DeepSeek R2 launch stalled as CEO balks at progress, The Information reports
June 26 (Reuters) - Chinese AI startup DeepSeek has not yet determined the timing of the release of its R2 model as CEO Liang Wenfeng is not satisfied with its performance, The Information reported on Thursday, citing two people with knowledge of the situation.
R2, a successor to DeepSeek's wildly popular R1 reasoning model, was planned for release in May with goals to produce better coding and reason in languages beyond English, Reuters reported earlier this year.
Over the past several months, DeepSeek's engineers have been working to refine R2 until Liang gives the green light for release, according to The Information, opens new tab.
However, a fast adoption of R2 could be difficult due to a shortage of Nvidia server chips in China as a result of U.S. export regulations, the report said, citing employees of top Chinese cloud firms that offer DeepSeek's models to enterprise customers.
A potential surge in demand for R2 would overwhelm Chinese cloud providers, who need advanced Nvidia chips to run AI models, the report said.
DeepSeek did not immediately respond to a Reuters request for comment.
DeepSeek has been in touch with some Chinese cloud companies, providing them with technical specifications to guide their plans for hosting and distributing the model from their servers, the report said.
Among its cloud customers currently using R1, the majority are running the model with Nvidia's H20 chips, The Information said.
Fresh export curbs imposed by the Trump administration in April have prevented Nvidia from selling in the Chinese market its H20 chips - the only AI processors it could legally export to the country at the time.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Reuters
2 hours ago
- Reuters
Countries agree 10% increase for UN climate budget
BONN, June 26 (Reuters) - Countries agreed on Thursday to increase the U.N. climate body's budget by 10% for the next two years, a move the body welcomed as a commitment by governments to work together to address on climate change, with China's contribution rising. The deal, agreed by nearly 200 countries - from Japan to Saudi Arabia, to small island nations like Fiji - at U.N. climate negotiations in Bonn, comes despite major funding cuts at other U.N. agencies, triggered in part by the U.S. slashing its contributions, and political pushback on ambitious climate policies in European countries. Countries agreed to a core budget of 81.5 million euros for the United Nations Framework Convention on Climate Change (UNFCCC) over 2026-2027, up 10% from 2024-2025. The core budget is funded by government contributions. The deal includes an increase in China's contribution, reflecting the country's economic growth. China, the world's second-biggest economy, would cover 20% of the new budget, up from 15% previously. Only the United States, the world's biggest economy, was allocated a bigger share, of 22%. However, President Donald Trump quit the U.N. Paris climate agreement and halted international climate funding. Bloomberg Philanthropies has pledged to cover the U.S. contribution to the UNFCCC budget. The U.S. did not attend the U.N. climate talks this week in Bonn, Germany where the budget was approved. UN climate chief Simon Stiell welcomed the increase as "a clear signal that governments continue to see U.N.-convened climate cooperation as essential, even in difficult times." The UNFCCC runs annual climate negotiations among countries and helps implement deals that are made - including the 2015 Paris Agreement, which commits nearly all nations to limit global warming. The body has faced a severe budget shortfall in recent years, as major donors including China and the U.S. had not paid on time, prompting the body to cut costs including by cancelling some events. The UNFCC's running costs and headcount - its core budget funded 181 staff in 2025 - are smaller than some other U.N. bodies facing sharp funding cuts, such as the U.N. trade and development agency's roughly 400 staff. Meanwhile, the U.N. Secretariat, the global body's executive arm, is preparing to slash its $3.7 billion budget by 20%, according to an internal memo.
