Technology
New Research Challenges GPU Utilization as Measure of Inference Infrastructure
Published
2 hours agoon
By
New NAND Research report underscores that busy GPUs aren’t an accurate measure of productive output; introduces new Inference Economics Scorecard to measure inference infrastructure by the useful work it delivers, as applied in an eight-hour benchmark with Nebius on NVIDIA HGX B300 GPUs and WEKA NeuralMesh.
CHICAGO, Nov. 16, 2026 /PRNewswire/ — From Supercomputing 2026: WEKA, the AI data and memory infrastructure company, in collaboration with market intelligence firm NAND Research, today introduced the Inference Economics Scorecard, a new common framework for measuring inference infrastructure productivity and economic value by the useful work it delivers rather than by how busy it keeps GPUs. The scorecard defines nine measures, including tokens per watt, cost per inference, and context retention. It sets four testing rules, including runs of at least eight hours and counting only successful output. Already applied in a benchmark with Nebius, the AI cloud company, it comes as inference is claiming a growing share of AI infrastructure spending and buyers hit power, memory, and rack-space limits.
Buyers Are Already Treating Inference Differently, New Research Finds
New research from 451 Research by S&P Global shows organizations are already adapting, projecting that inference will grow from 34% of AI infrastructure spending in 2026 to 46% by 2030 and finding that 41% of organizations running inference buy fundamentally different infrastructure for it than for the rest of their IT stack.
“Inference is where the value of AI is ultimately realized, and our research shows it’s reshaping how organizations buy infrastructure,” said Greg Macatee, Senior Analyst at 451 Research by S&P Global. “Agents are driving exponential growth in token volume; memory, power, and supply constraints greatly challenge the notion that organizations can keep up simply by adding more GPUs.”
Why GPU Utilization Falls Short for Inference
For years, training was the dominant AI workload. Because training keeps GPUs busy with large, predictable jobs, GPU utilization became the default measure of AI infrastructure productivity. Inference, especially agentic inference, places very different demands on GPUs. An agent does not simply answer a prompt; it completes a task over many steps, and at each step it re-reads the context it built earlier. After the GPU evicts that context from memory to make room for other users and sessions, it must recompute it when the task returns, and the task makes no further progress until that work is done. The result is a GPU that is fully utilized but unproductive because it wastes cycles recomputing the same tokens multiple times.
The Inference Economics Scorecard measures inference systems at production scale: completed requests across many concurrent users, sustained over hours rather than minutes.
“The industry has been grading inference with a training-era report card. GPU utilization tells you the chip was busy. It cannot tell you whether the work was finished, repeated, or thrown away, and with agents, a lot of it is repeated,” said Liran Zvibel, co-founder and CEO of WEKA. “Power, supply, and rack space are constraints now as much as capital. The new math required for inference infrastructure is useful work per watt, per rack, per dollar, at the response times your users expect. The Inference Economics Scorecard puts that math on paper, and WEKA will be measured by it like everyone else.”
From Busy to Productive: How the Scorecard Works
The Inference Economics Scorecard defines nine measures of inference performance and economics: tokens per watt, tokens per rack, cost per inference, GPU goodput, time to first token and end-to-end latency, memory efficiency, operational density, agent concurrency, and context retention. It also sets the discipline for testing against them:
The same workload and service targets across every system testedMeasurement of the whole serving stackRuns of at least eight hoursOnly successful output counted
“The most expensive mistake in AI infrastructure right now is buying for peak throughput and paying for repeated work,” said Steve McDowell, principal analyst and founder at NAND Research. “Agentic workloads make that mistake easy to miss, because every step of a long task re-reads context that a busy GPU may have to rebuild from scratch. We designed the scorecard so buyers can hold any system to the same standard: finished work, measured over hours, with every failure counted. The Nebius benchmark shows what that discipline reveals, because it ran for the full eight hours and counted every request that failed or timed out.”
Putting It to the Test: Real Agent Load, Sustained for Eight Hours
WEKA and Nebius recently applied the scorecard’s rules in a joint benchmark on NVIDIA HGX B300 systems and reported results on several measures. The test ran on one eight-GPU node serving DeepSeek-V4-Pro to 512 active sessions drawn from a corpus of more than 5,000 real agentic coding sessions over eight hours per configuration, counting every failed and timed-out request.
Using WEKA NeuralMesh with Augmented Memory Grid, which pools NVMe across the cluster as a shared context tier, the node produced 3.7x the output tokens per GPU-hour compared with HBM-only serving on the same GPUs, and 97% of successfully completed requests reached a first token within five seconds, up from 44%. On HBM-only serving, 60% of prefill work was recomputed from context the cluster had already processed earlier in the session; with Augmented Memory Grid, 6.8%. At the measured rate, the HBM-only node would need about 30 hours to produce what Augmented Memory Grid produced in eight. The full benchmark is published at weka.io and nebius.com blogs.
“Peak throughput shows what a system can do for a few minutes, but our customers run agents for hours, days, or longer, so what matters is sustained useful work,” said Narek Tatevosyan, Vice President of Product at Nebius. “Over a full shift, the HBM-only configuration steadily lost ground as context fell out of memory and timeouts climbed, while WEKA NeuralMesh kept first-token latency flat the entire run. That changes how we plan capacity: we size for finished work over a working day, not a benchmark peak, and the Inference Economics Scorecard gives us a common way to show customers that number.”
