Summary Bullets:

• Chinese providers are catching up with global leaders in three areas: generative AI, operationalized solutions, and agentic AI, giving enterprises broader options to diversify their AI partners.

• However, gaps in go-to-market, professional services, and model development remain large – crucial areas to drive adoption and accelerate deployments.

AI demand continues to grow rapidly. GlobalData Market Opportunity Forecasts to 2030: Artificial Intelligence (published on June 30, 2026) forecasts the global AI market to increase at 48.2% five-year-CAGR to reach $1.3 trillion in 2030. This is also reflected by 88% of the 3,069 global enterprises interviewed by GlobalData, who increased their AI budget this year (source: GlobalData ICT Customer Insights Survey 2026. Asia-Pacific shows a similar trend with a 48.6% CAGR, accounting for 36.2% of the global AI market in 2026. While growth is reported across the stack from the GPU to the application layer, cloud-based enterprise platforms remain crucial to drive adoption by providing broad and flexible capabilities to develop, govern, and scale AI deployments. GlobalData’s recent updates on Cloud-Based Enterprise AI Platforms: Competitive Landscape Assessment (September 17, 2026) shows that global hyperscalers (Google and Amazon) continue to lead the market, followed closely by Microsoft and IBM. However, Chinese players including Alibaba, Baidu, Huawei, and Tencent are closing the gap. This is important especially in markets like Asia where these players already have a strong presence. This gives enterprises broader options to diversify their AI partners, but also attracts cost-sensitive customers through their cost-leadership strategy.

The Competitive Landscape Assessment report also analyzes vendors’ capabilities, competitive comparison, market assessment, market drivers, buying criteria, and vendor recommendations. It covers 10 vendors: Alibaba, Amazon, Baidu, Dataiku, DataRobot, Google, Huawei, IBM, Microsoft, and Tencent Cloud.

Gaps Getting Smaller

Chinese providers are catching up with global leaders in three areas: generative AI (GenAI), operationalized solutions, and agentic AI. In the report, Alibaba is rated Very Strong in GenAI, the same as Amazon and one level behind the leader, Google. Alibaba’s latest Qwen family lets users toggle reasoning on and off, supports over 200 languages and dialects, and includes the agent-focused Qwen3.7-Max. Meanwhile, Baidu, Huawei, and Tencent are rated Strong. Baidu’s ERNIE 5.1 and Qianfan platform combine in-house models with DeepSeek, Kimi, and GLM. Tencent’s Hunyuan 3D and video models have over 3 million Hugging Face downloads, and Huawei has open-sourced a version of Pangu as OpenPangu.

Operationalized solutions are the clearest convergence. Alibaba and Huawei are both rated Very Strong, matching Amazon and ahead of several others. Alibaba offers vertical portfolios in finance, education, transportation, and retail while Huawei supports over 400 prepackaged AI applications across 500 industrial scenarios. Rated Strong, Tencent’s Productivity Agent Suite spans more than 20 verticals. Tencent is also extending into physical AI with its Hy-Embodied models, Tairos platform, and Unitree partnership, although deployments remain early. Industry-specific solutions are becoming more important for enterprises exploring the technologies as well as when moving from PoC to production.

The gap in agentic AI is closing more slowly. Alibaba, Baidu, and Tencent are rated Strong while Huawei is Competitive. Alibaba offers AgentRun, AgentTeams, and Agent Sandbox. Alibaba also offers a Skills portal, which converts cloud capabilities into MCP-compatible formats. Tencent has ADP 4.0 which is compatible with OpenAI’s Agents SDK) while Huawei offers AgentArts and AgentSphere. This shows not only serious investment by these players, but also increasing initiatives to support open standards rather than build closed ecosystems. However, their development platforms and tooling are rather new and less proven compared to global players.

This growing momentum is underpinned by their vertical integration strategies with investment in infrastructure such as Huawei’s Ascend 950 chips, the 1,024-card Atlas 950 SuperPod, Alibaba’s Zhenwu M890, and Baidu’s Kunlun chips. This is also reflected in their financial performance. Huawei reported $7.5 billion in AI chip revenue, Alibaba claimed 11 consecutive quarters of triple-digit AI revenue growth and over 290,000 GenAI customers, and Tencent said its AI cloud revenue doubles every year.

Where Gaps Remain

The widest gap is go-to-market. Alibaba, Baidu, Huawei, and Tencent are all rated only Competitive. While they have a strong presence in China supported by various references, their global footprints are relatively small: Alibaba has 32 regions in 17 countries, Huawei 34 in 16, and Tencent 21 in 10. All four have a predominantly Asia-centric or China-centric customer base. Geopolitical headwinds are a structural constraint, particularly in North America and Western Europe.

