Aussie Broadband H1FY25 Results: Business & Enterprise Growth Shines

Aussie Broadband H1FY25 Results: Business & Enterprise Growth Shines

Aussie Broadband (ABB) has released its interim results for the first half of the 2025 financial year, showcasing a period of steady growth and strategic expansion. While the residential NBN business continues its upward trajectory, the real story lies in the impressive growth of the business and enterprise segments.

Metric	Result	Significance
Overall Revenue	$588m (+7%)	Steady growth
EBIT	37% increase	Significant improvement
NBN Market Share	7.8% (target 10%)	Shows Aussie's growing market presence
Residential Revenue	$327m (+15%)	Core business remains strong
Residential Gross Margin	31% (slight increase)	Marginal improvement in profitability
Business Segment Revenue	$54m (+13%)	Rapid growth in business segment
Business Segment Customer Growth	~50% over 2 years	Strong customer acquisition
Enterprise & Government Revenue	$47m (+13%)	Expansion into higher-margin segments
NPATA	$22m (+35%)	Substantial increase in adjusted profit
Dividend	4cps (incl. 2.4cps special)	Returns to shareholders
Capital Expenditure	Increased to $75-80m	Higher investment for future growth
Strategic Focus	Shift beyond reselling	Investments in infrastructure and cloud

Key Financial Highlights:

• Revenue Growth:
  • Total revenue increased by 7% to $588 million.
  • Residential revenue rose by 15% to $327 million.
  • Business revenue grew by 13% to $54 million.
  • Enterprise and government revenue also increased by 13% to $47 million.
  • Wholesale revenue decreased by 22% to $56 million.
• Profitability:
  • EBITDA climbed by 16% to $62 million.
  • NPATA (Net Profit After Tax, adjusted for acquired intangibles) surged by 35% to $21 million.
  • NPAT(Net Profit After Tax) rose 6% to $12 million.
• Dividends:
  • The company paid a total dividend of 4 cents per share (cps), including a special dividend of 2.4 cps.
• Capital Expenditure:
  • Capital expenditure guidance increased from $55-60 million to $75-80 million.

Business Segment Performance: A Driving Force

• Business Segment Growth:
  • Revenue in the business segment, serving small and medium enterprises, saw a 13% increase.
  • Customer numbers have grown by approximately 50% over the past two years.
  • This segment's growth is significant as customers tend to adopt multiple products, suggesting potential for accelerated revenue growth in the future.
  • Although the gross margins shrunk marginally to 43%, the large customer growth is very positive.
• Enterprise and Government Segment:
  • This segment, targeting larger customers and utilizing Aussie Broadband's own fiber network, also experienced a 13% revenue increase.
  • Gross margins in this segment decreased to 49%.
  • This is a key area of growth as it means that ABB is moving away from being solely reliant on the NBN network.
• Residential NBN Business:
  • While the residential business saw a respectable 15% revenue growth, the focus is clearly shifting towards the higher-margin business and enterprise segments.
  • The NBN market share is planned to grow to 10% from the current 7.8%.
  • Gross margins in this segment increased marginally to 31%.

Financial Outcomes and Key Considerations:

• EBIT Growth:
  • The 37% increase in EBIT and improved group margins reflect the success of Aussie Broadband's strategy to expand its higher-margin business and enterprise segments.
  • The expected full year EBIT of 90 million dollars, and resulting EBIT multiple of less than 9, shows that the company is currently undervalued.
• Operating Cash Flow:
  • The weak operating cash flow, attributed to working capital requirements, is a point of concern that investors should monitor in future reports.
• Capital Expenditure Increase:
  • The increased capital expenditure guidance indicates Aussie Broadband's commitment to investing in its infrastructure, including fiber network expansion, new IP addresses, and internal cloud capabilities.
  • This investment is crucial for the company's long-term growth and its ambition to become more than just an NBN reseller.

Strategic Direction and Future Outlook:

• Investor Day:
  • Aussie Broadband will host an investor day next month, where further details of its strategic direction will be unveiled. This event will provide valuable insights into the company's future plans.
• Beyond Reselling:
  • The investments in fiber and cloud capabilities suggest that Aussie Broadband is actively working to diversify its services and move beyond its core NBN reselling business.

