In partnership with
Cut Lead Review From Hours To Minutes
Sign up for a free trial of Attio, the agentic CRM.
Ask Attio to build a daily workflow that surfaces the deals that need your attention today, like anything with a stage change, a recent reply, or a new signal in the last 24 hours.
Review your pipeline in Claude, synced live from Attio via MCP.
That's it.
The GPU Daily · #015 · covering Tuesday 28 July
The physical layer of AI, every day. Here's what moved on Tuesday 28 July 2026.
In this issue
Want The GPU Daily in your inbox?
Top story · HARDWARE
China · ▼ Bearish · 27 Jul 05:30 UTC · Newsroom · 19 sources
CXMT's Shanghai IPO closed with shares surging on debut, giving China's most credible domestic memory producer a public market war chest to accelerate HBM and DRAM capacity. The listing matters for GPU infrastructure because it funds the one part of China's AI stack that remains genuinely constrained: high-bandwidth memory that SK Hynix and Samsung currently dominate, and that US export controls are now explicitly targeting. Watch for CXMT's next capacity announcement and whether Chinese neocloud operators begin publicly committing to domestic memory procurement timelines.
MARKET MOVES · 3
US · ▼ Bearish · 28 Jul 22:45 UTC · Newsroom
CNBC's Jim Cramer reports that Wall Street investors are exiting AI-related stocks and rotating into alternative sectors.
Why this matters: Potential investor sentiment shift away from AI infrastructure plays, which could affect funding availability for neoclouds and data centre operators.
Global · ▼ Bearish · 28 Jul 10:15 UTC · Newsroom · 2 sources
Global stock markets showed mixed performance with South Korean chipmaker stocks declining, reflecting sector-wide pressure.
Global · ▲ Bullish · 28 Jul 22:00 UTC · Newsroom
Leading AI companies and investors are evaluating next-phase investments and acquisitions as the market matures beyond initial model releases.
NEOCLOUD · 8
India · ▲ Bullish · 27 Jul 11:16 UTC · Company/PR
Yotta launched Shakti Studio, an AI inference and deployment platform supporting NVIDIA L40S and H100 GPUs with serverless scaling, fine-tuning via LoRA/QLoRA/DPO, and OpenAI-compatible endpoints.
Why this matters: Expands India-based GPU inference capacity with managed MLOps, reducing latency for Indian enterprises and supporting IndiaAI Mission infrastructure goals.
India · ▲ Bullish · 27 Jul 11:08 UTC · Company/PR
Yotta's Shakti Studio supports modern alignment techniques including SFT, DPO, and GRPO, plus LoRA and QLoRA for efficient model adaptation, with flexible token-based and GPU-minute pricing.
Why this matters: Provides managed infrastructure for post-training alignment at scale, reducing barrier to entry for Indian AI labs building custom models without owning GPU clusters.
India · ▲ Bullish · 27 Jul 11:06 UTC · Company/PR
Yotta launched Shakti AI Workspace VM offering on-demand access to NVIDIA H100 SXM and L40S GPUs with project-level isolation, role-based access control, and optional NVLink-enabled H100 configurations.
Why this matters: Provides enterprise-grade GPU workspace isolation for Indian AI teams, supporting multi-tenant governance and compliance requirements for sovereign AI development.
India · ▲ Bullish · 27 Jul 11:03 UTC · Company/PR
Shakti Cloud Bare Metal offers NVIDIA HGX H100 and L40S GPUs with InfiniBand Quantum-2 NDR 400G switches delivering 3200 Gbps node-to-node connectivity, with plans to add B200, B300, and RTX PRO 6000 support.
Why this matters: Adds India-based bare metal GPU infrastructure with enterprise-grade networking, supporting distributed training for sovereign AI models without cloud abstraction overhead.
India · ▲ Bullish · 27 Jul 11:00 UTC · Company/PR
Yotta's Shakti Clusters support both Kubernetes and SLURM orchestration on dedicated bare metal with HGX H100 (80GB) and L40S (48GB) nodes, NVLink (900 Gbps) and InfiniBand (3200 Gbps) interconnect.
Why this matters: Enables Indian AI teams to choose orchestration based on workload type (inference vs. training), reducing lock-in to single scheduling paradigm and improving GPU utilisation efficiency.