Another key area is professional services. Rated Competitive, these players still lack professional and managed services as well as partner ecosystem, which is crucial in helping enterprises with deployments, but also lag in ethical and governance frameworks. IBM’s 21,000 data and AI consultants and Google’s ecosystem of over 220,000 GenAI-enabled partner consultants set the benchmark.

The third is model development. Chinese providers have end-to-end tooling, such as Alibaba’s PAI, Huawei’s ModelArts, and Tencent’s TI Platform, but the experience is less unified. Tencent’s tooling, for example, is fragmented across TI Platform, ADP, and TokenHub. Global players are also ahead in the governance and monitoring tools that enterprises need to move from PoC to production.

The post Chinese Cloud Giants Close Enterprise AI Platform Gap with Global Hyperscalers, but Execution Lags appeared first on IT Connection.

Summary Bullets:

• Concerns about missing the AI boat have many corporate executives pushing AI application development mandates.

• However, research indicates that far too few organizations are realizing the kind of expected dividends in terms of cost savings, revenue generation, productivity gains, and innovation. What can they do differently to make the most of their investments?

To say artificial intelligence (AI) has become ubiquitous is almost an understatement. Just under 90% of all enterprises today use AI to support at least one corporate function, according to consulting firm McKinsey and Company, with 56% of all organizations applying the technology across three or more discrete tasks. This pace of adoption is unprecedented. Stanford University’s 2026 AI Index Report says 53% of the global population had been using generative AI (GenAI) within three years of widespread availability. For comparison’s sake, it was 12 years before the personal computer reached 40% of the population.

The appeal is obvious: Executives see AI’s promise as an engine to accelerate cost savings and operational efficiencies while delivering data-driven analytics that produce better outcomes and rapid revenue expansion. On paper, the technology is a game-changer, the kind that has corporations pressing lines of business to quickly develop and deploy AI-driven applications to realize fast returns. But the reality, particularly given the nascent nature of the technology and the relatively limited experience of most organizations in developing and deploying it, presents a much more complicated reality.

Figures like the 95% AI project failure rate quoted by the Massachusetts Institute of Technology (MIT) in the research university’s 2025 study of 300 public AI implementations accounting for a $30 billion to $40 billion total investment raise widespread concerns about poor outcomes and an imminent AI bubble implosion. Broader studies show that while less drastic, early AI results have delivered less than stellar results. IBM Institute for Business Value (IBV) research reports that the 1,250 IT executives surveyed in mid-2026 had seen a return on their AI investment of 17%. The result: Nearly two-thirds of all AI projects fell short of corporate goals. Separate IBM research points to internal technical oversight and management issues often beyond IT’s control, which can cut into 20% of AI returns. These issues range from disjointed processes, erratic measurement, limited visibility, and high upfront costs.

The source of much of this friction is trying to apply AI project budgets across multiple departments where priorities and end goals may be wildly different. The simple guidance for enterprises, particularly those with limited AI experience and even lower success rates, is to allocate resources to smaller, more focused AI projects initially. The goals should closely align with the unit’s requirements, and should by their nature, be easier to monitor and measure and need fewer resources. Enterprises can leverage learnings from these individual engagements over time, which they can then scale for larger inter-departmental projects.

As with any cutting-edge technology investment, failures are bound to happen more frequently than any organizations want to see. The key is to strike the right balance of adequate resource allocation and manageable implementations with clearly established governance over monitoring, management, and overarching objectives.

The post Under Pressure to Demonstrate AI ROI, Too Often Organizations Come Up Short appeared first on IT Connection.

Summary Bullets:

• In August, Meta admitted liability regarding social media’s addictive qualities on children, reaching a $17.1 billion settlement with US states and the District of Columbia.

• While seen as a victory by critics, the settlement is insufficient to introduce the sweeping changes needed to have a measurable effect.

Just weeks after Meta, the parent company of social media giants Instagram and Facebook, promised to restrict access to some of the more controversial features of its platform to settle a multi-billion dollar lawsuit, California Governor Gavin Newsom signed a bill tightening social media controls for children. The California law, which goes into effect in 2028, mandates social media accounts for users under age 16 be private, limits algorithmic feeds for minors, masks children’s location data, and turns off overnight notifications for underage users. Unlike similar legislation passed by other states like Illinois, the California law allows parents to amend social media restrictions. Newsom also signed a bill that requires AI companies to reveal to minors whether they are communicating with a non-human entity. The legislation also prevents children from accessing risky content on subjects like self-harm.