Conclusion:

Aussie Broadband's H1FY25 results highlight a company strategically positioned for growth, particularly in its business and enterprise segments. While the residential NBN business remains a stable foundation, the focus on higher-margin services and infrastructure investments indicates a clear vision for the future. 

Related

Telstras H1FY25 Results - Investors Lens

Telstra's Dual Strategy: Dividends Rise as Buybacks Begin

Telstra's Dual Strategy: Investor Lens

Telstra's first half FY25 results have delivered a strategic surprise that signals management's growing confidence in the company's financial strength. With a clean 6% increase in underlying earnings, Telstra has not only raised its dividend as expected, but also announced a significant share buyback program. This dual approach to shareholder returns marks a new chapter for a telecommunications giant that has dramatically transformed recently.

Segment Performance: Mobile Leads with World-Class Margins

 
Mobile dominance continues: 

The Mobile division delivered 4% earnings growth, exceeding analyst expectations despite only four months of price increases and disruption from the 3G network shutdown.

Why it matters: 

With Mobile (>40% market share) accounting for over 60% of total EBITDA and maintaining an extraordinary 47% margin (among the highest globally), this segment remains Telstra's crown jewel. The division's ability to grow postpaid subscribers by 48k while experiencing minimal disruption from the 3G shutdown demonstrates operational resilience.

InfraCo exceeds expectations: 

The fixed infrastructure business grew EBITDA by 7% compared to a 2% expectation, with margins expanding to an impressive 62%. 

Why it matters: 

InfraCo's outperformance indicates that Telstra's structural separation strategy is yielding results, creating focused business units that can optimise operations independently.

The fixed segment shows dramatic 

Improvement: Fixed EBITDA surged 74% year-over-year to $183 million, primarily uplift in CS&B  and its digitisation.

Why it matters: 

This historically challenged segment is finally showing signs of stabilisation after years of 

NBN-related headwinds, contributing to a more balanced business portfolio.

EBITDA Growth and Profitability: Transformation Bears Fruit

 
Total EBITDA up 6% to $4.28 Bn, with 6/7 business segments showing growth.
Why it matters: 

The broad-based growth across multiple segments demonstrates that Telstra's transformation isn't dependent on a single division, creating a more sustainable earnings profile.

Return on invested capital (ROIC) reached 8.0%, supported by strong cost performance. 

Why it matters: 

This metric indicates that Telstra is deploying capital more efficiently, a key factor for long-term shareholder value creation and a sign of improved management discipline.

Earnings per share grew 6% to 8.9 cents, supporting the dividend increase.
Why it matters: 

Consistent EPS growth provides the foundation for sustainable dividend increases without compromising the company's financial position.

Capital Allocation: A New Framework Emerges

Management shifts to cash EPS focus for capital allocation decisions, moving away from traditional accounting EPS.
Why it matters: 

This change acknowledges the reality that Telstra's cash flows exceed accounting profits due to non-cash expenses like depreciation on leases, spectrum, and fibre assets.

$750 million share buyback announced, commencing March 12, 2025.
Why it matters: 

While some interpret buybacks as a sign management lacks investment opportunities, they can also indicate management believes the stock is undervalued. With Telstra, the buyback likely signals that management sees the current share price as disconnected from the company's intrinsic value, particularly given the stable cash flows from core businesses.

Capital reallocation to growth areas, including $800 million of capex redirected to mobile over the next four years.
Why it matters: 

This targeted investment demonstrates that while management is returning capital to shareholders, they're simultaneously investing in Telstra's highest-return business to maintain network leadership.

Cash Flow and Debt Position: Room to Maneuver
Full-year free cash flow guidance of $3.0-3.4 billion implies a free cash flow yield of nearly 4.5%.
Why it matters: 

This strong cash flow generation supports both dividends and buybacks while providing flexibility for strategic investments.