India · ▲ Bullish · 27 Jul 10:55 UTC · Company/PR
Yotta's Shakti Studio emphasises data residency and local governance under a 'Train Local, Scale Global' strategy for emerging-market AI development.
Why this matters: Demand for GPU infrastructure in India with data sovereignty constraints, relevant to neocloud expansion in regulated markets outside US/EU.
US · ▲ Bullish · 22 Jul 12:00 UTC · Company/PR
CoreWeave is providing AI cloud infrastructure to Anam for real-time photorealistic AI avatar development. Company-stated.
Why this matters: Neocloud demand from avatar/synthetic-media workloads, a growing inference-heavy use case requiring sustained GPU capacity.
Italy · ▲ Bullish · 28 Jul 14:15 UTC
Italian energy company Eni announced opening its supercomputing infrastructure to external partners, details sparse.
Why this matters: Enterprise compute capacity being monetised as a service, relevant to neocloud competitive dynamics and GPU utilisation models.
SOFTWARE & AI · 12
Global · ▲ Bullish · 28 Jul 12:00 UTC · Trade press
Anthropic released the largest MCP update since launch, transitioning to stateless architecture that eliminates sticky routing, enabling deployment behind standard load balancers and Kubernetes, with formal 12-month deprecation policy developed with Google, Microsoft, and Amazon.
Why this matters: Removes enterprise deployment blocker for multi-agent systems; stateless design reduces infrastructure complexity and enables horizontal scaling of agent workloads across distributed GPU clusters.
China · ◆ Neutral · 28 Jul 02:08 UTC · Trade press
Moonshot AI released full weights for Kimi K3, a 2.8T mixture-of-experts model with 1M-token context, requiring inference providers with >$20M annual revenue to negotiate separate commercial agreements.
Why this matters: Sets precedent for open-weight licensing with revenue-based commercial tiers, exposing open-source releases to US regulatory scrutiny under ICTS and sanctions mechanisms.
China · ▼ Bearish · 28 Jul 07:46 UTC · Trade press
China now accounts for two-thirds of global LLM API calls, with Xiaomi MiMo-V2.5 ranking first in weekly and monthly token usage worldwide.
Why this matters: Shift in inference compute demand toward Chinese models and domestic cloud infrastructure; implications for GPU demand distribution between US and China markets.
US · ◆ Neutral · 28 Jul 22:10 UTC · Trade press
Visa deployed Anthropic's Claude Mythos to identify critical-severity vulnerabilities across its payment network spanning 200+ countries and 5 billion credentials, then open-sourced the Visa Vulnerability Agentic Harness on GitHub.
Why this matters: Enterprise-scale AI agent deployment for security validation; Visa's open-source harness and Project Lightwell participation ($5B IBM/Red Hat initiative) signal growing standardisation of AI-driven vulnerability detection across financial infrastructure.
US · ▲ Bullish · 28 Jul 17:29 UTC · Trade press
GM's autonomous driving division tripled merged pull requests by redesigning engineering workflows around AI agents, with engineers spending only 15% of time writing code while agents automate vehicle data analysis, triage, and testing.
Why this matters: Enterprise-scale AI agent deployment reducing engineering time-to-output; GM's use of custom Model Context Protocol servers and petabytes of internal data growing demand for inference infrastructure supporting agentic workflows.
US · ◆ Neutral · 28 Jul 17:20 UTC · Trade press
Runway ML discovered a bug where AI avatars drifted off-center during real-time generation; instead of fixing the backend, the team wrapped it as a frontend feature called 'Optimize for Image Quality' that automatically re-centers input images, reducing generation time by 80-90% through distillation and adversarial post-training.
Why this matters: Illustrates inference optimisation trade-offs in production video generation; Runway's 8% API call drop traced to faulty GPUs in a single data centre hardware reliability as a bottleneck for real-time inference services.
US · ▲ Bullish · 28 Jul 16:27 UTC · Company/PR
AI2 launched OlmoEarth Platform for geospatial inference using foundation models pretrained on 10 terabytes of satellite data, processing continent-scale wildfire risk mapping across North America using 994 GPUs and 19,600 CPUs in parallel, reducing 4,737 hours of serial compute to 30.5 hours.