“Agentic inference is the workload that defines the AI factory, and it puts extraordinary demand on memory and context. The question at scale is no longer how much context fits on the GPU, but how far it can be extended across the cluster without being recomputed,” said Jason Hardy, VP, Storage Technology at NVIDIA. “The WEKA and Nebius results on NVIDIA HGX B300 systems show what a shared context tier can do for sustained throughput and responsiveness, and as inference grows, buyers will need a clearer way to compare systems on the work they actually finish.”
Reproducible by Design: Open Traces and Tooling
The traces and tooling behind the benchmark are open-sourced so anyone can run the same workload on their own systems and compare results on equal terms. Because the traces come from real agentic coding sessions rather than synthetic load, and include the varied timing and complexity of real work, they give any team a realistic stand-in for production. Others are already building on them: SemiAnalysis’s AgentX benchmark is built on the traces, and NVIDIA AIPerf benchmarking tool ships a loader for them.
“A benchmark for agentic inference must reflect real-world production traffic, including long-running agents, growing context, complex multi-agent workflows with subagents, and the bursty, unpredictable traffic patterns that characterize real-world deployments,” said Dylan Patel, founder and CEO at SemiAnalysis. “Too few benchmarks meet that bar, which is the gap we built AgentX to close, and it’s why AgentX runs on real-world agentic traces using WEKA’s open kvlab trace format, so anyone can replay the same multi-turn workflows and compare results on equal terms. As agents become the dominant inference workload, reproducible measurement like this is how the industry gets past peak numbers and sees what its infrastructure can sustain.”
SemiAnalysis plans to share AgentX results on WEKA NeuralMesh as testing progresses. For more information, visit https://inferencex.semianalysis.com/agentx.
Additional Resources
The Inference Economics Scorecard and NAND Research companion report are available HERE.Download a complimentary copy of the 451 Research by S&P Global report, “From inference to agentic AI” HERE. Read the blog post from WEKA CEO Liran Zvibel. To learn about the engineering behind the capture tooling and how the measurements were defined, visit HERE.Visit WEKA at SC26 in Chicago, November 16 to 20, booth #2519 to learn more about how to apply the Inference Economics Scorecard to your own inference workloads.
About WEKA
WEKA is the AI data and memory infrastructure company transforming the economics of agentic AI. The WEKA NeuralMesh platform unifies high-performance data storage with extended GPU memory, giving enterprises, AI cloud providers, and AI builders a single foundation for training, inference, and agentic workloads. With Augmented Memory Grid, NeuralMesh extends GPU memory capacity by 1000x, accelerates time to first token by up to 20x, and delivers 10x more concurrent users from the same GPU footprint, as proven in production benchmarks. WEKA enables organizations to scale production AI faster, get more useful work from every GPU, and reduce the cost of every token served. Learn more at www.weka.io or connect with us on LinkedIn and X.
WEKA and the WEKA logo are registered trademarks, and WEKA NeuralMesh and Augmented Memory Grid are trademarks of WekaIO, Inc. Other trademarks mentioned herein are the property of their respective owners.
View original content to download multimedia:https://www.prnewswire.com/in/news-releases/new-research-challenges-gpu-utilization-as-measure-of-inference-infrastructure-302939612.html
You may like
Return to loan growth, with net loans up 1.3% QoQ in real terms and accelerating to 3.7% MoM in September
NPL ratio improved to 5.4%, approximately 200 bps below industry, with net cost of risk down 100 bps to 4.6%
Lower cost to serve, with personnel expenses excluding severance down 14% QoQ and administrative expenses down approximately 10%
BUENOS AIRES, Argentina, Nov. 16, 2026 /PRNewswire/ — Grupo Supervielle S.A. (NYSE: SUPV; BYMA: SUPV), (“Supervielle” or the “Company”) a universal financial services group headquartered in Argentina with a nationwide presence, today reported results for the three- and nine-month period ended September 30, 2026.
Starting 1Q20, the Company began reporting results applying Hyperinflation Accounting, in accordance with IFRS rule IAS 29 (“IAS 29”) as established by the Central Bank.
Commenting on third quarter 2026 results, Patricio Supervielle, Grupo Supervielle’s Chairman & CEO, noted: “The third quarter marked a return to loan growth as we began to deliver on the priority we outlined last quarter: selectively expanding lending in strategic products and sectors with stronger credit profiles while maintaining disciplined risk management and a clear focus on profitability. Our net loan portfolio grew 1.3% sequentially in real terms, with momentum strengthening in September, when loans increased 3.7%, and 2% in October.
Asset quality continued to improve. Our NPL ratio edged down to 5.4%, remaining 200 basis points below the industry, while net cost of risk declined by 100 basis points sequentially to 4.6% and loan loss provisions fell 14.5%. Although conditions remain challenging, particularly for some SMEs, these results reinforce our confidence in our approach to managing credit risk and rebuilding lending momentum.
On the funding side, we actively managed our deposit base as interest rates increased, reducing high-cost institutional peso deposits (as interest rates increase) to better align funding with our lending needs and contain interest expense, while dollar deposits continued to grow.