Gearing ratio of 2.16x at FY24, with projections to decrease to 2.0x in FY25.
Why it matters: 

Telstra maintains headroom under both its own "comfort zone" debt ratio (1.5-2.0x) and rating agency thresholds, providing flexibility for additional capital returns without jeopardising its financial position.

Potential for $2.25 billion in cumulative share buybacks through to FY28 as a base case, according to a leading analyst
Why it matters: 

The multi-year buyback capacity indicates sustainable financial strength rather than a one-time return of capital.

Dividend Strategy: Sustainable Growth Ahead

Dividend increased to 9.5c from 9.0c per share, a 6% increase.
Why it matters: 

Dividend increases typically support share price appreciation by attracting income-focused investors and signalling management confidence in future cash flows.

 
Potential for dividend to reach mid-20c range by FY28, limited primarily by franking credit generation.
Why it matters: 

This long-term dividend growth trajectory provides investors with visibility into future returns, supporting the investment case for patient shareholders.

 
Dividends currently exceed EPS, but are supported by superior cash flows.
Why it matters: 

This highlights the importance of Telstra's shift to cash EPS for capital allocation decisions, ensuring dividend sustainability despite accounting metrics suggesting otherwise.

Asset Monetisation: Reducing Complexity and Information Asymmetry

Sale of Ventures business completed, with proceeds reflected in the first half cash balance.
Why it matters: 

Divesting non-core assets simplifies Telstra's story for investors, reducing the information asymmetry that often leads to valuation discounts for conglomerates.

 
Foxtel sale in progress, not yet reflected in cash balances.

Why it matters: Beyond the immediate cash proceeds, divesting media assets allows management to focus on core telecommunications infrastructure, where Telstra has sustainable competitive advantages.

 
Portfolio restructuring continues, with the NAS division identified as a focus area.
Why it matters: 

This ongoing portfolio optimisation process suggests management is continuously evaluating where Telstra can create the most value, rather than maintaining legacy businesses for size alone.

Future Growth Catalysts: Beyond the Core

 
Accenture's partnership to develop AI capabilities, with $700 million investment primarily focused on cost reduction rather than revenue growth.
Why it matters: 

This strategic initiative demonstrates management's willingness to invest in emerging technologies with clear ROI metrics, rather than pursuing technology for its own sake.

Intercity fibre rollout progressing, with seven routes under construction and the first two (Sydney-Canberra and Melbourne-Canberra) expected to be ready in late 2025.
Why it matters: 

These new routes position Telstra to capitalise on growing demand from hyperscalers and the AI industry, representing a strategic pivot toward higher-growth market segments.

Mobile ARPU expected to benefit from the full-period impact of price increases in 2025.
Why it matters: 

This pricing power demonstrates Telstra's dominant market position (>40% market share) and ability to translate network quality into financial returns.

 
Investment Thesis: 

A Transformed Telstra. Telstra has completed a remarkable transformation, evolving from a clumsy giant with unfocused international ambitions to a streamlined, efficient telecommunications infrastructure business. With industry-leading margins in mobile, growing infrastructure returns, and a disciplined approach to capital allocation, today's Telstra offers investors exposure to stable cash flows with modest growth potential. The combination of dividend growth and share buybacks creates a compelling total return proposition in an uncertain economic environment. While the stock may never generate the same level of excitement as high-growth tech names, its 4.5% free cash flow yield and clear path to higher dividends make it a worthy consideration for income-focused portfolios. 

As management continues to monetise non-core assets and reinvest in high-return businesses, Telstra appears well-positioned to deliver sustainable shareholder returns while maintaining the financial flexibility to respond to industry shifts. The upcoming investor day later this year is expected to provide further clarity on Telstra's strategy through 2030, potentially catalysing additional investor interest in this transformed telecommunications leader.

source: AFR, Telstra, WSJ, Reuters, Yahoo Finance

How DeepSeek's Challenge Actually Strengthened NVIDIA's Market Position

How DeepSeek's Challenge Actually Strengthened NVIDIA's Market Position

The emergence of DeepSeek, a Chinese AI startup, initially sent shockwaves through the market, raising fears of disruption to NVIDIA's dominance in the AI chip market. DeepSeek's claim of developing an advanced AI model with significantly lower computing costs triggered a sharp market correction, wiping out nearly $750 billion from NVIDIA's market capitalization in a week. However, this seemingly negative event has, paradoxically, strengthened NVIDIA's long-term market position.   

