Development

Amazon invests $10 billion in AI data centres

Amazon is investing $10 billion to build data centres in North Carolina, aiming to expand its AI and cloud infrastructure instead of outsourcing compute needs.

The initiative will create at least 500 high-skilled roles and includes support for education, broadband careers, and local development through a $150,000 community fund.

The company is also developing AI-powered humanoid robots for future delivery tasks, reportedly testing them in a newly constructed ‘humanoid park’ at its San Francisco office.

Although using third-party hardware for now, the long-term goal is to embed Amazon’s software into these robots, according to sources.

Experts say the investment underlines a growing concern: that only large firms can afford the infrastructure needed for cutting-edge AI.

‘It’s positive for growth but risks concentrating innovation in Big Tech’s hands,’ said Leo Fan, co-founder of Cysic, a blockchain-based AI firm. He argues that the shift could disincentivise smaller players and dampen broader AI progress.

Amazon is also rolling out Alexa+, a new generative AI-powered assistant, and has secured a licensing deal with The New York Times to integrate journalism, recipes, and sports content into Alexa and other AI products, further extending its ecosystem instead of relying solely on internal data.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

OpenAI hits 3 million business subscribers

OpenAI has added another 1 million paying business subscribers since February, bringing the total to 3 million across ChatGPT Enterprise, Team and Edu.

The milestone was shared during a company livestream and confirmed in interviews with outlets like CNBC.

Chief Operating Officer Brad Lightcap noted that the business tools are being adopted widely, even in regulated sectors like finance and healthcare.

He said growth among individual users has fuelled enterprise adoption instead of stalling it, highlighting a feedback loop between consumer and business uptake.

OpenAI launched ChatGPT Enterprise in August 2023, followed by Team in January 2024 and Edu in May 2024. Within a year of its first business product, the firm had already reached 1 million paying business users—a number that has now tripled.

Lightcap said AI is reshaping work across sectors—from student learning to patient care and public services—by increasing productivity instead of just automating tasks.

A separate PYMNTS Intelligence report found that 82% of workers using generative AI weekly believe it improves their output. OpenAI’s overall user base has reportedly reached 800 million people, with CEO Sam Altman claiming 10% of the global population now uses the company’s tools.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Google email will reply by using your voice

Google is building a next-generation email system that uses generative AI to reply to mundane messages in your own tone, according to DeepMind CEO Demis Hassabis.

Speaking at SXSW London, Hassabis said the system would handle everyday emails instead of requiring users to write repetitive responses themselves.

Hassabis called email ‘the thing I really want to get rid of,’ and joked he’d pay thousands each month for that luxury. He emphasised that while AI could help cure diseases or combat climate change, it should also solve smaller daily annoyances first—like managing inbox overload.

The upcoming feature aims to identify routine emails and draft replies that reflect the user’s writing style, potentially making decisions on simpler matters.

While details are still limited, the project remains under development and could debut as part of Google’s premium AI subscription model before reaching free-tier users.

Gmail already includes generative tools that adjust message tone, but the new system goes further—automating replies instead of just suggesting edits.

Hassabis also envisioned a universal AI assistant that protects users’ attention and supports digital well-being, offering personalised recommendations and taking care of routine digital tasks.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Ask Photos rollout stalls as Google tackles speed and UX

Google has temporarily paused the rollout of its Ask Photos feature in Google Photos, citing performance concerns.

The tool, powered by Gemini’s multimodal AI, was designed to enhance search by allowing users to find images using natural language—such as ‘pictures of my dog eating on the sofa’ or ‘me in a swimming pool.’

The feature began rolling out in September last year but has seen limited availability. Google Photos product manager Jamie Aspinall revealed the rollout had stalled at ‘very small numbers’ due to latency, quality and user experience issues.

He confirmed that an improved version is expected within two weeks, promising search speeds comparable to the app’s original functionality.

Users with access to Ask Photos can continue using it, though they may encounter slow response times. Those who prefer to wait can turn off the feature in the app’s settings until the update is released.

In the meantime, Google has improved its traditional photo search capabilities. You can now search for specific text in your images by enclosing keywords in quotation marks. This refined approach delivers more accurate results for users who prefer a faster and more direct search experience.

As part of its tenth-anniversary celebrations, Google Photos has also introduced new features, including an updated image editor with AI tools, the ability to share albums via QR codes, and broader access to formerly Pixel 9-exclusive tools like Reimagine and Auto Frame.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Salt Typhoon and Silk Typhoon reveal weaknesses

Recent revelations about Salt Typhoon and Silk Typhoon have exposed severe weaknesses in how organisations secure their networks.

