Re-reading a One-Year AI Forecast at the Half-Year Mark - NIA's 2026 Trend Report

Every December, trend reports pour out. Trend Korea, the house views from each securities firm, the year-ahead decks from consulting firms - they all land in the same month. NIA’s (National Information Society Agency) “2026 Top 12 AI & Digital Trends,” published December 31, 2025, sits in that same current.

Reading it, the thought that stuck with me was about form rather than content. It is a snapshot taken once a year, but AI moves in weeks. In the six months between the report’s release at the end of 2025 and now, NVIDIA revealed Vera Rubin, a real agent-security breach happened, and data-center power bottlenecks filled the headlines. Six months is neither too short nor too long a window to see how well a forecast holds.

This post does no scoring. Whether each individual trend hit or missed is for the later parts. Part 0 organizes what the report predicted and how it produced that prediction, on the report’s own terms. And at the end, one point: why the analysis method the report chose is structurally bound to lag. That becomes the baseline for the scoring that follows.

What the report is - what, by whom, why

This report is Volume 6 of NIA’s “IT & Future Strategy (IF Strategy)” series. IF Strategy is, as the name says, a planning publication that captures “strategies for the what-ifs” of the future - issued to analyze paradigm shifts in 21st-century Korean society and to explore IT-based responses.

Item	Detail
Title	NIA’s Outlook: 2026 Top 12 AI & Digital Trends
Series	IT & Future Strategy, Volume 6
Published	December 31, 2025
Publisher	National Information Society Agency (NIA)
Planning	NIA AI Policy Office, Future Strategy Team
Research	Insight Plus Inc.

Planning sits with NIA’s in-house Future Strategy Team, while the actual survey and analysis are carried out by an external research firm. Worth noting is the arrangement in which the analytical labor is outsourced while NIA holds planning, publication, and the attribution of authority - a point the series will return to later.

The report has three parts.

• Part 1, Major Sector Outlook: environment analysis across four axes - economy, industry, society, policy. It first lays the macro backdrop: roughly 3% global growth in 2026, strong cycles in semiconductors, IT, and shipbuilding, and a technology-sovereignty contest among the US, China, and the EU.
• Part 2, AI & Digital Technology Outlook: where the methodology shows. It analyzes 50 forecast reports as text to extract 10 core technologies.
• Part 3, Top 12 Trends Outlook: it bundles those technologies through a convergence lens into 12 trends.

Compressing the social outlook into one sentence, the report’s own line reads:

2026 enters the era of agentic AI - AI that judges and acts on its own - and society begins to reorganize around hyper-personalization, hyper-connectivity, and hyper-intelligence.

Agentic AI is the axis, and beneath it sit hyper-personalization (on-device AI, emotional AI, digital identity), hyper-connectivity (spatial computing, autonomous driving, intelligent eco-cities), and hyper-intelligence (AI coworkers, super-intelligent industrialization, AI governance). In broad direction, it is not far from the narrative the industry has repeated all year.

The macro backdrop it lays down - the 2026 outlook

Before going into trends, Part 1 first lays a macro backdrop across economy, industry, society, and policy. Knowing what soil the report expected trends to grow in sets the context for later scoring, so this part too is organized on the report’s own terms.

On the economy, the weight falls on the slowdown side. Major institutions saw the 2026 global economy growing in the low 3% range, and Korea recovering modestly at 1.8-2.1%.

Segment	2024	2025(e)	2026(e)	Source
World (IMF)	3.3%	3.2%	3.1%	IMF WEO (‘25.10)
World (OECD)	3.3%	3.2%	2.9%	OECD EO (‘25.12)
Korea (IMF)	2.0%	0.9%	1.8%	IMF WEO (‘25.10)
Korea (OECD)	2.0%	1.0%	2.1%	OECD EO (‘25.12)

The report splits the fork into positive and negative factors. On the positive side: high growth in AI semiconductors and steady growth in AI-transformation (AX) services such as cloud, 6G, and cybersecurity. On the negative side: protectionism spreading from US-China friction, a manufacturing slowdown, and security-driven supply-chain reshaping (reshoring). AI on its own reads as high growth, but the macro environment it sits in leans toward low growth and bloc formation - a dual structure the report concedes at the outset.

