ICT industry trends 2027+

Summary: China dominates the mining and refining of gallium, germanium, antimony, graphite and rare earths and now uses export licences as a negotiating lever. Pauses granted in trade talks can be revoked at short notice, which makes mineral access a recurring supply shock for chip, telecom and electronics makers.

Current situation: China produces about 94% of the world's gallium and dominates refining of germanium, antimony, graphite and rare earths. In November 2025, after the Trump-Xi meeting in Busan, Beijing suspended until 27 November 2026 its ban on exporting gallium, germanium and antimony to the US and paused rare-earth controls announced in October 2025. The metals remain on China's dual-use list, so every shipment still needs a licence, and in January 2026 China imposed new dual-use controls on medium and heavy rare earths aimed at Japan.

Expected development: The suspension expires on 27 November 2026, which puts gallium, germanium and several rare earths back under US-specific licensing unless Beijing extends it. Through 2028, Western governments fund stockpiles, recycling and non-Chinese refining, while chipmakers dual-source materials and redesign to reduce exposure. Each new round of Western chip export controls risks a mineral counter-move.

Challenges: Building refining capacity outside China takes years and faces cost disadvantages and environmental permitting hurdles. Heavy rare earths face acute bottlenecks through 2027. Companies struggle to map their exposure beyond direct suppliers.

Time to impact: 0-2 years

Potential impact: Very high

STEEP segment: Political

This year's trend development: New

Summary: The US, Europe and Asia spend hundreds of billions to build domestic chip capacity, while leading-edge manufacturing remains concentrated in Taiwan. Europe's 20% market-share goal has proved out of reach, and policy shifts from subsidising fabs toward securing demand, emergency powers and selective dependencies.

Current situation: Europe's 2023 Chips Act mobilised more than EUR 52 billion in investment, yet the EU still produces less than 10% of global semiconductors and depends on Taiwan for more than 90% of advanced chips. An April 2026 audit found the 20% market-share target for 2030 will be missed by about half after Intel scrapped its planned German fabs. On 3 June 2026, the Commission proposed a Chips Act 2.0 that prioritises AI chips, speeds up permitting for strategic investments and gives Brussels emergency powers to reprioritise chip orders during shortages.

Expected development: Through 2030, sovereignty strategies narrow to defensible niches: Europe in equipment, power semiconductors and automotive chips, the US in leading-edge logic built by TSMC and Intel on American soil. Governments take equity stakes and use procurement to guarantee demand for domestic fabs. Supply security comes more from allied partnerships than from complete domestic supply chains.

Challenges: Subsidies cannot overcome shortages of skilled workers and anchor customers for new fabs. Budgets for the next EU funding period remain uncertain. Competing national programmes duplicate capacity and raise costs.

Time to impact: 4-6 years

Potential impact: Very high

STEEP segment: Political

This year's trend development: Unchanged

Summary: AI data centers are adding electricity demand faster than grids can connect it. Power availability, more than chips or capital, now decides where and how fast AI capacity gets built, and it shifts bargaining power to utilities, grid operators and regions with spare capacity.

Current situation: The IEA projects global data center electricity consumption to more than double to about 945 TWh by 2030, slightly more than Japan's total electricity use today. Morgan Stanley forecasts US data center demand could reach 74 GW by 2028, with a shortfall of about 49 GW in available power access. The IEA estimates that one-fifth of global data center investment is at risk of delay because of grid bottlenecks, and lead times for power transformers have stretched to about 128 weeks.

Expected development: Through 2028, utilities introduce large-load tariffs and connection rules that make data centers pay for grid upgrades, and flexible operation that curtails load during peaks becomes a condition for faster connection. Capacity shifts toward regions with surplus power, such as the Nordics, parts of the US Midwest and the Gulf. On-site generation and batteries fill the gap where grids cannot keep up.

Challenges: Grid expansion takes up to a decade because of permitting and supply chains for transformers and switchgear. Rising household electricity prices turn data center growth into a political issue. Demand forecasts vary widely, which risks overbuilding generation for loads that never materialise.