Net interest margin decreased to 16.0%, while the year-to-date margin stood at 18.0%, within our full-year guidance range. As anticipated, the temporary benefits we highlighted last quarter unwound as asset yields repriced, following the prior quarter decline in funding costs. Lower inflation reduced income from inflation-linked loans, and lower investment securities gains also contributed to margin compression. Renewed loan growth is an important step toward strengthening our earnings capacity, with its contribution is expected to build as origination volumes recover. We reported net income of AR$6.8 billion, equivalent to 2.2% ROAE. Excluding extraordinary severance charges associated with the completion of our rightsizing plan, adjusted net income was AR$10.7 billion, equivalent to 3.4% adjusted ROAE.
The transformation of our operating model is translating into a lower cost to serve. Personnel expenses excluding severance charges declined 14% sequentially, while administrative expenses fell approximately 10%, reflecting savings from branch consolidation and a leaner operating structure. Although lower revenues weighed on efficiency during the quarter, the year-to-date efficiency ratio excluding extraordinary severance charges stood at 55.5%. Building revenues on this lower cost base remains central to improving operating leverage and sustainable profitability.
On November 5, we marked an important milestone in Supervielle’s transformation, unveiling a new brand identity that brings to life the integrated financial ecosystem we have been building over the past two years. Our renewed SuperApp connects everyday banking, family finances and investing through “SuperFamilia”, closer integration with IOL Inversiones and AI-assisted customer service needs within a simpler, more personalized experience. We also launched “Pollen X”, our Banking as a Service platform, enabling companies, fintechs and digital platforms to embed our financial services into their own offerings. Together, these capabilities strengthen our role in customers’ everyday financial lives and open new opportunities for growth through a broader range of channels.
At IOL, assets under custody increased 20% sequentially in U.S. dollar terms to US$3.5 billion, while total accounts reached 2.5 million. Assets under management in mutual funds managed by IOL also continued to grow, reaching US$ 492 million at quarter-end. This growth underscores the opportunity to connect our banking and investment capabilities more closely and support a broader range of our customers’ financial needs.
Looking ahead, our priorities remain clear: sustain the improvement in asset quality, build on the return to lending growth and continue our permanent agenda of lowering the cost to serve into stronger, sustainable returns. Argentina’s low level of financial penetration presents a significant long-term opportunity. With a solid CET1 ratio of 13.5%, we are well positioned to pursue it through disciplined risk management, close customer relationships and a more efficient operating model. As we mark ten years as a publicly listed company, we look forward to sharing our Five-Year vision and strategic priorities at our Supervielle Day on November 23,” concluded Mr. Supervielle.
Third quarter 2026 Highlights
The Company reported an Attributable Net Income of AR$.6.8 billion in 3Q26, compared to Ps.13.6 billion in 2Q26 and an Attributable Net Loss of AR$.67.0 billion in 3Q25. In 9M26, the Company posted an Attributable Net Income of Ps.1.1 billion, compared to an Attributable Net Loss of Ps.35.9 billion in 9M25. Group headcount was 16% below year-end 2025 levels at quarter-end, following the headcount rightsizing plan implemented in 1H26 across its Banking ecosystem to align its workforce with a more efficient distribution model, as a growing share of customer activity migrates to digital and virtual hub service channels. Excluding severance charges, which in 3Q26 were limited to residual charges related to the headcount rightsizing plan implemented in 1H26, the Company posted an Adjusted Net Income of AR$.10.7 billion in 3Q26 and AR$.56.5 billion in 9M26.
During 3Q26, the structural improvements implemented in recent quarters became increasingly visible in the Company’s efficiency metrics, with a lower cost base driving the cost-to-assets ratio to 7.9%, its lowest level in the last five quarters. Credit costs also continued to decline, reflecting the sustained impact of portfolio management, collection and refinancing initiatives undertaken since 2025, together with a disciplined and cautious origination strategy. These trends translated into improved asset quality indicators, with the NPL ratio improving to 5.4% and net cost of risk easing to 4.6%.
However, these improvements were more than offset by lower Net Financial Income, reflecting the anticipated unwinding of the temporary spread benefits observed in the previous quarter, together with lower returns on UVA loans and CER-adjusted securities as inflation slowed. As a result, Net Income declined sequentially. In 9M26, results continued to reflect a transitional period, with underlying cost and credit trends improving.
3Q26 ROAE was 2.2% while adjusted ROAE was 3.4%, and Structural ROAE, which reflects the full run-rate of the salary savings associated with the headcount reductions under the voluntary retirement program, as if those reductions had been fully in effect throughout the period, was 3.5%. ROAA was 0.3%. 9M26 ROAE was 0.1%, while adjusted ROAE was 6.1% and structural ROAE was 7.2%.
In 3Q26, the Company reported Profit Before Income Tax of AR$.12.2 billion, compared to a Profit Before Income Tax of AR$.20.4 billion in 2Q26 and a Loss Before Income Tax of AR$.108.3 billion in 3Q25. This result included approximately AR$.5.9 billion in residual severance charges, compared with AR$.38.1 billion in 2Q26. Excluding these charges, Adjusted Profit Before Income Tax, was AR$.18.1 billion in 3Q26.