The California law and the Meta settlement come at a time when governments across the US and around the world are imposing regulations on social media and AI companies. The goal is to protect minors from damaging content and addictive online platform characteristics that interfere with the mental health of minors. Governments ranging from Australia and Indonesia to the EU have enacted restrictions barring children and young teenagers from unlimited content access. In the US, government social media regulatory and legal activity is occurring at the state and local levels.

Social media platforms have been under fire for years as parents and researchers voice concerns about hazardous effects of social media on younger users. Research from both public and private sector entities finds that social media use can have wide-ranging repercussions on children’s mental health and brain development. Advisories from the US Surgeon General and the American Psychological Association also cite negative physical and social consequences to minors from sites like Instagram and Snapchat, including sleep deprivation and exposure to content promoting eating disorders.

While social media companies reject accusations that their platforms propagate addictive usage patterns, successful litigation and settlements favoring plaintiffs such as the August Meta resolution show they are on the defensive. As part of the resolution of that settlement, Meta agreed to set a two-hour daily platform limit for users under 18 on Facebook and Instagram. The company will also block overnight access to the online sites and eliminate some features such as ‘like’ counts that have negative effects on users’ mental health. Meta will pay roughly $12 billion initially to the states, with the remaining $5 billion disbursement dependent on TikTok, Google (YouTube’s parent), and Snap (SnapChat’s parent) also settling and agreeing to pay financial penalties and make similar feature access changes.

On paper, experts critical of the effects of social media use on children and teenagers welcome the feature changes. But while large-scale settlements and platform adaptations show at least implied admission of responsibility by platform vendors, it is not necessarily an indicator of future feature development plans or end user behavior. Regulatory restrictions imposed by governments such as Australia have had limited impact on user access as teenagers find ways to navigate around controls.

The post Despite Meta’s $17.1 Billion Lawsuit Settlement, Questions Linger About Reducing Negative Social Media Effects for Children appeared first on IT Connection.

Summary Bullets:

• Conversational AI is driving a major voice resurgence in customer service by improving the IP voice experience.

• Long call queues on account of human agents being thinly stretched across large call volumes have led to customer frustrations and poor CX, but modern versions of conversational AI are set to change that by taking away the frictions associated with voice.

It was believed that voice would diminish in value as digital channels started to emerge as an alternative to the voice channel. While messaging and chatbots offer cost and scale advantages, GlobalData’s research shows that voice continues to remain a vital channel for connecting with customers despite the frictions. Voice provides instant gratification and tends to be more straightforward to use as it does not involve navigating through messaging channels, websites and so and having to type messages. While the elderly demography tends to prefer voice due to greater familiarity, voice is seen to be popular even with the younger generation.

The modern versions of conversational AI are reinforcing the value of voice by taking away the frictions, leading to improved experience, at least for some use cases. Conversational AI is not new, but the older versions were highly programmed, rigid, and could only answer limited, non-interactive queries. Modern conversational AI relies on probabilistic models, making it highly interactive and capable of closely emulating human agents. This technology has given birth to virtual AI receptionists that handle calls, greet customers, carry out authentication, identify intent, and route calls. This eliminates long queues and grants customers 24/7 access. The value multiplies with real-time translation, allowing AI receptionists to speak multiple languages. This is a game-changer for diverse ethnic groups who do not speak English as a first language, especially when accessing critical local government public services where expressing needs in a native language is vital.

Driven by rapid tech advancements, conversational AI is expanding from basic receptionists into autonomous AI agents. These agents handle tasks like answering queries, booking appointments, and taking orders. To do this, they require robust technical foundations, integrating seamlessly with knowledge bases and CRMs (systems of record) to provide targeted responses. Linked to the right workflows, AI agents can check calendars to schedule medical visits or process restaurant orders. Combined with 24/7 availability and translation, this dramatically elevates the voice experience for customers. Furthermore, non-technical teams can easily build these bots in agent builder studios by simply describing the desired persona and goals, removing the need for complex coding. As 8×8 CEO Samuel Wilson noted at the company’s analyst summit this June, “The next billion voice numbers will go to AI agents.”

Despite the promise, architectural challenges remain. Latency is a primary hurdle; the traditional multi-step process (like speech-to-text) can create latency, leading to awkward pauses that disrupt conversational flow. Robust guardrails are also critical to stop AI from hallucinating false information, which carries severe financial and legal risks. Additionally, backend integrations can be clunky due to poorly documented APIs or failing connectors. Finally, the compute costs for generative (GenAI) and agentic AI remain high and can be prohibitive for many businesses. While agentic conversational AI is still in its developing stages, steady breakthroughs are addressing initial operational hurdles to gradually enhance the contact center voice experience. This ongoing progress will drive deeper adoption of the voice channel, counteracting previous assumptions that voice would decrease in value.