Time to impact: 0-2 years

Potential impact: Very high

STEEP segment: Ecological

This year's trend development: Upgraded

Summary: Chipmakers, cloud providers and AI labs increasingly finance each other's purchases through equity stakes, vendor financing and off-balance-sheet vehicles. The structure accelerates the build-out and concentrates risk, so a revenue shortfall at a few AI labs could cascade through suppliers, lenders and data center developers.

Current situation: Nvidia signed a letter of intent in September 2025 to invest up to USD 100 billion in OpenAI tied to deploying 10 GW of Nvidia systems, while OpenAI agreed a reported USD 300 billion cloud contract with Oracle and plans about USD 1.4 trillion of infrastructure spending. S&P Global warned that a third of Oracle's revenue could be tied to OpenAI by 2028, and a measure of Oracle's credit risk reached levels last seen in 2009. In August 2026, a gauge of Nvidia's own credit risk nearly doubled within three weeks over its customer financing commitments.

Expected development: Through 2027, more data center capacity is funded through private credit and special purpose vehicles that keep debt off the balance sheets of the main sponsors. If AI revenue keeps growing, the loop sustains the build-out; if it stalls, overbuilt capacity, idle GPUs and stressed lenders follow, a pattern analysts compare to the telecom fibre bust of 2001. Investors increasingly separate profitable hyperscalers from leveraged neoclouds and single-customer suppliers.

Challenges: Opaque structures make exposure hard to assess for investors and regulators. UBS notes that the largest deals are performance-linked and moderate relative to Nvidia's revenue, so risk concentrates in specific leveraged players. Rapid GPU depreciation shortens the window to earn a return on the assets being financed.

Time to impact: 0-2 years

Potential impact: Very high

STEEP segment: Economic

This year's trend development: New

Summary: Grid connection queues of up to seven years are pushing AI data centers to build their own power plants, mostly gas turbines. The ICT sector's decarbonisation narrative reverses as new fossil capacity is locked in for decades to serve compute demand.

Current situation: Global Energy Monitor counted 189 GW of gas-fired capacity in development in the US specifically to power data centers by mid-2026, almost double the 97 GW it identified at the end of 2025; in Texas alone, 77 GW of planned gas capacity is tied directly to data centers. Cleanview's analysis for AFP found about 60 data centers planning behind-the-meter generation into the early 2030s, and AI labs have signed leases for more than 10 GW of on-site power. Amazon confirmed a 7.65 GW gas plant at its GW Ranch campus in Texas.

Expected development: SemiAnalysis expects more than 40 GW of behind-the-meter data center capacity in the US by 2028, powering more than half of new data centers that year. Hyperscalers reconcile net-zero pledges with renewable certificates and carbon capture pilots, a practice institutional investors increasingly question. Gas burned for compute becomes a material driver of US power-sector emissions into the 2030s.

Challenges: Turbine and transformer lead times of three to four years cap how fast on-site generation can scale. Air-quality permits and community opposition add delays. Stranded-asset risk rises if efficiency gains or a slowdown in AI demand reduce compute needs.

Time to impact: 2-4 years

Potential impact: High

STEEP segment: Ecological

This year's trend development: New

Summary: AI already writes most new code at the most advanced companies, and agents are moving up from functions to whole systems. Software architecture, long the domain of senior engineers, becomes a task agents perform within human-defined guardrails, shifting the bottleneck to review, testing and governance.

Current situation: Google said in April 2026 that about 75% of its new code is generated by AI and then reviewed by engineers, up from 25% in October 2024, and that a complex migration run by agents and engineers together finished six times faster than a year earlier. Industry surveys estimate that 41% of all code is now AI-generated, yet engineers report being able to fully delegate only a small share of their tasks. AI-generated code produces about 1.7 times more issues than human-written code.

Expected development: Between 2029 and 2032, agents design, build, test and maintain complete applications from specifications, including architecture choices within guardrails set by humans. Engineering teams shift to defining requirements, reviewing designs and owning reliability and security. The volume of software grows sharply, and testing, review and governance become the bottlenecks.

Challenges: Generated code carries more defects and security issues, and review capacity is limited. Architecture decisions depend on business context that agents often lack. Large volumes of generated code create maintenance debt that someone must own.