These state-affiliated hacking groups have demonstrated that modern cyber threats come from well-resourced and coordinated actors instead of isolated individuals.

Salt Typhoon, responsible for one of the largest cyber intrusions into US infrastructure, exploited cloud network vulnerabilities targeting telecom giants like AT&T and Verizon, forcing companies to reassess their reliance on traditional private circuits.

Many firms continue to believe private circuits offer better protection simply because they are off the public internet. Some even add MACsec encryption for extra defence. However, MACsec’s ‘hop-by-hop’ design introduces new risks—data is repeatedly decrypted and re-encrypted at each routing point.

Every one of these hops becomes a possible target for attackers, who can intercept, manipulate, or exfiltrate data without detection, especially when third-party infrastructure is involved.

Beyond its security limitations, MACsec presents high operational complexity and cost, making it unsuitable for today’s cloud-first environments. In contrast, solutions like Internet Protocol Security (IPSec) offer simpler, end-to-end encryption.

Although not perfect in cloud settings, IPSec can be enhanced through parallel connections or expert guidance. The Cybersecurity and Infrastructure Security Agency (CISA) urges organisations to prioritise complete encryption of all data in transit, regardless of the underlying network.

Silk Typhoon has further amplified concerns by exploiting privileged credentials and cloud APIs to infiltrate both on-premise and cloud systems. These actors use covert networks to maintain long-term access while remaining hidden.

As threats evolve, companies must adopt Zero Trust principles, strengthen identity controls, and closely monitor their cloud environments instead of relying on outdated security models.

Collaborating with cloud security experts can help shut down exposure risks and protect sensitive data from sophisticated and persistent threats.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

HMRC got targeted in a £47 million UK fraud

A phishing scheme run by organised crime groups cost the UK government £47 million, according to officials from His Majesty’s Revenue and Customs.

Criminals posed as taxpayers to claim payments using fake or hijacked credentials. Rather than a cyberattack, the operation relied on impersonation and did not involve the theft of taxpayer data.

Angela MacDonald, HMRC’s deputy chief executive, confirmed to Parliament’s Treasury Committee that the fraud took place in 2024. The stolen funds were taken through three separate payments, though HMRC managed to block an additional £1.9 million attempt.

Officials began a cross-border criminal investigation soon after discovering the scam, which has led to arrests.

Around 100,000 PAYE accounts — typically used by employers for employee tax and national insurance payments — were either created fraudulently or accessed illegally.

Banks were also targeted through the use of HMRC-linked identity information. Customers first flagged the issue when they noticed unusual activity.

HMRC has shut down the fake accounts and removed false data as part of its response. John-Paul Marks, HMRC’s chief executive, assured the committee that the incident is now under control and contained. ‘That is a lot of money and unacceptable,’ MacDonald told MPs.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Reinforcement learning enables robot to master badminton

A Swiss-led team at ETH Zurich has developed an AI-powered legged robot capable of playing badminton against human opponents with impressive precision and agility.

The project uses reinforcement learning, a type of AI that enables the robot to refine its movements and decisions through repeated trial and error.

The robot can accurately track the shuttlecock, predict its trajectory, and position itself effectively to return shots during high-speed rallies. Its ability to navigate the court and respond in real-time demonstrates significant progress in applying AI to dynamic, physical tasks.

Lead researcher Yuntao Ma said the project highlights the potential for AI to drive legged robots in increasingly complex activities.

The work represents a step forward in developing autonomous and intelligent robotic systems, including future humanoids capable of interacting in real-world environments.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Tebra introduces AI Note Assist to speed up clinical documentation

Tebra, a platform for independent healthcare providers, has launched AI Note Assist, an ambient documentation tool designed to streamline clinical note-taking significantly.

Integrated directly into Tebra’s electronic health record (EHR) system, the HIPAA-compliant solution promises to reduce the time spent on documentation by up to 50%.

The AI tool listens during patient consultations and automatically generates structured, editable clinical notes. It supports various formats, including SOAP notes, therapy progress notes, and psychiatric documentation, and even suggests ICD-10 codes based on conversation content.

All notes remain under the provider’s control, with the ability to review, edit and approve each entry to ensure clinical accuracy.

Tebra’s chief product and technology officer, Kyle Ryan, described the launch as a ‘significant milestone’ in supporting independent practices.