Finextra
2 hours ago
- Finextra
The infrastructure of trust: building AI foundations for inclusive, explainable finance: By Diederick Van Thiel
The time is now to focus on AI infrastructure, which will enable companies to scale AI and build a future where humans and multiple AI agents successfully work together. In this blog I share some insights on how we at AdviceRobo do this so you can learn from it and build your own infrastructure of trust with AI. As the world accelerates toward an AI-first economy, one truth is becoming inescapable: no digital transformation will succeed without the right infrastructure. For financial services, especially those addressing underserved markets, the stakes are even higher. AI must be both inclusive and explainable—capable of reaching those locked out of traditional credit systems, while remaining transparent and accountable. This is the future AdviceRobo has long been building toward: a future where infrastructure meets empathy, and where AI augments human decisioning without undermining human dignity. Rethinking risk in the age of AI Traditional credit systems have failed billions of people globally—those without formal income, credit history, or access to mainstream banking. AdviceRobo has pioneered the use of psychometric data, behavioral analytics, and alternative data to assess risk far beyond FICO scores. These methods have already shown substantial lift: AdviceRobo's research reveals a 20–30% increase in acceptance rates among thin-file customers, while reducing defaults with 20% through better predictive power. But even the most advanced algorithms are only as scalable as the systems supporting them. So I invite you to enter the AI infrastructure revolution! Why Infrastructure is the next growth frontier McKinsey estimates a $5 trillion investment is needed over the next five years to support the growing appetite of AI across industries. But this isn't just about GPUs or cooling systems—it's about how you build AI that earns trust, scales cost-effectively, and adapts to regulatory demands. As Rodrigo Liang of SambaNova puts it, 'You're going to see a tenfold increase in investment for inferencing… and if it's not efficient, it won't scale.' For AdviceRobo and neo-lenders, this means focusing on three foundational pillars: 1. Hybrid AI deployment models AdviceRobo's clients—credit bureaus, retailers, digital banks and credit platforms —often span multiple jurisdictions and compliance regimes. A hybrid AI model, combining cloud-based inference with on-premises secure learning, is no longer optional; it's table stakes. This flexibility allows financial institutions to deploy AdviceRobo's scoring agents locally, while retraining them globally—a balance between data sovereignty and model innovation. 2. Agentic AI for dynamic (credit) decisioning Agentic AI—the use of multiple autonomous, specialized AI agents working in coordinated workflows—is redefining real-time decision-making. Imagine a suite of AdviceRobo agents: One analyzing psychometric data, Another adjusting for behavioral shifts, A third scanning macroeconomic risks. Each agent contributes to a unified, explainable decision. And with response times as fast as 0.03 seconds, these agents feel instantaneous to both lenders and borrowers. This is the dawn of always-on, always-fair credit decisioning. 3. Explainability as core infrastructure AI in finance must be auditable, not a black box. AdviceRobo has led the change in integrating explainable AI (XAI)—highlighting which behavioral traits influenced a risk score, and why. We work with powerful models that can capture the complexities of today's world far more effectively than traditional logistic regression models. At the same time, we use tools like SHAP values and LIME to open up these 'black boxes' and ensure that our AI-driven decisions remain transparent and explainable. This capability isn't just about compliance. It's about empowerment. Borrowers can gain insights into how to improve their profiles, and lenders build trust with regulators and stakeholders. In my opinion: 'We have to cross this S-curve, where we have enough infrastructure that convinces us the models are behaving correctly and the outputs are being securely managed.' Toward inclusive finance at scale The next chapter for AdviceRobo lies not in building bigger models, but smarter infrastructure—systems that: Are multi-lingual and culturally contextual , , Run efficiently even in data-constrained environments, And prioritize inclusion, from UX to underwriting. The ambition? To AI-ify credit and democratize access to credit for the 1.7 billion people currently unbanked—not with brute-force computation, but with infrastructure designed for empathy, precision, and global scale. And we're just getting started. We're also developing AI agents tailored for the embedded finance space, supporting industries like insurance, telecommunications, and retail—each with its own unique challenges and opportunities. In insurance, our agents help underwrite policies using behavioral data, making coverage more accessible for underserved or high-risk segments. In telecom, AI-driven financial profiling can enable dynamic credit limits for prepaid-to-postpaid transitions, or personalized device financing offers. And in retail, our technology powers embedded lending at the point of sale—enabling instant credit decisions and personalized repayment options that drive conversion and loyalty. These use cases all build on AdviceRobo's core strengths: behavioral data science, explainable AI and api-based scalable infrastructure. This ensures our partners can scale inclusive financial services—seamlessly integrated into their customer journeys. Our models go far beyond traditional statistical methods, and our use of explainability tools like SHAP and LIME ensures that every prediction can be trusted, audited, and acted upon—whether you're approving a loan, hiring a candidate, or retaining a customer. Final word: infrastructure is the new differentiator In a world where AI is ubiquitous, the companies that win will be those who build infrastructure with purpose. AdviceRobo is already ahead of the curve—blending cutting-edge AI, ethical risk profiling, and scalable delivery – and there to help others to drive this transformation succesfully too. Because the future isn't just algorithmic. It's agentic, explainable, and radically inclusive.
Reuters
2 hours ago
- Reuters
Damaged unit of Iran's South Pars refinery back operating, Nour News says
June 26 (Reuters) - A damaged unit of the South Pars refinery's Phase 14, which was hit by Israel in its first strike on Iran's oil and gas sector, has returned to operation, the Iranian state-run agency Nour News reported on Thursday. South Pars, the world's biggest gas field, is located offshore in Iran's southern Bushehr province and is responsible for the lion's share of the country's gas production. Iran is the world's third largest gas producer after the United States and Russia. Gas production was not halted and the damaged unit was repaired in about 10 days, the agency added. On June 14, the Israeli strike caused a fire, which has been extinguished, in one of the four units of Phase 14. Israel launched an air offensive against Iran on June 13 killing commanders and scientists and bombing nuclear sites, saying it wanted to stop Tehran building an atomic weapon. After a 12-day air war between Israel and Iran, U.S. President Donald Trump announced on Monday a ceasefire between them.