The sequential decline primarily reflects a 22.4% QoQ decrease in Net Financial Income, to AR$.241.8 billion, as the lagged repricing of interest-earning assets to lower market rates and to lower inflation, compressed spreads following the exceptional margin expansion recorded in 2Q26. This was partially offset by a 26.5% QoQ reduction in personnel and administrative expenses, to AR$.141.5 billion (down 12.2% excluding severance charges in both quarters), and a 14.5% QoQ reduction in Loan Loss Provisions, which declined to AR$.61.5 billion, extending the downward trend from the AR$.133.9 billion peak reached in 4Q25. Additionally, the loss from the net monetary position (RECPAM) declined 1.8% QoQ as inflation moderated. Net Service Fee Income decreased 6.7% QoQ to AR$.53.2 billion, primarily reflecting lagged fee repricing, as account package repricing was only implemented in September, and lower credit card commissions amid lower loan volumes, together with a weaker performance in asset management fees, partially offset by a 4.1% increase in net fee income from the IOL brokerage business.
On a cumulative basis, the Company posted a Profit Before Income Tax of AR$.7.4 billion in 9M26, compared to a Loss Before Income Tax of AR$.73.7 billion in 9M25. Excluding extraordinary severance charges, Adjusted Profit Before Income Tax amounted to AR$.92.6 billion in 9M26.
3Q26 Net Financial Income totaled AR$.241.8 billion in 3Q26, decreasing 22.4% QoQ and increasing 45.0% YoY. The sequential decline reflects the unwinding of the temporary spread benefits observed in the previous quarter, as anticipated. In 2Q26, the sharp decline in market rates, with average TAMAR falling to 22.8% from 32.0% in 1Q26, was passed through quickly to funding costs, while asset yields adjusted with a lag, temporarily widening spreads. In 3Q26, average TAMAR edged up slightly to 23.5%, keeping funding costs broadly stable, while interest-earning assets continued to reprice to the lower rate levels reached in the prior quarter, compressing spreads, together with lower inflation, which reduced the accrual on inflation-linked (UVA) loans and CER-adjusted securities. YoY, the increase reflects the recovery from the exceptionally compressed margins of 3Q25, when the monetary tightening ahead of the mid-term elections drove the average TAMAR to 46.4%, alongside historically high reserve requirements. Net Financial Income plus the Result from exposure to changes in the purchasing power of the currency totaled AR$. 208.5 billion, decreasing 24.9% QoQ, but increasing 63.1% YoY.
In 9M26, Net Financial Margin amounted to AR$841.1 billion, increasing 16.4% from 9M25.
Net Interest Margin (NIM) was 16.0% in 3Q26, declining 430 bps QoQ from the exceptionally high level recorded in 2Q26, but 520 bps above YoY. As anticipated in the previous quarter, the sequential compression reflects the lagged repricing of interest-earning assets to the lower rate environment, after funding costs had adjusted faster in 2Q26, as well as lower inflation, which reduced the accrual on inflation-linked (UVA) loans and CER-adjusted securities. The impact was reflected across the balance sheet: Loan Portfolio NIM was 17.2%, down 360 bps QoQ, and Investment Portfolio NIM was 13.7%, down 410 bps QoQ. AR$ NIM was 19.4% in 3Q26, 480 bps lower QoQ but 770 bps higher YoY. In pesos, loan portfolio NIM fell 450 bps to 21.8%, as loan yields caught up with lower market rates, while the AR$ cost of funds remained broadly stable. The peso investment portfolio NIM contracted 560 bps to 14.4% reflecting lower yields on AR$ government securities although 1,250 bps above 3Q25 level.
YoY, the expansion in NIM reflects the normalization of funding costs from the exceptionally elevated levels of 3Q25, when the monetary tightening ahead of the mid-term elections drove average TAMAR to 46.4%, compared to 23.5% in 3Q26.
The Total NPL Ratio was 5.4% as of September 30, 2026, down from 5.5% as of June 30, 2026 and approximately 200 bps below the industry average. While the ratio continues to reflect the carry-over effects of credit stress from prior quarters in a challenging macroeconomic environment, portfolio dynamics continued to improve, with delinquency indicators declining consistently since March. The 90-day delinquency ratio was 4.4% at the end of 3Q26, 100 bps below the reported NPL ratio. This gap reflects exposures classified as non-performing under Central Bank criteria that were performing or less than 90 days past due according to the Bank’s records. The Retail NPL ratio declined to 8.6% from 9.6% in June, while the retail 90-day delinquency ratio improved to 7.1% from 7.9%.
Quarterly NPL formation declined for the third consecutive quarter and stood approximately 25% below the 4Q25 peak, with retail NPL formation down 44% from that peak.
Measured in constant pesos, the stock of retail loans more than 90 days past due declined every month since February 2026, ending the quarter 27% below its peak and 16% below June. This reflects improved collection and refinancing dynamics resulting from active portfolio management and a disciplined origination strategy in place since early 2025, with origination cohorts since 4Q25 continuing to perform meaningfully better. In contrast, the commercial portfolio showed some deterioration during the quarter, with higher delinquency and NPL formation, reflecting the lagged impact of the high interest rate environment in prior months on SME customers. While conditions remain challenging, the sustained improvement in retail and the stronger performance of recent origination cohorts reinforce our confidence in our credit risk management approach.
