][9]
- Nebius's AI Cloud Annual Recurring Revenue hit $3.0 billion in Q2, rising 56–58% within a single quarter. [6][9]
- Total contract value for deals closed in Q2 grew 4x quarter-over-quarter, with annual contract value (ACV) per megawatt escalating sharply: from a ~$12M base in the 2026 baseline to over $20M in Q2 deals, and exceeding $40M for short-term capacity deals in Q3. [3][8]
- New capacity deals are now being priced at $40–50 million per megawatt, with Nebius signing its first contract at that price level during the quarter. [1]
- Nebius reported a payback period of approximately one year and ten months on new deals, measured excluding any prepayments, with the improvement driven by the rising ACV per megawatt. [4]
- Nebius's stock rose approximately 15–30% in after-hours trading following the earnings release. [2][5]
Nebius GroupAI cloud revenue growthannual recurring revenuedata center capacity pricingcontract value per megawattearnings beatEBITDA turnaroundhyperscaler infrastructuredeal payback period
7.18
Hetzner launches free experimental AI inference API with open-weight models
- Hetzner, the German cloud and hosting provider, launched a free experimental AI inference service featuring an OpenAI-compatible API, available to its customers during the testing/beta phase. [2][3][4]
- The service offers access to several open-weight models, specifically DeepSeek-V4-Flash-0731, GLM-5.2-NVFP4, Kimi-K2.7-Code, and Qwen3.6-35B-A3B-FP8. [1][4]
HetznerDeepSeek-V4-Flash-0731GLM-5.2-NVFP4Kimi-K2.7-CodeQwen3.6-35B-A3B-FP8OpenAIAI inference serviceopen-weight modelscloud hostingOpenAI-compatible APIbeta testingfree tier accessGerman cloud provider
7.19
AI data center investment accelerates as hyperscalers race for compute capacity
- Meta and Oracle are making notable moves to secure large-scale data center capacity, with their decisions being examined as indicators of broader shifts in data center economics and how hyperscalers approach infrastructure procurement at scale. [2]
- Colocation data centers are playing an increasingly central role in enabling rapid AI expansion, as companies turn to third-party facilities to meet surging compute demand without bearing the full burden of building proprietary infrastructure. [3]
- Investment in AI data centers is accelerating, with capital expenditure into compute facilities framed as a critical strategic bottleneck for AI development, raising expectations around the infrastructure build-out required to support continued AI growth. [1][3]
MetaOracledata center infrastructurecolocation facilitieshyperscaler procurementAI compute capacitycapital expenditureinfrastructure build-outthird-party facilitiesAI expansion