Why this matters: Large-scale GPU inference workload (994 GPUs) for geospatial analysis; OlmoEarth's three-stage hardware-matched pipeline and custom metadata indexing signal emerging patterns in inference infrastructure design for scientific computing.
Global · ▲ Bullish · 28 Jul 05:00 UTC · Company/PR
Moonshot AI's 2.8T parameter Kimi K3 model is available via Dell Enterprise Hub for on-premises deployment, enabling enterprises to run frontier AI without cloud dependencies.
Why this matters: Expands deployment options for open-weight models beyond cloud endpoints, reducing vendor lock-in for enterprises building sovereign AI infrastructure.
US · ▲ Bullish · 28 Jul 17:30 UTC · Company/PR
Google Cloud moved Conversational Analytics from preview to general availability for BigQuery and Looker, with support for multi-cloud environments including AWS S3 and Apache Iceberg REST catalogs, plus new enterprise security controls and agentic workflows.
Why this matters: Expands inference platform footprint across multi-cloud data estates; agentic workflows with scheduled reporting and anomaly detection signal growing demand for continuous GPU-backed inference at scale.
US · ◆ Neutral · 28 Jul 17:00 UTC · Company/PR
OpenAI released a field report on AI coding agents in scientific computing, claiming agents accelerate software development and discovery in genomics, company-stated.
US · ◆ Neutral · 29 Jul 00:00 UTC · Company/PR
Together AI released documentation on configuring dedicated model inference endpoints for enterprise workloads.
US · ◆ Neutral · 28 Jul 17:30 UTC · Company/PR
AWS released the Spark Upgrade Agent, a fully managed tool that automates Apache Spark 3.x to 4.0 migration for Scala, PySpark, and Spark SQL workloads on Amazon EMR, using Model Context Protocol to analyse code and validate jobs against live EMR Serverless clusters.
HARDWARE · 6
US · ▲ Bullish · 28 Jul 21:00 UTC · Trade press
Cadence Design Systems certified AI reference design flows for Intel's 18A-P and 14A process nodes, enabling AI chip development on Intel Foundry Services.
Why this matters: Removes design barriers for AI accelerator development on Intel's advanced nodes, potentially enabling new competitors to NVIDIA and AMD in the GPU market.
US, Vietnam, Malaysia · ▲ Bullish · 28 Jul 10:00 UTC · Trade press
OCI Holdings plans to double polysilicon production from 35,000 to 70,000 metric tons by 2029, targeting US AI infrastructure and space industries.
Why this matters: Polysilicon supply chain repositioning toward US-bound AI infrastructure, with OCI establishing non-prohibited foreign entity supply chain through Vietnam and Malaysia operations.
Global · ▲ Bullish · 29 Jul 00:00 UTC · Company/PR
ASMPT won orders for more than 50 advanced packaging tools for chip-to-substrate bonding, a critical step in GPU and accelerator assembly.
Why this matters: Sustained demand for advanced packaging capacity; thermo-compression bonding is essential for high-density GPU interconnects and chiplet assembly.
Global · ▲ Bullish · 29 Jul 00:00 UTC · Company/PR
ASMPT announced strong interim results for 2026 with multiple advanced packaging products driving performance, company-stated.
Why this matters: Sustained demand for GPU and accelerator packaging equipment; strong results suggest continued investment in next-generation chip assembly capacity.
Global · ◆ Neutral · 28 Jul 15:00 UTC · Company/PR
Cisco's Zeus AI Lab benchmarked a scale-across Ethernet fabric designed to unify distributed GPU clusters, claiming AI workloads are outgrowing standalone data centres. Company-stated, not independently verified.
Why this matters: If validated, addresses multi-cluster orchestration bottleneck that constrains large-scale distributed training and inference deployments.
Europe · ▲ Bullish · 28 Jul 14:15 UTC
Qualcomm announced its Innovation Fellowship Europe programme recognising research in AI and cybersecurity.
HYPERSCALER · 3
India · ▲ Bullish · 27 Jul 10:58 UTC · Company/PR
Sarvam AI trained PARAM-1 and PARAM-2 models end-to-end on Yotta's Shakti Cloud, scaling from single node to hundreds of H100 GPUs with stable performance, now serving over 1 million daily inference requests.