Loan Loss Provisions (LLPs) declined 14.5% QoQ to AR$61.5 billion in 3Q26. This reduction was consistent with easing delinquency trends throughout the quarter and reflects the early benefits of the collection and refinancing initiatives implemented since December 2025, together with disciplined risk-adjusted loan origination. The Coverage Ratio was 92.8% as of September 30, 2026, compared to 98.9% as of June 30, 2026, and 112.2% as of September 30, 2025. The decline reflects the contraction of the retail portfolio over recent quarters, where loans require higher provisioning. At the same time, NPL formation increased in the commercial portfolio, where loans require lower allowances relative to their outstanding balance under the IFRS 9 expected credit loss model. On an adjusted basis, considering loans to individuals more than 90 days past due at the Bank, the coverage ratio was 105%.
Efficiency Ratio was 61.3% in 3Q26, compared with 63.4% in 2Q26. The sequential improvement mainly reflects lower severance charges, which declined to approximately AR$5.9 billion in 3Q26 from AR$38.1 billion in 2Q26, with 3Q26 charges mainly reflecting residual costs related to the voluntary retirement program. Excluding these items, the adjusted efficiency ratio would have been 59.2% compared with 52.3% in 2Q26. The increase in the adjusted efficiency ratio reflects the 19.7% sequential decline in revenues, driven by lower net financial income as asset yields continued to reprice downward while funding costs remained broadly stable, compressing spreads. This more than offset the 9.2% decline in operating expenses excluding severance charges, reflecting continued cost discipline and the benefits of the structural efficiency initiatives implemented across the organization. In 9M26, the efficiency ratio was 64.6%. Excluding AR$ 85.2 billion of extraordinary severance costs, the 9M26 efficiency ratio would have been 55.5%.
The Loans to Deposits Ratio was 82.8% as of September 30, 2026, compared to 72.6% as of June 30, 2026, and 67.3% as of September 30, 2025. The QoQ increase mainly reflects the 11.6% decline in deposits during the quarter reflecting asset and liability management, combined with the 0.9% growth in the net loan portfolio.
Total Deposits were AR$ 5,587.6 billion at quarter-end, decreasing 11.6% QoQ and 17.2% YoY. The sequential decrease was driven by asset and liability management initiatives reducing high-cost institutional peso deposits and a seasonal decrease in savings accounts.
Total private sector deposits were AR$ 5,367.6 billion, decreasing 11.5% QoQ and 17.8% YoY. AR$ deposits totaled AR$ 3,676.0 billion, decreasing 15.9% QoQ and 17.8% YoY.
The QoQ performance of AR$ Deposits, down 15.9%, or AR$ 695.2 billion, primarily reflects lower institutional funding, together with the reversal of the seasonal increase in retail deposits recorded in the previous quarter. During the quarter, the Company continued to actively manage its funding mix, reducing higher-cost institutional time deposits as well as corporate and public sector deposits, and partially replaced them with alternative funding sources, mainly repo transactions. As a result, the blended AR$ and US$ cost of funds declined approximately 50 bps QoQ, while the AR$ cost of funds increased only 10 bps sequentially, significantly less thanthe 70 bps increase in average TAMAR, limiting the impact of higher market rates on funding costs.
Foreign currency deposits amounted to US$1.3 billion, increasing 1.5% QoQ and 0.9% YoY.
Total Assets were AR$9,114.6 billion as of September 30, 2026, decreasing 1.3% QoQ and 8.4% YoY. The sequential performance mainly reflected: i) a 9.3%, or AR$201.0 billion, decrease in the public sector investment portfolio and repo transactions reflecting asset and liability management decisions following the reduction in institutional deposits, partially offset by: ii) a 17.1%, or AR$43.1 billion, increase in the corporate notes investment portfolio. In addition, iii) the loan portfolio (net of provisions) increased 1.3%, or AR$ 58.1 billion, driven by 13.4% increase in US dollar–denominated loans partially offset by a 1.7% decline in peso-denominated loans, as the Company prioritized disciplined origination and attractive risk-adjusted returns over volume, while iv) cash and due from banks remained broadly stable declining 1.0%, or AR$ 14.1 billion, reflecting lower regulatory liquidity needs at quarter-end. The leverage ratio (Assets to Shareholders’ Equity) was 7.3x, decreasing 10 bps QoQ, from 7.4x as of June 30, 2026, and 60 bps YoY, from 7.9x as of September 30, 2025. The average leverage ratio in 3Q26 was 6.9x, stable compared with 2Q26 and 3Q25.
Total Loans amounted to AR$4,624.0 billion as of September 30, 2026, increasing 0.9% QoQ and 1.8% YoY and 155.5% since March 31, 2024. Loan growth since March 31, 2024, has significantly outpaced the industry’s 142% increase, while remaining in line with the industry on a YoY basis. Net loans amounted to AR$ 4,383.0 billion as of September 30, 2026, increasing 1.3% QoQ and 1.1% YoY.
The sequential performance was primarily driven by a 13.4% increase in U.S. dollar loans, mainly to corporate clients in export-oriented sectors, which continued to lead economic activity, while peso–denominated loans decreased 1.7% reflecting soft loan demand and a disciplined origination strategy focused on attractive risk-adjusted returns.