Time to impact: 4-6 years

Potential impact: Very high

STEEP segment: Technological

This year's trend development: Upgraded

Summary: Chinese labs such as Alibaba's Qwen team, DeepSeek, Moonshot and Zhipu release capable models with open weights at a fraction of frontier prices. They now dominate open-model downloads and routine token volume, capping what closed-model vendors can charge and raising security and sovereignty questions for enterprises.

Current situation: The ATOM Report by Nathan Lambert and Florian Brand documents that Chinese open models overtook US ones in 2025, with Alibaba's Qwen the most-used open model family and past one billion cumulative Hugging Face downloads by March 2026. Chinese models accounted for about 41% of Hugging Face downloads over the trailing year and roughly 61% of tokens consumed on OpenRouter by May 2026, while Meta's Llama, the open-weight leader two years earlier, dropped out of the top rankings.

Expected development: Through 2028, enterprises run a barbell: premium closed models for high-stakes reasoning and cheap open-weight models for high-volume tasks. Token prices for routine work keep falling, which compresses margins for model API providers and shifts value to applications, proprietary data and distribution. US and EU policymakers debate restrictions on Chinese-origin models in critical sectors.

Challenges: Chinese data-sharing laws and possible hidden behaviours raise trust concerns in regulated industries. Chip export controls may slow future Chinese releases. Western open-weight models trail in capability, which leaves few alternatives at the low-cost tier.

Time to impact: 0-2 years

Potential impact: High

STEEP segment: Economic

This year's trend development: New

Summary: Europe has postponed the AI Act's high-risk rules to 2027 and 2028, while US states pass AI laws faster than Washington can preempt them. Companies face a patchwork of obligations that differ by jurisdiction, sector and risk level.

Current situation: In May 2026, the Council and Parliament agreed the AI part of the Digital Omnibus, delaying high-risk obligations to 2 December 2027 for standalone systems and to 2 August 2028 for AI embedded in regulated products, while Article 50 transparency rules still apply from August 2026. In the US, an executive order of 11 December 2025 created a Justice Department task force to challenge state AI laws, yet 29 states had enacted 109 AI laws by mid-2026. California's SB 53 on frontier model transparency has applied since January 2026, and Colorado's AI Act took effect on 30 June 2026.

Expected development: Between 2027 and 2029, EU high-risk obligations take effect and conformity assessment becomes routine for AI used in hiring, credit and critical infrastructure. In the US, courts decide whether state laws survive federal challenges, while Congress debates broad preemption. Multinationals converge on internal AI governance programmes that satisfy the strictest applicable rules.

Challenges: Repeated delays undermine regulatory credibility and planning certainty. The federal-state conflict in the US leaves companies complying with laws that may later be struck down. Standards for testing and documenting AI systems remain incomplete.

Time to impact: 2-4 years

Potential impact: High

STEEP segment: Political

This year's trend development: Upgraded

Summary: AI Overviews, AI Mode and chat assistants answer questions directly and end the journey on the platform. Publishers, comparison sites and any business built on organic search referrals lose traffic, which breaks the ad-funded economics of the open web.

Current situation: Chartbeat data show Google search traffic to publishers fell by about a third globally in the year to November 2025. A preregistered field experiment published in April 2026 found that AI Overviews reduced outbound organic clicks by 39.8% and raised zero-click searches by 34.5%. In the UK, the Association of Online Publishers measured a 7.1% quarterly fall in Google referrals, with how-to content losing more than 20%.

Expected development: Publishers surveyed by the Reuters Institute expect search traffic to fall a further 43% on average within three years. Visibility inside AI answers replaces ranking as the key metric, and content owners shift toward licensing, paywalls, direct audiences and paid access for AI crawlers. Smaller publishers without licensing leverage consolidate or exit.

Challenges: Referrals from ChatGPT and other assistants remain a tiny share of traffic and fall far short of replacing search volume. Opting out of AI summaries risks losing search visibility altogether, which gives platforms leverage over publishers. Competition authorities are only beginning to examine whether AI answers extend search dominance.

