SendTech Times
العربية
Cloud & Data CentersNews|July 8, 2026, 03:54 PM
CAPACITY TEST:

AI Times Says Infrastructure Captures 82% Of Generative AI Value

BySendTech Times Cloud & Infrastructure DeskNewsroom-edited, source-reviewed coverage|Source: AI Times Korea
Newsroom brief

AI Times Korea cited Exponential View data showing 82% of measured AI-economy value going to cloud, GPU and inference infrastructure in Q1 2026, while foundation-model companies accounted for 11% and applications 7%.

Verified against source materialEdited by SendTech Times Cloud & Infrastructure Desk
AI Times Says Infrastructure Captures 82% Of Generative AI Value
Image source: AI Times Korea

Exponential View Reports 82% Of AI Value In Infrastructure

AI Times Korea cited an Exponential View report showing that cloud, GPU and inference infrastructure captured the largest measured share of the generative AI economy, even as model developers remain the most visible companies in the sector.

The Korean technology outlet said Exponential View's State of the AI Economy 2026 report measured where value accrues across the AI stack, including infrastructure, foundation models and applications.

AI Times Korea said Exponential View put 82% of captured AI-economy value in the hosting layer in the first quarter of 2026.

The hosting layer includes cloud services, GPUs and inference infrastructure.

For the most recent 12 months, AI Times Korea said generative AI recorded $110 billion in actual revenue.

It also said the annualised revenue run rate had expanded to $175 billion, framing generative AI as a fast-growing enterprise technology market rather than only a consumer chatbot trend.

Foundation Models Account For 11% In The Report

AI Times Korea said the Exponential View report assigned 11% of measured value to the foundation-model layer, including companies such as OpenAI and Anthropic.

The application layer accounted for the remaining 7%, according to the same article.

AI Times Korea said user payments for ChatGPT-like services still push much of the economic value towards GPUs, data centres and cloud operations needed to serve models at scale.

AI Times Korea also said applications increased from a 4% share a year earlier to 7% in the latest period.

AI Times Korea linked the increase to growth in AI agents, coding tools and enterprise AI services that perform work on top of foundation models.

FT Questions Model-Company Valuations

AI Times Korea also referenced Financial Times commentary on possible OpenAI and Anthropic listings.

The Financial Times commentary cited research, development, GPU investment and cloud spending as costs that could make durable profitability harder to prove.

AI Times Korea wrote that the Exponential View report was analysing value across the whole AI industry, not publishing financial statements for individual companies.

That means the foundation-model layer figure should not be read as a direct company-level share for OpenAI or Anthropic.

AI Times Korea said OpenAI and Anthropic operate services beyond foundation models, including ChatGPT, Claude, enterprise AI services and API platforms.

Their revenue can therefore sit across model, application and platform layers rather than only inside the foundation-model category.

AI Times Korea said narrower performance gaps between advanced models could also shift competition towards enterprise services, AI agents and productivity platforms.

The valuation debate stays tied to disclosed model economics rather than a claim that foundation-model companies have already lost customer demand.

OpenAI Infrastructure Costs Remain A Missing Company-Level Figure

AI Times Korea said OpenAI is pursuing the Stargate project and in-house chip development as ways to lower infrastructure costs and increase supply-chain control.

AI Times Korea described those moves as part of the broader pressure created by compute and hosting costs.

AI Times Korea did not identify OpenAI or Anthropic profit figures, cash-flow figures, named customers, infrastructure-spending totals or IPO schedules.

AI Times Korea also did not disclose the exact spending commitments behind Stargate or any measured cost reduction from in-house chip work.

Share this article
inXf

Related articles

More
Meta Compute Prepares AI Cloud Push Against AWS And Google
Cloud & Data Centers

Meta Compute Prepares AI Cloud Push Against AWS And Google

Meta is considering external sales of AI compute and model access through Meta Compute. Zuckerberg said the idea is on the table, but customer names, pricing, GPU inventory and a Muse Spark release date remain undisclosed.

India's 8.33 GW Data-Center Pipeline Tests AI Infrastructure Readiness
Cloud & Data Centers

India's 8.33 GW Data-Center Pipeline Tests AI Infrastructure Readiness

Knight Frank India says the country's data-center development pipeline has reached 8.33 GW, with Mumbai, Hyderabad and Chennai showing how AI demand is turning capacity, power, fibre and subsea links into strategic constraints.

Iren Plans 800MW Australia AI Data Center Campus as Power Becomes the Capacity Gate
Cloud & Data Centers

Iren Plans 800MW Australia AI Data Center Campus as Power Becomes the Capacity Gate

Iren signed a transmission connection agreement for a planned 800MW data center campus in Bundey, South Australia. The project is Iren's first Australian foray and is expected to be energized in 2028 as the company shifts more cash flow toward AI cloud infrastructure. The practical question is whether Iren can turn grid-connected power, financing and GPU capacity into energized AI cloud campuses on the announced timelines.

CAS Star’s Photonics Bet Turns Into an AI Infrastructure Test
Cloud & Data Centers

CAS Star’s Photonics Bet Turns Into an AI Infrastructure Test

CAS Star founder Mi Lei says the AI boom has validated a decade-long investment thesis around photonics and other hard-tech fields. The firm has more than 200 photonics-related companies among roughly 600 portfolio companies, spanning sensing, communications, computing, storage and display. The next test is whether optical links, laser chips and photonic computing companies can turn AI data-centre demand into durable commercial scale.

Oracle AI Buildout Raises Debt Test After 19% Stock Drop
Cloud & Data Centers

Oracle AI Buildout Raises Debt Test After 19% Stock Drop

Oracle shares fell 19% in a week as investors weighed about $130 billion in debt, nearly $56 billion in fiscal 2026 capital spending and a financing plan tied to AI data-center expansion.