By segment, quarterly performance was driven by commercial loans, which increased 4.2%, while retail loans declined 5.1%. Lending activity gained momentum toward quarter-end, with total loans increasing 3.7% in September compared to August. Retail loan origination also picked up, with origination volume in September 20% higher than in August. On a YoY basis, commercial loans grew 24.6%, while retail loans declined 19.4%, reflecting the disciplined approach to retail origination maintained since early 2025.
Common Equity Tier 1 Ratio (CET1) was 13.5% as of September 30, 2026, declining from 14.2% in the prior quarter but remaining 30 basis points higher than a year earlier. The sequential decline primarily reflects higher RWA density, driven by an increase in loans and an increase in market risk associated with the expansion of the corporate notes portfolio.
View original content to download multimedia:https://www.prnewswire.com/news-releases/grupo-supervielle-reports-3q26-results-302941254.html
SOURCE Grupo Supervielle S.A.
Technology
QuiX Quantum Advances the Industrial Foundations for Fault-Tolerant Quantum Computing Through the Dutch-Taiwanese QSPAD Project
Published
48 minutes agoon
November 17, 2026By
With Taiwanese partners Artilux Inc. and its affiliate Brillink Inc., and supported by a TechBridge grant from the Netherlands Enterprise Agency (RVO) and Taiwan’s Industrial Development Administration (IDA), QuiX Quantum will develop room-temperature, chip-integrated single-photon detectors — removing the cryogenic bottleneck that constrains scalable, fault-tolerant photonic quantum computing.
ENSCHEDE, Netherlands and HSINCHU, Nov. 17, 2026 /PRNewswire/ — QuiX Quantum, a Dutch-German developer of universal photonic quantum computers, together with Taiwanese partners Artilux Inc. and its affiliate Brillink Inc., has been awarded funding under the TechBridge Taiwan Call 2026 – Photonics Integrated Circuits, a joint programme of the Netherlands Enterprise Agency (RVO) and Taiwan’s Industrial Development Administration (IDA). The 36-month project, QSPAD, will run from January 2027 to December 2029 to develop, integrate and demonstrate photonically integrated single-photon detectors that operate near room temperature.
For QuiX Quantum, QSPAD represents an important industrial building block for its long-term roadmap toward scalable, fault-tolerant quantum computing — removing one of the last hardware barriers to quantum systems that can be deployed beyond specialised laboratories.
Removing the Cryogenic Bottleneck in Photonic Quantum Computing
Today, the most advanced photonic quantum computers depend on superconducting nanowire single-photon detectors that operate only below about 2.5 kelvin — colder than deep space. This requirement binds an expensive, bulky cryogenic system to the heart of every machine and keeps the technology out of environments where such cooling is impractical.
QSPAD is designed to remove that constraint. The project will develop and demonstrate waveguide-based germanium-silicon (GeSi) single-photon avalanche diode (SPAD) arrays that operate at or near room temperature — down to –40 °C using conventional thermoelectric cooling — co-integrated with QuiX Quantum’s low-loss silicon nitride photonic integrated circuits. The detectors are CMOS-compatible and sensitive at the 1310/1550 nm telecom wavelength, and the detector and photon-source circuits are joined through an engineered optical coupling packaged in a compact Photonic Assembly Control Unit.
Eliminating cryogenic detection removes the single largest cost item from a photonic quantum system and opens applications that are structurally closed to cryogenic hardware, from commercial data centres to field-deployed quantum sensing, while supporting a route to European volume manufacturing for the underlying photonic circuits.
Supporting the Path Toward Fault-Tolerant Quantum Computing
QSPAD’s objectives align closely with QuiX Quantum’s development roadmap. Building on CARINA, the universal photonic quantum computing platform the company delivered in 2026, QuiX Quantum is advancing its next-generation Dedalo architecture, with loss-tolerant logical-qubit capabilities planned throughout 2028–2030 and commercially deployable fault-tolerant systems targeted beyond 2030.
Dedalo is designed around modular photonic hardware, quantum error correction, photon-loss protection, energy-efficient operation, and integration with classical HPC and data-center infrastructure. Achieving these objectives requires not only advances in quantum-system architecture but also the industrial manufacturing and integration of high-performance photonic components — including efficient, scalable single-photon detection.
By developing non-cryogenic, chip-integrated detectors across the 2027–2029 period, QSPAD helps translate the system-level requirements of this roadmap into concrete, manufacturable technology, supporting the industrial foundations on which increasingly scalable photonic quantum computers can be built.
A Netherlands–Taiwan Partnership Strengthening the Photonic Quantum Supply Chain
QSPAD pairs two ecosystems whose strengths are highly complementary. QuiX Quantum contributes its silicon nitride photonic platform, photon-source design and quantum-system integration expertise; Artilux contributes GeSi single-photon detector technology, CMOS-compatible fabrication at commercial volumes, while Brillink bridges the region’s world-class advanced packaging and testing supply chain. Neither partner could deliver QSPAD alone.