On industry, it classifies semiconductors and IT/telecom as booming, autos as stable, displays and secondary batteries as recovering, and refining, petrochemicals, and construction as depressed. AI-infrastructure demand pulls up semiconductors, data centers, and telecom, while traditional heavy industries sink under oversupply and high rates - a split running underneath the backdrop.

On policy, the frame is a technology-sovereignty contest among three axes: the US, China, and the EU.

Axis	Direction	Representative measures
US	America First, tech sovereignty	Stargate Project, CHIPS and Science Act, Paris Agreement withdrawal
China	State-led tech self-reliance	15th Five-Year Plan (2026-2030); quantum, bio, hydrogen, 6G self-reliance
EU	Strategic investment, regulatory leadership	AI Act enforcement, EUR 1B AI ecosystem, Digital Identity Wallet

The report then names five global issues: geopolitical fragmentation and neo-protectionism, AI-driven industrial restructuring, entrenched low growth, AI-driven labor-market upheaval and polarization, and the weaponization of climate resources. Worth a second look is the “November 2025 digital simplification package” under the EU entry - a move toward loosening some regulation, not tightening it. The report’s overall tone leans toward “strengthening AI governance and norms,” yet its own policy table already carries a deregulation signal. That mismatch reappears when we score the tech-sovereignty trend later.

How it was built - from 50 reports to 12 trends

This is the part to read most closely in Part 0. The procedure by which the report extracted its 12 trends sets the standard for the half-year scoring to come.

The report discloses the derivation in eight steps.

---
config:
  look: handDrawn
  theme: neutral
---
flowchart TD
    A["1) Collect institutional forecasts<br/>50 reports total"] --> B["2) Preprocess data<br/>cleaning, lemmatization, normalization"]
    B --> C["3) Build trend dataset<br/>corpus"]
    C --> D["4) Text network analysis<br/>word frequency & co-occurrence"]
    D --> E["5) Extract core technologies<br/>10 techs by connectivity"]
    E --> F["6) Topic modeling<br/>categorize content, derive themes"]
    F --> G["7) Derive trends<br/>20, weighing convergence points"]
    G --> H["8) Refine trends<br/>12, via expert consultation"]

The heart is steps 1 and 4.

Step 1 collects 50 already-published forecast reports. The source set the report discloses spans trend books (Trend Korea 2026, 2026 Trend Monitor, 2026 Trend Note), public research institutes and bodies (UNCTAD, KISTEP, WEF, Stanford HAI, MIT, IEEE, and more), government (the White House), consulting firms (McKinsey, Deloitte, KPMG, IDC, Accenture, EY, and others), and private research bodies and companies (J.P. Morgan, Gartner, Forbes, Adobe, and others). In effect, it gathers in one place the forecasts major bodies at home and abroad put out through the second half of 2025.

Step 4, text network analysis, measures word frequency and the association between words across these 50 documents. Centering on words that appear in five or more reports, it extracts and links the keywords that show up often and cluster closely together. That word map is then qualitatively grouped into 10 core technologies (step 5), themes are categorized by topic modeling (step 6), 20 trends are derived weighing convergence points among technologies (step 7), and expert consultation narrows them to 12 (step 8).

The method itself is transparent and reproducible. Disclosing which inputs went in and by what procedure they were narrowed puts it, among trend reports, on the diligent end. But one structural property is embedded in this procedure, to be dealt with below: the input is “forecasts already published,” and the core operation is “word frequency.”

What it saw - 10 core technologies and 12 trends

The 10 core technologies derived in steps 4-5 are below. The 12 trends are the result of recombining these technologies through a convergence lens.