The Initial Shock and the Correction:

The initial market reaction was swift and decisive, with NVIDIA's stock price dropping by 6% while the Nasdaq remained relatively flat. This correction, while painful in the short term, has resulted in a more attractive valuation for NVIDIA. The forward P/E ratio has compressed from a lofty 40x six months ago to a more reasonable 32x. This reset makes NVIDIA more appealing to investors seeking a balance of growth and value.

Beyond the Headlines: Customer Spending Speaks Volumes:

While the market reacted to the DeepSeek news, the actions of NVIDIA's largest customers tell a different story. Amazon, Meta, and Alphabet, the very companies that would be most likely to benefit from a cheaper alternative to NVIDIA's hardware, have all announced massive increases in capital expenditure for AI infrastructure. Crucially, they've all reaffirmed their strong partnerships with NVIDIA. This demonstrates that despite the emergence of potentially competitive solutions, NVIDIA remains the preferred and trusted provider for these industry giants. Amazon's commitment of $100 billion in capital expenditure for 2025, with a significant portion earmarked for AI infrastructure, is a powerful testament to this reality.   

Financial Metrics Tell the Tale:

The financial impact of the DeepSeek event, coupled with continued customer spending, paints a compelling picture for NVIDIA:

• Valuation Reset: The stock price correction has created a more attractive entry point for investors.
• Sustained Growth: Despite the lower P/E ratio, NVIDIA's projected revenue growth of 53% significantly outpaces its mega-cap peers, who are expected to average around 12.2%.
• Unwavering Customer Commitment: The substantial investments in AI infrastructure by major tech companies underscore NVIDIA's indispensable role in their AI strategies.
• Dominant Market Share: Projected data center revenue of $113 Bn for the fiscal year further solidifies NVIDIA's market leadership.

Navigating the Blackwell Transition – A Short-Term Hurdle:

While the transition to the Blackwell chip family presents a potential short-term challenge, with some analysts predicting a slowdown in growth during this period, NVIDIA's CEO has stated that production is "in full steam." The anticipated high-volume shipments in the second half of the fiscal year should mitigate any negative impact and pave the way for continued growth. The lowered market expectations due to the DeepSeek news could actually work in NVIDIA's favour if they manage to exceed forecasts.

The DeepSeek Silver Lining:

The DeepSeek episode, rather than being a threat, has served as a valuable market correction, bringing NVIDIA's valuation down to more sustainable levels while simultaneously highlighting the company's fundamental strengths. It has reinforced the fact that even with advancements in AI model efficiency, the demand for high-performance computing infrastructure remains robust. The incident has also provided a stress test for NVIDIA, demonstrating the company's resilience and its deep entrenchment within the AI ecosystem.

Looking Ahead:

NVIDIA's future prospects remain bright. The company is poised to capitalize on the continued explosive growth of AI, driven by its technological leadership, strong customer relationships, and dominant market share. The DeepSeek challenge, while initially perceived as a threat, has ultimately served to strengthen NVIDIA's position, creating a more compelling investment case and solidifying its role as the backbone of the AI revolution.

AI in Telecommunications: Transforming the Industry

Artificial intelligence (AI) is revolutionising the telecommunications industry, enabling communications service providers (CSPs) to tackle growing network complexity, meet evolving customer expectations, and drive innovation. As highlighted by Manish Singh, CTO of Telecom Systems Business at Dell Technologies, AI is being strategically implemented across three critical domains: sovereign AI deployment, enhanced customer experience, and automated network operations. Additionally, an emerging theme, Edge AI for real-time applications, is gaining traction, offering CSPs new opportunities to deliver low-latency, high-performance services. Below, we explore each theme in detail, including specific use cases, pros and cons, real-world examples from global telecom operators, and considerations for total cost of ownership (TCO).   