The collaboration also strengthens a resilient, internationally competitive quantum-photonics supply chain — combining Taiwan’s detector leadership with a European manufacturing pathway and deepening a long-term partnership between the Dutch and Taiwanese photonics ecosystems. QuiX Quantum and its partners look forward to beginning the joint work in early 2027.
“Room-temperature single-photon detection is one of the missing industrial building blocks on the path to deployable fault-tolerant quantum computers. With Artilux and Brillink, we are combining Taiwan’s leadership in GeSi detectors and advanced packaging and testing supply chain with our photonic-integration and system expertise to remove the cryogenic bottleneck — and we are proud to be building this bridge between the Dutch and Taiwanese photonics ecosystems,” said Stefan Hengesbach, CEO of QuiX Quantum.
“Together with our Dedalo roadmap toward logical-qubit demonstrations between 2028 and 2030, QSPAD helps lay the manufacturing and integration foundations for quantum systems that can operate beyond the laboratory. It is another step in turning European — and now European and Taiwanese — scientific excellence into scalable industrial capability.”
“This collaboration is an important step in translating our GeSi technology into a key building block for future quantum computing and network. Together with QuiX Quantum and Brillink, we look forward to realizing integrated, manufacturable solutions for next-generation quantum systems.” said Dr. Erik Chen, Co-founder & CEO, Artilux Inc.
“The future of photonic quantum computing depends not only on advances in individual devices but also on reliable integration and scalable packaging. At Brillink, we are committed to advancing photonic integration and packaging technologies that bridge innovative components with practical system-level solutions” said Dr. Fang Yen Hsiang, COO, Brillink Inc.
About QuiX Quantum
QuiX Quantum is a European photonic quantum computing company founded in Enschede, the Netherlands, in 2019. The company develops integrated photonic quantum computing hardware and describes its approach as full-stack and fabless, with systems designed for modularity, scalability, and compatibility with data center and HPC environments. QuiX Quantum has offices in the Netherlands and Germany and is developing universal photonic quantum computing systems based on its silicon nitride photonic technology. More information: www.quixquantum.com
About Artilux
Founded in 2014, Artilux is a global leader in GeSi photonics, delivering a CMOS-compatible platform engineered for the AI era. The company powers three pillars of intelligence: Cloud AI, through high-speed optical interconnects (Connect Enterprise), hybrid electro-photonic computing (Inception), and scalable quantum photonics (Odyssey); Edge AI, through low-power on-device processing (Inception) and device-level optical connectivity (Connect Consumer); and Physical AI, through wide-spectrum sensing (Aware) and 3D spatial imaging (Explore)—advancing sharper perception, faster connectivity, and scalable compute to enlighten the path from information to intelligence. For more information, please visit www.artilux.ai
About Brillink
Founded in 2025, Brillink delivers AI-driven optical solutions featuring proprietary 2D array surface-coupling technology. We offer end-to-end capabilities—from AI-assisted PIC design to optical engines integrating Micro LEDs and Meta-optics. As a comprehensive provider bridging chip development and system design, we are “Connecting Intelligence at Light Speed” to power the next generation of AI infrastructure. For more information, please visit www.brillink.ai
About TechBridge
TechBridge is an RVO programme, commissioned by the Dutch Ministry of Economic Affairs, that supports Dutch companies and knowledge institutions in developing key technologies with international partners. The Taiwan Call 2026 on Photonics Integrated Circuits is run jointly with Taiwan’s Industrial Development Administration (IDA), with each country funding its own partners.
Media contact:
QuiX Quantum
Artilux Inc.
Brillink Inc.
Communications@quixquantum.com
Angela Wang
Linda Wu
+886-910-030-255
+886-910-383-527
View original content to download multimedia:https://www.prnewswire.com/apac/news-releases/quix-quantum-advances-the-industrial-foundations-for-fault-tolerant-quantum-computing-through-the-dutchtaiwanese-qspad-project-302941060.html
SOURCE QuiX Quantum; Brillink Inc.; Artilux Inc.
Technology
Entrepreneur Universe Bright Group Reports Third Quarter 2026 Financial Results
Published
48 minutes agoon
November 17, 2026By
XI’AN, China, Nov. 16, 2026 /PRNewswire/ — Entrepreneur Universe Bright Group (“EUBG” or the “Company”), a Nevada corporation, reported its unaudited financial results for the three and nine months ended September 30, 2026. For more information, please refer to the Company’s Quarterly Report on Form 10-Q for the quarter ended September 30, 2026, filed with the U.S. Securities and Exchange Commission (the “SEC”) on November 16, 2026.
Third Quarter and Nine-Month 2026 Financial Results Summary
Revenue: $680,624 for the three months ended September 30, 2026, compared to $1,217,966 during the same period of 2025.Net Loss: The Company recorded a net loss of $221,730 for the three months ended September 30, 2026, compared to net income of $395,113 during the same period of 2025.Nine-Month Revenue: $2,164,581 for the nine months ended September 30, 2026, compared to $3,323,026 during the same period of 2025.Nine-Month Net Loss: The Company recorded a net loss of $178,346 for the nine months ended September 30, 2026, compared to net income of $1,001,450 during the same period of 2025.Cash and cash equivalents were approximately $8.94 million as of September 30, 2026.