‘Our research shows that 82% of independent providers see more patient time as their key competitive advantage,’ said Ryan. ‘AI Note Assist helps them reclaim that advantage by reducing administrative workload, improving efficiency and minimising burnout.’

Clinically validated by practising healthcare professionals, the solution has been tested in real-world settings to ensure reliability and effectiveness. Tebra customers can use AI Note Assist immediately without requiring additional hardware or software installations.

By helping healthcare professionals cut down on documentation time while maintaining quality and compliance, Tebra aims to support providers in delivering more focused and personalised patient care.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Naver and Siam.AI launch Thai-language LLM and tourism agent

South Korean tech giant Naver is entering Thailand’s AI market through a strategic partnership with local technology firm Siam.AI Cloud.

The two companies have signed a memorandum of understanding to jointly develop a Thai-language large language model (LLM) and a tourism-specific AI agent, with the first outputs expected by the end of 2025.

The partnership aims to foster Thailand’s technological self-reliance, enabling the country to develop and run AI models domestically rather than relying on Western or Chinese technologies.

The move is a practical step toward achieving sovereign AI — a concept gaining momentum across Southeast Asia. One-third of Asia’s governments are expected to adopt sovereign cloud services by 2026, reflecting regional concerns about ‘digital colonialism’ and the need for local data control.

Thailand, and countries like Vietnam and Indonesia, are already implementing policies to retain data within national borders.

The Thai-Naver collaboration reflects a broader commitment to embedding AI into the healthcare, public services, and education sectors. The timing is strategic, with 77% of Thai businesses already deploying AI, especially in customer communication and acquisition.

The country’s AI market is projected to grow significantly through 2031, with machine learning, natural language processing, and computer vision leading the charge.

Beyond tourism, banking, manufacturing, retail, and agriculture sectors are primed for AI transformation.

For example, Thailand’s agricultural AI market is projected to grow from $80.3 million in 2023 to nearly $114 million by 2029, driven by initiatives like the AGROWTH platform that supports deep-tech agricultural innovation.

This partnership underscores a shift from importing AI tools to creating solutions tailored to national needs—backed by domestic infrastructure, local data, and region-specific expertise.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Meta inks 20-year nuclear deal to power AI expansion

Meta has entered a landmark 20-year agreement with Constellation to purchase 1.1 gigawatts of nuclear power from the Clinton Clean Energy Center in Illinois, starting in 2027.

The deal is designed to support the company’s rapidly growing AI infrastructure and data centres as energy demands surge across the tech industry.

Once facing closure due to financial losses, the Clinton plant’s future is secure — without relying on Illinois’ Zero Emission Credit programme. The agreement will keep over 1,100 local jobs, boost grid capacity by 30 megawatts, and generate an estimated $13.5 million in annual tax revenue.

Illinois lawmakers have praised the deal for its economic and environmental benefits, with Republican Regan Deering calling it ‘a forward-thinking investment.’

The partnership is part of Meta’s broader strategy to build a nuclear-powered AI ecosystem. With clean energy targets of 1 to 4 gigawatts, Meta has been negotiating with multiple nuclear providers and says further agreements are in the final stages.

According to the International Atomic Energy Agency, global data centre energy use is set to more than double by 2030 — potentially outstripping Japan’s entire electricity consumption. Meta alone plans to invest $65 billion in AI infrastructure in 2025.

The Clinton plant deal also serves as a hedge against the environmental impact of fossil fuels. A 2024 study by the Brattle Group estimated that closing the facility would have led to an additional 34 million metric tons of carbon emissions over two decades. It would also have dealt an annual $765 million blow to Illinois’ GDP.

Constellation, the plant’s operator, said consistent, carbon-free baseload power is essential for the AI-driven future. With its reliability and scale, nuclear energy is increasingly seen as critical to supporting always-on AI systems.

Meanwhile, Meta continues advancing its AI vision. The company plans to fully automate ad creation by late 2026, generating images, videos, and text tailored to user location and timing.

This automation effort has already boosted ad performance, with Q1 2025 results showing a 30% rise in AI-generated ad use, a 10% increase in average ad prices, and $42.31 billion in revenue — a 16% year-over-year jump.

However, the push for AI-generated content has unsettled the advertising industry. Firms like Omnicom Group have seen share prices dip over fears disrupting to traditional creative and production models.

Zuckerberg’s long-term AI vision includes automating marketing and enhancing user experience through AI companions and virtual therapists — part of Meta’s goal to integrate machine learning into everyday life while ensuring its platforms run on clean, scalable energy.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Diplo chatbot can make mistakes. Consider checking important information.