1. Sovereign AI Deployment: Localising AI for Cultural and Regional Relevance

Overview

Sovereign AI involves developing and deploying AI systems within a nation or region, tailored to local contexts, languages, and cultural nuances. CSPs, with their deep regional presence and customer relationships, are uniquely positioned to lead by creating AI models that deliver personalised, culturally relevant experiences whilst ensuring data sovereignty.  

Use Cases

• Localised Conversational AI: Deploy AI-powered chatbots or virtual assistants that understand regional dialects, slang, and cultural preferences for tailored customer support and marketing.
• Sovereign AI Factories: Build AI infrastructure (e.g., data centres with GPU clusters) to train and deploy localised AI models, ensuring compliance with regional data regulations.   

Pros

• Enhanced Customer Engagement: "Localised AI improves interactions by offering culturally relevant responses, increasing satisfaction and loyalty."   
• Regulatory Compliance: Sovereign AI ensures adherence to data privacy laws (e.g., GDPR in Europe, CCPA in the US), reducing legal risks.   
• Revenue Opportunities: CSPs can offer sovereign AI-as-a-Service (AIaaS) to enterprises and governments, creating new income streams.
• Data Security: Keeping data within national borders mitigates risks of cross-border data breaches.   

Cons

• High Initial Investment: "Building sovereign AI infrastructure requires significant capital for data centres, GPUs, and skilled personnel."   
• Complexity in Scaling: Developing AI models for diverse regions increases complexity and costs.
• Limited Interoperability: Localised models may not integrate seamlessly with global AI ecosystems.
• Talent Shortage: Finding AI experts with regional expertise can be challenging, especially in emerging markets.

Global Telecom Examples

• SoftBank (Japan): SoftBank leverages AI-RAN to offer distributed GPU-as-a-Service (GPUaaS), supporting sovereign AI workloads with low latency, aligning with Japan’s focus on technology independence.
• Reliance Jio (India): Jio integrates AI with its 5G Standalone network to provide vernacular language chatbots, enhancing engagement in India’s diverse linguistic landscape.   
• BT (United Kingdom): BT’s managed SASE service uses AI to provide secure, localised network solutions for UK enterprises, ensuring compliance with UK data regulations.   

Total Cost of Ownership (TCO) Considerations

• Capital Expenditure (CapEx): High upfront costs for AI infrastructure (e.g., Dell PowerEdge servers with NVIDIA GPUs, $500,000–$2 million per data centre for small-to-medium setups).   
• Operational Expenditure (OpEx): Ongoing costs include energy (10–20% of TCO), cloud services, and AI model training/tuning ($50,000–$200,000 annually per model). Maintenance and licensing add 5–10%.
• Savings Potential: "Sovereign AI reduces reliance on third-party cloud providers, saving 10–15% on data processing costs. AIaaS can generate $1–$5 million in annual revenue per enterprise client."
• Break-even Period: 3–5 years, depending on scale and monetisation success.

2. Enhanced Customer Experience: Leveraging GenAI - Powered Personal Assistants

Overview

Generative AI (GenAI)-powered personal assistants, built on fine-tuned large language models (LLMs), transform customer interactions by providing 24/7 contextual support, predicting needs, and integrating with backend systems to drive revenue and satisfaction.  

Use Cases

• AI-Powered Chat Agents: Deploy GenAI chatbots to handle inquiries, recommend personalised data plans, and upsell services.
• Customer Sentiment Analysis: Use AI to analyse interactions (e.g., call centre data, social media) to predict churn and tailor retention strategies.   
• Integrated BSS Automation: Link AI assistants with business support systems (BSS) to automate plan changes, billing adjustments, and upgrades.

Pros

• Improved Customer Satisfaction: "AI chatbots reduce resolution times by up to 40% (e&’s Autonomous Store Experience)."
• Revenue Growth: Personalised recommendations boost sales conversions by 15–20%.
• Cost Efficiency: Automating support reduces call centre staffing needs, saving 15–20% on operational costs.
• Scalability: AI assistants handle thousands of simultaneous interactions.   