Business Overview
EUBG provides digital marketing consultancy services through its indirect wholly-owned PRC subsidiary. The Company focuses on delivering marketing consulting and related services to start-up and small-sized enterprises in China, supporting brand development and e-commerce through online traffic generation and integrated advisory solutions. The Company also conducts commercial lending activities in Hong Kong through its indirect wholly-owned Hong Kong subsidiary.
During the third quarter of 2026, the Company continued to develop its digital marketing services, which generated revenue of $107,080 for the quarter. In addition, HengYing International Investment Limited, the Company’s indirect wholly-owned Hong Kong subsidiary, progressively commenced its Credit Services business following the renewal of its Money Lenders License in June 2026. Credit Services generated $7,516 in loan interest income during the quarter from a secured loan of approximately $2.04 million made to a single third-party borrower in September 2026.
Strategic Outlook
EUBG continues to focus on strengthening its consulting capabilities, improving operational efficiency, and enhancing long-term competitiveness.
The Company is also continuing to evaluate and develop strategic expansion opportunities, including selected fintech-related initiatives through its Hong Kong subsidiary, HengYing International Investment Limited.
In addition, on February 25, 2026, the Company effected a 1-for-10 reverse stock split.
Management Commentary
Mr. Guolin Tao, CEO of EUBG, stated:
“Our third quarter results reflected contracting demand in certain consulting service lines, while digital marketing services and our newly established Credit Services segment contributed revenue during the period. We will continue to manage our operations prudently, strengthen our service capabilities and pursue opportunities to expand our revenue base in line with our long-term business objectives.”
About Entrepreneur Universe Bright Group
Entrepreneur Universe Bright Group is a Nevada holding company that conducts its operations through its wholly-owned subsidiaries in Hong Kong and mainland China. The Company primarily engages in digital marketing consultation services in China and also conducts credit services in Hong Kong.
For more information, please visit: www.eubggroup.com
Safe Harbor Statement
This press release contains projections and “forward-looking statements” as defined by the Private Securities Litigation Reform Act of 1995 related to the Company’s business. Forward-looking statements include statements concerning plans, objectives, goals, strategies, future events or performance, and underlying assumptions and other statements that are not historical facts. When the Company uses words such as “may,” “will,” “intend,” “should,” “believe,” “expect,” “anticipate,” “project,” “estimate,” or similar expressions that do not relate solely to historical matters, it is making forward-looking statements.
Forward-looking statements are not guarantees of future performance and involve risks and uncertainties that may cause actual results to differ materially from those discussed in the forward-looking statements. These statements are subject to uncertainties and risks including, but not limited to, the following: the Company’s goals and strategies; future business development; financial condition and results of operations; product and service demand and acceptance; competition and pricing pressures; changes in technology; government regulations; the PRC government’s significant oversight and discretion over the conduct of the Company’s business and restrictions on the transfer of cash into and out of the PRC and Hong Kong; the Company’s ability to develop its credit services business and to collect on loans it makes; the Company’s reliance on a limited number of customers; fluctuations in economic and business conditions in China; and assumptions underlying or related to any of the foregoing and other risks contained in the Company’s filings with the SEC. Investors are cautioned not to place undue reliance on any forward-looking statements in this press release. Additional factors are discussed in the Company’s filings with the SEC, which are available for review at www.sec.gov. The Company undertakes no obligation to publicly revise these forward-looking statements to reflect subsequent events or circumstances.
View original content:https://www.prnewswire.com/news-releases/entrepreneur-universe-bright-group-reports-third-quarter-2026-financial-results-302941252.html
SOURCE Entrepreneur Universe Bright Group
Grupo Supervielle Reports 3Q26 Results
QuiX Quantum Advances the Industrial Foundations for Fault-Tolerant Quantum Computing Through the Dutch-Taiwanese QSPAD Project
Entrepreneur Universe Bright Group Reports Third Quarter 2026 Financial Results
Send Rakhi to UK swiftly with UK Gifts Portal
Whiteboard Series with NEAR | Ep: 45 Joel Thorstensson from ceramic.network
Nullmax and Renesas Forge Strategic Partnership to Deliver Competitive ADAS Solutions for the Global Market
Why You Should Build on #NEAR – Co-founder Illia Polosukhin at CV Labs
Whiteboard Series with NEAR | Ep: 45 Joel Thorstensson from ceramic.network
NEAR End of Year Town Hall 2021: The Open Web World, MetaBUILD 2 Hackathon and 2021 recap
Trending
-
Technology4 days agoDomoAI Launches InkoAI Agent – One Sentence to Ready-to-Post Video
-
Near Videos4 days agoIllia Polosukhin on NEAR’s answer to Big Tech AI
-
Coin Market5 days agoSouth Korea to permit crypto withdrawal freezes during hacking threats
-
Technology3 days agoInnovative Immunology Biotech Windward Bio Lists on Nasdaq
-
Coin Market4 days agoGemini’s quarterly retail trading volume sinks to 3-year low
-
Near Videos3 days agoIllia Polosukhin on NEAR as encrypted Ethereum
-
Coin Market4 days agoDOJ moves to seize $1.2B from Binance hack, disrupts FaceStealer scam
-
Technology2 days agoThe First Cartridge-Free Smart Waterless Diffuser Is Here. What Would You Fill It With?