Time to impact: 0-2 years

Potential impact: Very high

STEEP segment: Economic

This year's trend development: New

Summary: Generative AI lets anyone publish music, video and text at near-zero cost, and synthetic uploads now rival human output on some platforms. Most of it attracts little genuine attention but a lot of fraud, so platforms shift from distributing everything to filtering, labelling and demonetising machine-made content.

Current situation: Deezer reported receiving about 90,000 fully AI-generated tracks a day in June 2026, more than half of all daily uploads at peak, up from about 10,000 a day in January 2025. Such tracks account for only 1 to 3% of streams, and up to 85% of those streams in 2025 were fraudulent and were demonetised. Deezer removes AI tracks from algorithmic recommendations and editorial playlists, and Tidal has demonetised AI audio across the board.

Expected development: Through 2028, platforms tag synthetic content by default, cap uploads and route royalties and ad revenue toward verified human creators. Recommendation systems weight provenance and engagement quality over volume. A premium market emerges for content certified as human-made, while AI-assisted production becomes standard for professional creators.

Challenges: Detection becomes harder as generated content improves. Drawing the line between fully AI-generated and AI-assisted work is contested. Smaller platforms lack the resources to filter at scale.

Time to impact: 0-2 years

Potential impact: High

STEEP segment: Economic

This year's trend development: Upgraded

Summary: Operators burned by 5G investments demand that 6G arrive as an evolution of existing networks. The first 6G standard will be finished in 2029, commercial launches follow around 2030, and early deployments in existing bands rely on software upgrades, with AI-native operation as the main selling point.

Current situation: At its June 2026 plenary, 3GPP fixed the Release 21 timeline for the first 6G standard: work item approval in March 2027, functional freezes in late 2028 and the final code freeze in March 2029, delivered as a single release. Operators in the NGMN alliance call for 6G to be deployable as a software upgrade with no intrinsic need for a hardware refresh, and 3GPP agreed to largely reuse 5G channel codes. Qualcomm notes that 6G in existing bands is possible through software upgrades, while new spectrum requires hardware evolution.

Expected development: Commercial 6G launches start around 2030 in the US, China, Japan and Korea, with Europe following as spectrum from WRC-27 becomes available. Value comes from AI-native network operation, integrated sensing, satellite integration and better spectral efficiency more than from headline speeds. Operators phase in 6G through software on 5G-Advanced equipment, which caps vendor revenue growth.

Challenges: Operators lack a clear revenue case beyond efficiency gains. Upper mid-band spectrum must be cleared and auctioned on time. Geopolitical splits over vendors and standards could fragment the ecosystem.

Time to impact: 6-8 years

Potential impact: High

STEEP segment: Technological

This year's trend development: Downgraded

Summary: AI agents are moving from product search into payment execution, as card networks, payment processors and AI platforms build tokenized rails that let software complete purchases on a shopper's behalf. Whoever controls the agent controls product selection, which shifts margin and customer access away from retailer storefronts toward AI platforms and payment networks.

Current situation: Card networks and AI platforms built the rails for agent-initiated payments during 2025 and 2026. Visa and more than ten partners introduced the Trusted Agent Protocol in October 2025 to help merchants tell legitimate AI agents from malicious bots, OpenAI and Stripe launched the Agentic Commerce Protocol, and in December 2025 Visa and partners announced completed secure agent-initiated transactions ahead of mainstream adoption in 2026. Volume is still small: Juniper Research puts agentic commerce transaction value at USD 8 billion in 2026, with card payments taking the early lead.

Expected development: Agent checkout becomes a standard payment channel by 2028, with networks issuing agent-specific tokens and merchants exposing catalogs, prices and inventory through machine-readable interfaces. Juniper projects transaction value to reach USD 3.5 trillion by 2031, and McKinsey estimates agents could handle USD 3 to 5 trillion of consumer commerce by 2030. Retailers that agents cannot parse lose visibility at the moment of purchase, while payment networks gain a new control point over which agents may transact.