Cons

• Integration Challenges: "Linking AI with legacy BSS systems may delay deployment by 6–12 months."
• Data Privacy Risks: Handling sensitive customer data increases breach risks, necessitating robust security.
• Customer Resistance: Some customers prefer human agents, potentially leading to dissatisfaction.  
• Maintenance Costs: Continuous LLM training requires ongoing investment ($100,000–$500,000 annually for large deployments).

Global Telecom Examples

• SK Telecom (South Korea): With Dell Technologies, SK Telecom developed an AI-powered chat agent, reducing resolution time by 40% and improving customer effort scores by 35%.   
• AT&T (United States): AT&T’s AI chatbots streamline customer service, routing high-value prospects to sales teams and increasing conversions by 15%.   
• e& (United Arab Emirates): The e& Autonomous Store Experience (EASE) uses AI-powered cameras and LLMs, increasing digital channel adoption by 28%.   

Total Cost of Ownership (TCO) Considerations

• Capital Expenditure (CapEx): AI platforms (e.g., Dell AI Factory with NVIDIA GPUs) cost $200,000–$1 million. BSS integration adds $100,000–$500,000.
• Operational Expenditure (OpEx): Annual costs include cloud hosting ($50,000–$200,000), LLM training ($50,000–$150,000 per model), and security ($20,000–$100,000). Staff training adds 2–5%.
• Savings Potential: "Reduced call centre costs save $500,000–$2 million annually. Increased conversions add $1–$3 million in revenue."
• Break-even Period: 2–4 years, driven by cost savings and upselling.

3. Automated Network Operations: Building Autonomous Networks with AI

Overview

AI enables autonomous networks that enhance reliability, reduce costs, and improve performance through anomaly detection, predictive maintenance, and closed-loop automation, paving the way for “Dark NOC” operations with minimal human intervention.   

Use Cases

• Anomaly Detection: Real-time monitoring to identify unusual network patterns (e.g., traffic spikes, security threats).   
• Predictive Fault Detection: AI forecasts equipment failures for proactive maintenance.   
• Closed-Loop Automation: AI autonomously detects, diagnoses, and resolves issues.   
• Network Engineer CoPilot: AI-driven tools (e.g., Dell and Kinetica’s solution) analyse 5G core and RAN data to accelerate troubleshooting.   

Pros

• Enhanced Reliability: "Predictive maintenance reduces incidents by up to 35%, improving uptime."
• Cost Reduction: Automation lowers operational costs by 10–15% through reduced downtime and labour.
• Improved Performance: AI optimises spectral efficiency and resource allocation.
• Energy Efficiency: "AI reduces RAN energy consumption (73% of network energy) by enabling 'zero traffic, zero watts' operations."   

Cons

• High Complexity: Implementing AI in 5G and RAN environments requires advanced expertise.
• Data Overload: Processing petabytes of network data demands high-performance infrastructure.
• Security Risks: AI-driven automation introduces new vulnerabilities.   
• Resistance to Change: Engineers may resist AI tools, necessitating training.

Global Telecom Examples

• XL Axiata (Indonesia): With Ericsson, XL Axiata implemented AI-based Virtual Drive Testing, reducing site report generation time by 60%.   
• T-Mobile (United States): T-Mobile’s AI-RAN, developed with NVIDIA, Ericsson, and Nokia, optimises spectral efficiency and reduces energy consumption.   
• Vodafone (Global): Vodafone uses AI for predictive maintenance, reducing downtime by 20% and costs by 15%.   

Total Cost of Ownership (TCO) Considerations

• Capital Expenditure (CapEx): AI-ready servers (e.g., Dell PowerEdge XR8000) and GPUs cost $300,000–$1.5 million per site. 5G core/RAN integration adds $200,000–$800,000.
• Operational Expenditure (OpEx): Annual costs include energy (10–15%), software licenses ($50,000–$200,000), AI model maintenance ($100,000–$300,000), and cybersecurity (5–10%).
• Savings Potential: "Reduced downtime saves $1–$5 million annually. Energy efficiency cuts RAN costs by 10–20% ($500,000–$2 million)."
• Break-even Period: 3–5 years, depending on scale and automation level.