The quest for ever-more automation within customer contact solutions continues apace. There will always be scenarios where human-to-human conversations are the preferred and best option. But to enable companies to offer voice interaction at the scale that customers demand, AI will have to carry a significant share of the burden. GenAI- and agentic AI-powered conversational AI is fundamentally shifting the voice experience in contact centers and customer service.

The post A New Breed of Conversational AI is Leading to Voice Resurgence in Customer Service appeared first on IT Connection.

Summary Bullets:

• Fast-growing use of agentic AIs within organizations has triggered agentic orchestration/governance prioritization among platform providers

• Controversy remains over two distinct approaches to orchestration: control plane construct or orchestration frameworks

Enterprises deploying AI in 2026 are turning their attention from deployment of agentic AIs to the management of growing numbers of agents being released across organizations. Companies are struggling with how to manage the hundreds or thousands of individual agents built within their organizations–agents built by different teams, running on different platforms, with inconsistent security and governance. This is problematic, considering most organizations lack visibility into agent inventory, purpose, and authorization.

The practice of addressing agentic orchestration at the control plane platform layer is moving to the forefront of the conversation, spurred by the lack of visibility, security, and management associated with agentic sprawl. Control planes sit above the agent layer, governing, observing, and enforcing policy across agents regardless of where they originated. The advantage is in their ability to ensure identity and security enforcement and enable cross-vendor interoperability regardless of what framework they use for coordinating agents. This approach contrasts with orchestration frameworks which are features of agentic solutions that simply coordinate agents.

GlobalData’s research captures the scale of the agentic market, and therefore the urgency of the situation. The report “Market Opportunity Forecasts to 2029: Agentic AI” puts the global agentic AI market at a 50.6% CAGR for 2024–2029, reaching $45.4 billion by 2029, driven by enterprise demand for autonomous decision-making, multi-agent orchestration, and scalable cloud-native AI infrastructure. GlobalData reports that early adopters are even replacing traditional robotic process automation (RPA) with goal-driven, self-adapting agent systems — and that the shift from pilots to production-grade systems is accelerating.

Vendor Strategies

A control plane market is emerging, positioned as framework-agnostic. Leading AI and platform providers are announcing strategies and solutions to address this evolving branch of agent orchestration:

• IBM is positioning the next generation of watsonx Orchestrate as an agentic control plane for the multi-agent era. It supports IBM-native agents alongside LangGraph, Langflow, and agents built on the open A2A protocol, with consistent policy enforcement.

• Salesforce has built its orchestration strategy on MuleSoft’s Agent Fabric. This has been helped by its ability to consolidate multiple data sources into a single source following Salesforce’s Informatica acquisition last November. A trust and data security layer serves as the key component of its new Agent Fabric control plane.

• ServiceNow is featuring its AI Control Tower as the governance layer spanning every AI agent, model, and action running across the enterprise, regardless of which vendor built them. The company is repositioning from being a workflow automation vendor to an enterprise AI operating system, shored up by its recent acquisition of IT/OT security provider Armis, which leans heavily into its new AI Control Tower solution.

• Boomi’s control plane approach is addressed via the Boomi Enterprise Platform, which sits between disparate systems, agents, frontier models, and data sources. Boomi’s acquisition of Lunar.dev, AI/MCP gateway, plays heavily into its strategy as the prompt routing layer for governing MCP servers and access.

Yet controversy over how to govern the fast-growing agentic AI market segment remains. Some rival platform providers are taking a different tact and keeping agentic orchestration within the confines of their own platforms and product ecosystems. They are not positioned as supporting cross-vendor governance layers in the same way as competitors:

• Microsoft has been reshaping Copilot Studio from an agent-building tool into an agent governance layer. It describes the new governance features as having centralized policy enforcement, agent lifecycle oversight, and cross-ecosystem governance spanning Microsoft 365 and partner-built agents. However, Microsoft’s architecture is embedded in and distributed across its popular platforms, including Power Platform and Azure, versus a discrete, specific control plane layer that sits above disparate agents.

• Oracle OCI’s strategy for management and governance also currently bypasses a control plane architectural model and remains within the confines of its own ecosystem. OCI Enterprise AI embeds agentic orchestration natively as a feature across the Oracle technology layers rather than positioning a discrete governance layer above them. Enterprise AI’s three integration layers are: Enterprise AI Models, Enterprise AI Agents, and Enterprise AI Governance.