Challenges: Liability for mistaken or unauthorised agent purchases is unsettled, and dispute processes were designed for human buyers. Competing protocols from OpenAI, Google, Visa and Mastercard fragment merchant integration effort. Payment analysts expect a flatter adoption curve than vendor forecasts, because most agent volume converts existing purchases, and consumer trust remains the largest barrier.

Time to impact: 0-2 years

Potential impact: Very high

STEEP segment: Economic

This year's trend development: New

Summary: The US and EU have both ended duty exemptions for low-value parcels, the loophole that powered Shein, Temu and AliExpress. Cross-border e-commerce shifts from millions of direct parcels to bulk imports and local warehouses, narrowing the price gap with domestic retailers.

Current situation: The US ended its USD 800 de minimis exemption for all imports at the end of August 2025. On 1 July 2026, the EU replaced its EUR 150 duty-free threshold with a flat EUR 3 customs duty per item category on low-value parcels, after 4.6 billion such parcels entered the EU in 2024, more than 90% of them from China; a handling fee of about EUR 2 per parcel is planned from November 2026. In July 2026, Chinese small-parcel e-commerce exports to Europe fell 25% year on year, and French transaction data showed Temu volumes down 50% from June.

Expected development: From 1 July 2028, the new EU Customs Authority and data hub replace the flat fee with category-specific duties from the first euro. Platforms move inventory into European warehouses, which turns them into conventional importers subject to product safety and tax rules. Logistics flows shift from air parcels to sea freight and bulk shipments.

Challenges: Moving stock into EU warehouses lets platforms keep much of their price advantage. Customs authorities need data quality and capacity to check billions of declarations. Higher costs also hit EU consumers and small non-EU sellers.

Time to impact: 0-2 years

Potential impact: High

STEEP segment: Political

This year's trend development: Upgraded

Summary: Device-bound passkeys, unlocked by fingerprint or face, are replacing passwords and SMS codes for login and payment confirmation. Card networks are building passkeys into checkout, which cuts fraud and friction at once and makes biometric authentication routine in e-commerce.

Current situation: The FIDO Alliance estimates that 5 billion passkeys were in use worldwide in 2026, and its State of Passkeys 2026 survey found that 75% of consumers had enabled a passkey on at least one account and 49% use them regularly. Organisations deploying passkeys report on average a 73% reduction in sign-in time and an 81% drop in login-related support tickets. Visa and Mastercard both offer payment passkey services, Mastercard aims to phase out passwords by 2030, and several financial regulators moved in 2025 and 2026 to phase out SMS one-time passwords.

Expected development: Through 2028, passkeys become the default for authenticating online card payments in Europe and other markets, replacing one-time codes in strong customer authentication. Passwords persist mainly as recovery fallbacks. Payments initiated by AI agents use delegated, device-bound credentials built on the same standards.

Challenges: Account recovery after a lost device remains a weak point attackers exploit. Merchants and banks must update legacy authentication systems. Cross-platform syncing ties users more closely to the Apple, Google and Microsoft ecosystems.

Time to impact: 2-4 years

Potential impact: High

STEEP segment: Technological

This year's trend development: Upgraded

The trends with very high impact and the shortest time to market include grid access as the hard limit on AI capacity, AI memory demand ending the era of cheap consumer electronics, China's critical-mineral export controls, circular financing in AI infrastructure, AI shopping agents at the checkout and AI answer engines draining the open web's traffic.

All six reach substantial market impact within two years.

Yes, and power sets the pace. Amazon, Alphabet, Microsoft, Meta and Oracle budgeted close to USD 700 billion for data centers in 2026, and the IEA expects data center electricity use to more than double to about 945 TWh by 2030. Grid connections, memory and copper now decide how fast that money turns into capacity.

Agents move from assisting people to acting for them. Google already generates about 75% of its new code with AI, and Visa and its partners built a protocol that lets AI agents pay at checkout. Juniper Research expects agentic commerce to grow from USD 8 billion in 2026 to USD 3.5 trillion by 2031.

The main risks are memory shortages driven by AI demand, China's export controls on gallium, germanium, antimony and rare earths, and US tariffs on advanced computing chips. China's suspension of its export ban to the US runs until 27 November 2026, and Europe still makes less than 10% of the world's semiconductors.