Why this matters: India-based GPU infrastructure for production AI workloads at scale, reducing latency and regulatory risk for Indian AI labs compared to US cloud providers.
China · ▼ Bearish · 28 Jul 07:47 UTC · Trade press
China Telecom deployed its first domestic TPU cluster in Hangzhou using Zhonghao Xinying Chana TPU with 400 PFLOPS, part of a national-grade computing chain from chip to platform.
Why this matters: China's domestic AI accelerator deployment at hyperscaler scale; alternative to NVIDIA GPUs for training workloads within Chinese cloud infrastructure.
US · ◆ Neutral · 28 Jul 16:31 UTC · Company/PR
DATA CENTRE · 1
UAE, Guam, Iran · ▼ Bearish · 28 Jul 13:00 UTC · Company/PR
AWS's me-central-1 region in the UAE and Bahrain suffered physical damage from drone strikes in April 2026, leaving it unable to reliably support customer applications; separate incidents included an 80% traffic drop in Guam from typhoon and Iran's 88-day Internet shutdown.
Why this matters: Geopolitical and natural disaster risk to data centre infrastructure in conflict zones and typhoon-prone regions; raises questions about regional redundancy and disaster recovery planning.
ENERGY & POWER · 15
US · ▲ Bullish · 28 Jul 12:02 UTC · Trade press
CenterPoint Energy submitted 17GW of large-load projects to ERCOT's new interconnection process and expects 14GW to qualify in Batch Zero, representing a 65% increase from the utility's current 21GW Houston-area peak demand.
Why this matters: Rapid data centre load growth in Texas; CenterPoint's $1.2B capex increase to $66.7B through 2035 grid operator confidence in sustained large-load demand.
US · ▲ Bullish · 27 Jul 22:10 UTC · Trade press
Xcel Energy proposed a clean transition tariff to Colorado's PUC modelled on Google's 2023 Nevada geothermal deal; Google committed $50M to support distributed capacity procurement and is piloting virtual power plant contracting with NV Energy.
Why this matters: Establishes template for hyperscaler-backed clean energy procurement outside traditional PPAs, reducing grid interconnection delays for data centre load growth.
US · ◆ Neutral · 28 Jul 21:21 UTC · Trade press
PJM Interconnection, the largest US electricity market, is moving toward structural reforms to address grid capacity constraints driven by data centre and AI compute load.
Why this matters: PJM reform directly impacts data centre interconnection timelines and power availability for GPU clusters across the Mid-Atlantic and Midwest; delays or restrictions could slow neocloud and hyperscaler expansion.
US · ▲ Bullish · 28 Jul 14:14 UTC · Trade press
Port of Long Beach signed a memorandum of cooperation with the Department of Transportation's Maritime Administration and a separate lease with BlueCore Energy to develop and test 10-MWe SMRs on floating power barges.
Why this matters: Establishes regulatory pathway for floating SMRs at US ports, potentially enabling distributed power generation for coastal data centre clusters.
US · ▲ Bullish · 28 Jul 17:46 UTC · Trade press
The US Department of Energy selected Idaho, Louisiana, Oklahoma, Tennessee, and Utah as finalists to host Nuclear Lifecycle Innovation Campuses, which could attract up to $50 billion in capital investment and host new reactors, advanced manufacturing, and data centres.
Why this matters: Opens pathway for data centre co-location with nuclear fuel cycle facilities; DOE's explicit mention of data centres as a campus use case federal recognition of compute's energy demands and potential for SMR deployment.
Australia · ▲ Bullish · 28 Jul 12:00 UTC · Trade press
Australia added over 700MW of rooftop solar in Q2 2026, lifting cumulative NEM capacity to 26.4GW, with 14 new grid-scale projects reaching full output totalling 3.9GW and wholesale prices falling 47% year-on-year to AUD 74/MWh.
Why this matters: Renewable-driven price compression in Australian grid; relevant to data centre power procurement costs in APAC region.
Taiwan · ▲ Bullish · 28 Jul 12:30 UTC · Trade press
Taiwan will require new, expanded, or renovated buildings of 1,000 square metres or larger to install solar PV at 1kW per 20 square metres, expected to add 660MW annually and support doubling of solar capacity by 2035.