4. Edge AI for Real-Time Applications: Powering Low-Latency Services (Emerging Theme)

Overview

Edge AI involves deploying AI models at the network edge (e.g., base stations, edge data centres) to process data closer to the source, enabling real-time, low-latency applications. This emerging theme is critical for use cases like IoT, smart cities, autonomous vehicles, and immersive services (e.g., AR/VR). CSPs can leverage 5G and edge computing to deliver these services, creating new revenue streams and enhancing network efficiency.   

Use Cases

• IoT and Smart Cities: Deploy Edge AI to process data from IoT devices (e.g., smart meters, traffic sensors) in real time, optimising urban services like traffic management.
• Immersive Services: Use Edge AI to support AR/VR applications, such as virtual concerts or gaming, with ultra-low latency (<10ms).
• Private 5G Networks: Implement Edge AI in enterprise settings (e.g., factories, hospitals) to enable real-time analytics for automation and patient monitoring.   
• Content Delivery Optimisation: Use Edge AI to cache and process content (e.g., video streaming) at the edge, reducing backhaul traffic and improving user experience.   

Pros

• Ultra-Low Latency: "Edge AI reduces latency to <10ms, critical for real-time applications like autonomous vehicles and AR/VR."
• Bandwidth Efficiency: "Processing data at the edge reduces backhaul traffic by 20–30%, lowering network congestion and costs."
• New Revenue Streams: "CSPs can offer Edge AI services to enterprises (e.g., smart manufacturing) and municipalities, generating $1–$10 million per client annually."
• Energy Efficiency: Localised processing reduces data centre energy consumption by 10–15% compared to cloud-based AI.

Cons

• Infrastructure Costs: "Deploying edge nodes (e.g., micro data centres, AI-enabled base stations) requires significant investment."
• Scalability Challenges: Managing thousands of edge nodes increases operational complexity and maintenance costs.
• Security Risks: Edge devices are more vulnerable to physical and cyber threats, requiring advanced security measures.   
• Limited Compute Power: Edge hardware (e.g., NVIDIA Jetson, Intel Xeon) has lower processing capacity than cloud GPUs, limiting complex AI workloads.

Global Telecom Examples

• Verizon (United States): Verizon’s 5G Edge with AWS Wavelength uses Edge AI to support real-time applications like autonomous drones and AR/VR, reducing latency by 50% for enterprise clients.
• Deutsche Telekom (Germany): Deutsche Telekom’s Edge AI platform supports smart city initiatives, such as AI-powered traffic management in Berlin, improving congestion by 15%.
• China Mobile (China): China Mobile leverages Edge AI in its 5G network to enable real-time analytics for industrial IoT, increasing factory automation efficiency by 20%.
• Telefonica (Spain): Telefonica’s Edge AI solution for private 5G networks supports real-time patient monitoring in hospitals, reducing response times by 30%.

Total Cost of Ownership (TCO) Considerations

• Capital Expenditure (CapEx): Edge AI infrastructure (e.g., Dell PowerEdge XR servers, NVIDIA Jetson modules) costs $100,000–$500,000 per edge node, with large networks requiring hundreds of nodes ($10–$50 million total). 5G integration adds $200,000–$1 million per site.
• Operational Expenditure (OpEx): Annual costs include edge node maintenance ($20,000–$50,000 per node), energy (5–10% of TCO), AI model optimisation ($50,000–$150,000 per use case), and security ($10,000–$50,000 per node). Managing distributed nodes adds 5–10% to operational costs.
• Savings Potential: "Reduced backhaul traffic saves $500,000–$2 million annually for large CSPs. Energy efficiency cuts edge processing costs by 10–15% ($200,000–$1 million). Enterprise contracts generate $1–$10 million per client."
• Break-even Period: 3–6 years, depending on edge node density and monetisation. "CSPs must prioritise high-value use cases (e.g., smart cities, private 5G) to accelerate ROI."