Summary
The concept of a control plane architecture construct is still being defined by the market. Vendors operating in the agentic orchestration space have varying opinions and product strategies. Pioneering activities and offerings suggest this type of AI operating system will quickly become the fundamental layer for agentic AI. Operational guardrails are critical for bringing to production environments that are built around ambitious agentic AI projects.

It is worth noting that players in this market segment have generally adopted or endorsed MCP and A2A as the underlying interoperability layer, serving as the common protocol layer, while the control planes above it remain proprietary and competitive. Therefore, much of the agentic AI battle will be won or lost according to who controls the management, orchestration, and governance of disparate agents across enterprise environments.

For more on this topic and other cloud trends including escalating cloud costs, please see Cloud Watch Q2 2026: Reassessing On-Demand Economics in the Era of Escalating Cloud Costs

The post Lack of AI Agent Oversight Brings Dueling Approaches appeared first on IT Connection.

Summary Bullets:

• The EU’s MSS 2 GHz proposal strengthens regulation, security, and competition, but implementation through 2027–2029 will be phased, not immediate.
• Small spectrum block sizes favor IoT, messaging, emergency services; large-bandwidth applications face bottlenecks.

When the European Commission unveiled its plan late last month to reassign the 2 GHz mobile satellite service (MSS) spectrum at EU-level, it initiated more than a regulatory proposal. With the dust now settled on the announcement, it is clear that this represents a geopolitical and commercial realignment. Signals in the market suggest the framework is firming up around core principles: sovereignty, security, restricted eligibility, spectrum caps, and wholesale access. But beneath those pillars lies a battlefield of interests that will define not just who wins licenses, but which services Europe values the most – IoT or in-flight broadband, messaging or full-fledged device-to-device (D2D) connectivity.

Implementation is no longer a reactive guesswork exercise; it is becoming a long game. While the proposal’s two-year incumbency extension mitigates the cliff edge in May 2027 (when existing licenses expire), it also reveals the commission’s understanding: that legislative processes, spectrum allocation, compliance criteria, and defining “EU provider” are unlikely to be resolved by that deadline. Toward end-2027 and into 2028, observers should expect spectrum awards, provisional licenses, and a cascade of certification, rollouts, and enforcement activities stretching well into 2029. This also coincides with ITU’s WRC-27 conference, in which key spectrum allocations – including MSS – and updates to radio regulations are expected to be finalized and will offer clarity on the international trajectory of spectrum allocations for MSS.

Commercial implications are now emerging with greater clarity. The planned 30 MHz paired allocation in 5 + 5 MHz blocks is modest; sufficient for low-bandwidth IoT, messaging, and emergency services, yet wholly inadequate for high-throughput, low-latency use cases such as in-flight broadband or widespread D2D applications. Telcos and satellite operators should be preparing for trade-offs: either focus on niche verticals where limited spectrum suffices, or scale via partnerships, capacity leasing, or wholesale models. The missing “IoT reservation” is not a bug; rather, it is a structural risk: IoT may get parceled out only where extra capacity exists, and only if stakeholders ensure selection criteria explicitly to protect it. D2D has won primacy as per the initial tone of the Commission in its press release.

Definitions will matter, especially in the case of the term “EU Provider,” which is morphing into a fulcrum for competition. The 2 GHz is currently occupied by two US firms: Viasat for European inflight connectivity, and EchoStar for IoT and mobile connectivity. Post-May 2027, such entities will need to operate through European-owned subsidiaries or JVs. AST SpaceMobile has already localized itself via Satellite Connect Europe and is expected to qualify. But expectation of strict control over decision-making, governance, and spectrum transfer will push non-EU-based providers – SpaceX (which owns EchoStar spectrum), Viasat, and Amazon Leo – to engineer complex JV structures or seek regulatory exemption. In the case of the latter, GlobalData expects both litigation and political pressure from incumbents based in the US, particularly via the US government or trade bodies, and especially within the incumbent Trump administration window. Retaliation against European satcos is also a possibility, though it is unlikely that the bloc will collectively budge.

For telcos, the strategic moment is now. Whether to bid directly, partner with incumbents, or secure wholesale access depends on scale, access to capital, regulatory risk appetite, and technical readiness. Device certification, security compliance, antenna deployment, and supply chain capacity remain gating factors. By late-2028, the winners will be those who have not only secured licenses or partnerships, but who align with the EU’s overriding narrative: resilience, sovereignty, and competition.

The post EU Proposal for MSS Spectrum Seeks to Balance Bloc’s Commercial and Sovereignty Aspirations appeared first on IT Connection.