Why this matters: Increases renewable generation capacity available for data centre procurement in Taiwan, a key semiconductor and AI infrastructure hub.
Germany · ▲ Bullish · 28 Jul 09:40 UTC · Trade press
Zenobē acquired German BESS developer SDP Energie on 27 July, gaining six transmission-level projects totalling 1.75GW in Germany.
Why this matters: Consolidates BESS development capacity ahead of 4 August 2029 German regulatory deadline, which triggered 15GWh of large-scale BESS announcements since May.
Italy · ◆ Neutral · 28 Jul 13:37 UTC · Trade press
RINA and Ansaldo Energia launched an initiative to help Italian companies enter the nuclear supply chain, following the Chamber of Deputies' approval of a bill to reintroduce nuclear energy after 34 years.
Why this matters: Opens domestic supply chain for nuclear projects; Italy's nuclear restart could eventually support data centre power procurement, though timeline remains uncertain.
Malawi · ▲ Bullish · 28 Jul 08:52 UTC · Trade press
Malawi commissioned a 20MW/40MWh grid-scale BESS facility in Kanengo with a US$20 million grant, expected to unlock approximately 100MW of previously constrained renewable generation.
Why this matters: BESS deployment model for low-income countries; facility includes grid-forming capabilities to stabilise feeder voltages, relevant to emerging-market data centre power infrastructure.
Europe · ▲ Bullish · 28 Jul 05:01 UTC · Trade press
TenneT invested €6.85 billion in the grid during the first half of 2026, supporting European transmission infrastructure expansion.
Why this matters: Continued European grid capex to support data centre and AI infrastructure load growth; TenneT serves major data centre hubs in Netherlands and Germany.
US · ◆ Neutral · 28 Jul 18:16 UTC · Trade press
Lincoln Electric System deployed a 3-MW/12-MWh zinc-based battery from Eos Energy Enterprises to support a 30-MW microgrid serving Nebraska's state Capitol and critical infrastructure, increasing capacity by approximately 10%.
Why this matters: Non-lithium battery deployment for grid-scale storage; Eos Energy's gigawatt-scale supply contracts (36 MW in Missouri, 100 MW in Texas) signal growing availability of alternative battery chemistry for data centre microgrids.
US · ◆ Neutral · 28 Jul 13:03 UTC · Trade press
Viridi installed its RPS 150 battery energy storage system at Oak Ridge National Laboratory's Grid Research Innovation and Development Center for DOE resource integration testing.
Why this matters: Battery storage integration with grid infrastructure; relevant to data centre backup and demand-response capabilities, though not directly compute-focused.
China · ◆ Neutral · 28 Jul 10:19 UTC · Trade press
CATL's Tener sodium-ion system retains over 92% capacity at -20°C with 15,000 cycles, and the company allocated RMB 5 billion to expand production by 40GWh annually.
Global · ◆ Neutral · 28 Jul 09:54 UTC · Trade press
Arenko's Nimbus platform coordinates solar and battery assets in real time across forecasting, control, market, and grid signal layers to avoid curtailment and clipping losses.
REGULATION & POLICY · 3
US · ▼ Bearish · 28 Jul 23:15 UTC · Newsroom
Scientists from OpenAI and Anthropic have asked the US government for regulatory tools to slow AI development timelines.
Why this matters: If adopted, development-pacing measures could reduce training cluster utilisation and GPU demand, affecting infrastructure investment timelines.
US, China · ▲ Bullish · 28 Jul 20:15 UTC · Newsroom
Major AI companies stated they want Chinese AI models to remain available, though observers question the sincerity of these claims.
Why this matters: Tension between US export controls on advanced chips and industry desire for open model access, affecting GPU allocation to Chinese labs.
US, China · ◆ Neutral · 28 Jul 00:13 UTC · Trade press
Anthropic CEO Dario Amodei stated his company has never advocated banning open-weight models but called for restricting China's chip access and formal crackdowns on AI distillation to prevent CCP military superiority.
Why this matters: Industry fracture on open-weight policy: Anthropic diverges from Meta, Mistral, and Hugging Face on export controls, raising stakes for US AI regulation and open-source governance.