Building the Future of Telecom: A Holistic Approach to AI

To fully leverage AI, CSPs must adopt a holistic approach that integrates AI architecture, data readiness, and secure infrastructure. Dell Technologies’ Dell AI for Telecom initiative, powered by the Dell AI Factory, provides a comprehensive ecosystem for deploying AI solutions across all four themes. This includes high-performance computing (e.g., PowerEdge servers with NVIDIA, Intel, or AMD chips) and partnerships with telecom-specific vendors like SK Telecom, Kinetica, and NVIDIA.  

Key Recommendations for CSPs

• "Start Now: Begin with high-impact use cases (e.g., AI chatbots, predictive maintenance, Edge AI for IoT) to build momentum and demonstrate ROI."
• Invest in Infrastructure: Deploy scalable, on-premises, and edge AI solutions to ensure data sovereignty, performance, and low latency.
• Prioritise Security: Implement zero-trust architecture and encryption to protect sensitive customer, network, and edge data.
• "Foster Partnerships: Collaborate with ecosystem partners (e.g., Dell, NVIDIA, Ericsson, AWS) to accelerate AI deployment and innovation."  
• Train Talent: Upskill teams to work alongside AI tools, ensuring smooth adoption across customer service, network operations, and edge applications.  

TCO Summary Across Themes

• Sovereign AI: High CapEx ($500,000–$2 million) but long-term revenue potential ($1–$5 million annually). Break-even in 3–5 years.
• Customer Experience: Moderate CapEx ($200,000–$1 million) with quick ROI from cost savings ($500,000–$2 million) and revenue ($1–$3 million). Break-even in 2–4 years.
• Automated Networks: High CapEx ($300,000–$1.5 million) but significant savings ($1–$5 million) and energy efficiency ($500,000–$2 million). Break-even in 3–5 years.
• Edge AI: High CapEx ($10–$50 million for large networks) but strong revenue potential ($1–$10 million per client) and bandwidth savings ($500,000–$2 million). Break-even in 3–6 years.

Conclusion

"AI is a transformative force in telecommunications, enabling CSPs to innovate, compete, and deliver value in a rapidly evolving landscape." By strategically focusing on sovereign AI deployment, enhanced customer experience, automated network operations, and the emerging Edge AI for real-time applications, telecom operators can achieve measurable results, from improved satisfaction to reduced costs. Global examples like SK Telecom, Verizon, Deutsche Telekom, and Vodafone demonstrate AI’s power, whilst Dell Technologies’ AI for Telecom initiative provides the tools and expertise to accelerate adoption. CSPs that act now, invest strategically, and embrace a holistic approach will gain a competitive edge, positioning themselves as leaders in the AI-native telecom era.

Deepseek's Architecture Adaptation of Export Controls

Deep Seek's GPU Infrastructure

• Initially acquired 10,000 GPUs in 2021
• Estimated to have grown to around 50,000 GPUs in total
• Used 2,000 H800 GPUs specifically for V3 model pre-training
• Share infrastructure with their quantitative trading fund operations

Initial Export Control Framework

• US government initially restricted two parameters:
  • Computing power (FLOPS)
  • Interconnect bandwidth between GPUs
• This two-factor restriction created an opportunity for optimisation

H800 GPU Restrictions and Adaptations

• H800 was China's version of the H100 GPU
• Two key restriction factors from the US government:
  • Chip compute (FLOPS)
  • Interconnect bandwidth
• H800 was designed with:
  • Full FLOPS capability (same as H100)
  • Restricted interconnect bandwidth
• Deep Seek developed specialized SM (Streaming Multiprocessor) scheduling techniques to work around interconnect limitations
• Managed to achieve full GPU utilisation despite interconnect restrictions

Export Control Evolution

1. First Phase:
   • Dual restrictions on FLOPS and interconnect
   • H800 was allowed in China with limited interconnect
2. Second Phase:
   • The government identified flaws in the dual-restriction approach
   • Simplified to focus only on FLOPS restrictions
   • H800 eventually banned completely in late 2023

H20 Architecture Adaptation

• Newer H20 chip designed specifically for the Chinese market:
• Has restricted FLOPS (to comply with controls)
• Improved memory bandwidth and capacity
• Maintained interconnect capabilities
• In some ways performs better than H100 on memory operations

Source: Gemini, Seekingalpha, Forrester, SemiAnalysis