Eco Wave Links Shoreline Power To The Energy Strain Behind AI Infrastructure
Cloud & Data Centers

Eco Wave Links Shoreline Power To The Energy Strain Behind AI Infrastructure

Eco Wave Power is using NVIDIA AI infrastructure and digital twins to model wave-energy equipment near ports and industrial zones, where future AI facilities may face grid and permitting limits.

Keep Reading

More Stories

Latest
Klarna Seeks U.S. Industrial Bank Charter As Partner-Bank Model Faces ReviewFintech & Digital PaymentsJul 8, 2026Klarna Seeks U.S. Industrial Bank Charter As Partner-Bank Model Faces ReviewBanking Dive reported that Klarna applied for a U.S. industrial bank charter after the company said it had extended $91.3 billion in credit to U.S. consumers; the application still needs FDIC and Utah approval.SambaNova Raises $1 Billion For AI Inference Hardware Without IPO TimingChips & SemiconductorsJul 8, 2026SambaNova Raises $1 Billion For AI Inference Hardware Without IPO TimingTechCrunch reported that SambaNova Systems raised $1 billion at an $11 billion valuation, named JPMorganChase as an inference-infrastructure partner and did not give IPO timing or SN50 customer shipment dates beyond the second half of 2026.FuriosaAI Starts RNGD Accelerator Deployment At Equinix Lisbon DatacenterChips & SemiconductorsJul 8, 2026FuriosaAI Starts RNGD Accelerator Deployment At Equinix Lisbon DatacenterFuriosaAI has begun deploying its RNGD AI accelerators at Equinix’s LS2 datacenter in Lisbon as the South Korean chip startup looks for European sovereign AI demand. The company describes 48 GB of HBM3, 1.5 TB/s memory bandwidth and 512 teraFLOPS of dense FP8 performance per card, but its Broadcom-linked next-generation accelerator still depends on HBM4 and HBM4e timing.Kraken Targets European Banking Licence In 10-Year Licence PushFintech & Digital PaymentsJul 8, 2026Kraken Targets European Banking Licence In 10-Year Licence PushKraken is pursuing a full European banking licence with Lithuania as the focus jurisdiction, according to a person familiar with the plans. CEO Arjun Sethi said Kraken’s parent company aims to acquire licences over the next 10 years, while Kraken declined to comment and Lithuania’s central bank said licensing processes are confidential.Meta Muse Image Uses Public Instagram Posts Unless Users Opt OutCapital & PolicyJul 8, 2026Meta Muse Image Uses Public Instagram Posts Unless Users Opt OutMeta has launched Muse Image with Instagram integration that allows public profile photos and videos to be used in AI image generations by default. Users can switch off sharing in Instagram settings, but Meta says existing AI images made with their content will not be deleted and users will not be notified when someone creates them.IBM Adds Rackmount z17 And LinuxONE 5 Systems Without Customer CountsChips & SemiconductorsJul 8, 2026IBM Adds Rackmount z17 And LinuxONE 5 Systems Without Customer CountsIBM is extending its z17 and fifth-generation LinuxONE families with single-frame and rackmount systems, including LinuxONE Express as an 18U entry point. The smaller hardware brings Telum II and Spyre AI capability to lower-scale deployments, but IBM did not disclose customer counts, pricing or shipment volumes for the new models.Dubai Data Manual Adds AI-Ready Rules Without Entity Rollout DatesCapital & PolicyJul 8, 2026Dubai Data Manual Adds AI-Ready Rules Without Entity Rollout DatesDigital Dubai, a government entity, launched an updated Dubai Data Manual for AI-ready public-sector data governance, but the official announcement did not disclose entity-level rollout dates, penalties, budgets, maturity scores or measured service improvements.Insilico Takes AI-Discovered IPF Drug Into Phase III Without Approval DatesAIJul 8, 2026Insilico Takes AI-Discovered IPF Drug Into Phase III Without Approval DatesAI News reported that Insilico Medicine is moving its AI-discovered IPF drug rentosertib into Phase III after a 71-patient Phase IIa trial across 22 Chinese sites, but the company did not disclose Phase III enrolment, completion timing or approval dates.Microsoft Finds Cheaper AI Model Rates Can Still Raise Agent CostsAIJul 8, 2026Microsoft Finds Cheaper AI Model Rates Can Still Raise Agent CostsMicrosoft said lower token prices for Claude Sonnet 5 did not remove AI agent cost spikes when it compared Claude models inside GitHub Copilot.Dubai Gaming Retreat Sets Five Priorities Without Funding PlanEconomyJul 8, 2026Dubai Gaming Retreat Sets Five Priorities Without Funding PlanDubai Media Council convened more than 100 industry leaders and more than 300 gaming community members to shape policy recommendations, but it has not disclosed funding amounts, deadlines, incentives or company commitments.Chinese Brain-Mapping Chip Claims 478 Times NVIDIA A100 Speed Without Clinical ProofChips & SemiconductorsJul 8, 2026Chinese Brain-Mapping Chip Claims 478 Times NVIDIA A100 Speed Without Clinical ProofA Peking University and Chinese Academy of Sciences team reported a 40-nanometre computing-in-memory chip that maps the brain’s folded cortex in less than 0.5 seconds, but the study did not disclose clinical validation or a commercial release plan.Dubai VDX Vertiport Gets GCAA Certification For eVTOL Air TaxisEconomyJul 7, 2026Dubai VDX Vertiport Gets GCAA Certification For eVTOL Air TaxisGCAA and Skyports said Dubai’s VDX vertiport has received regulatory certification for eVTOL aircraft operations, with capacity for up to 170,000 passengers a year but no first service date disclosed.