Core technology	Key trend keywords
AI semiconductors	HBM (high-bandwidth memory) essential, PIM (processing-in-memory) commercialization, high-performance/low-power
Generative AI	Domain-specific SLM (small language model) rising, multimodal, workflow automation
AI agents	Multi-agent collaboration, autonomy/reasoning maturity, agentic RAG
Physical AI	Humanoid use imminent, RL-based autonomous judgment, edge real-time control
Next-gen telecom	Terrestrial-satellite networks, 6G spectrum, open RAN, AI-based networks
AI governance	AI legislation, mandated trustworthy AI, explainable AI (XAI)
AI security	Generative-AI attack response, model-attack defense, autonomous security agents
Quantum tech	Quantum key distribution (QKD) in practice, qubit scaling, ultra-precise quantum sensing
AI energy	Energy-supply strain, AI-based smart grids, carbon-aware computing
AI bio	Drug-discovery expansion, personalized medicine, AI-based protein-structure prediction

Here SLM is a small language model, HBM is high-bandwidth memory, PIM is a memory-centric chip that computes inside memory (processing-in-memory), XAI is AI that explains its reasoning in human-understandable terms (explainable AI), and QKD is quantum key distribution, sharing encryption keys via quantum states.

The 12 trends finalized in Part 3 are below. Each is described around two key predictions, and later parts hold those two predictions against six months of what actually happened. For now, with no scoring, only the predictions are fixed in place.

#	Trend	Key prediction 1	Key prediction 2
1	AI infrastructure hegemony	National semiconductor strategy, AI-chip market diversification	AI-compute alliances and bloc formation
2	Self-working AI agents	”Digital employees” running work autonomously	Small agents cooperating in multi-agent systems
3	Physical AI on the shop floor	Robots, autonomous vehicles, smart manufacturing accelerate AI fusion	Reduced labor dependence, structural cost savings
4	6G and satellite convergence	Mobile + satellite 3D Coverage expands	Terrestrial-satellite traffic optimized by an AI-Native Network
5	Smarter AI security	AI cyberattacks routine, autonomous preemptive defense needed	Demand for hybrid security talent grows
6	National tech sovereignty	Tech sovereignty, including data sovereignty, in focus	Diplomatic contest to set AI standards and norms
7	Vertical AI spread	Finance/legal/medical specialized vertical AI rises	On-device AI paired for real-time, tailored services
8	Quantum leap	Major powers make quantum a core capability, astronomical investment	Quantum + AI national R&D for hard problems
9	Sustainable energy transition	Energy efficiency and sustainability as the new edge	Low-power AI chips, green data centers
10	On-device AI hyper-personalization	Lightweight models and dedicated chips run large AI on small devices	IoT/edge AI embedding, all-around hyper-personalization
11	AI bio revolution	AI dramatically improves drug-discovery time and cost	Privacy tech such as anonymization and homomorphic encryption becomes essential
12	AI media content big bang	Lower barriers drive exponential content growth	Watermark/blockchain provenance tracking mandated internationally

Scanning the 12, most are subjects that circulated across the industry through 2025: agents, physical AI, on-device, quantum, tech sovereignty. No newly invented frame appears; already-discussed keywords are tidied into twelve slots. Given the method was “word frequency across 50 already-published forecasts,” that is a natural outcome.

One thing worth flagging - why this method lags

This is where the reason for starting the series comes in.

Text network analysis is a method for finding words that appear often and travel together across already-written documents. A good tool, but with a clear character. A word absent from the input cannot appear in the output, and a concept whose frequency was still low in late-2025 discourse is not a candidate for extraction in the first place. In other words, what this method captures is the consensus formed up to the time of writing, not the variables that will newly surface after it.

Why that matters shows when you overlay six months of what actually happened. Several of the things that actually moved the AI board in the first half of 2026 have no name anywhere in the 12 trends.

• Power and grid bottlenecks: in the US, a sizable share of planned data centers were canceled or delayed by power, not chips. The report treats energy as trend 9, but in a positive “sustainability is the edge” frame; the angle where power itself becomes the ceiling on growth is weak.
• AI capex circular financing: the mutual investment-and-purchase structure running through NVIDIA-OpenAI-Oracle was flagged as systemic risk. This is a matter of capital allocation, not an individual technology trend, so word frequency does not catch it.
• Agent security crisis: prompt-injection attacks surged and led to real institutional breaches. The report treats AI security as trend 5, but does not capture the frame in which agent autonomy itself widens the attack surface as a counterforce.

What the three share is that they were subjects still low in frequency in late-2025 forecast reports. Low frequency pushed them out of the text network, and being pushed out kept them out of the 12. Apart from the method’s diligence, as long as the input is the past, the output has trouble escaping the past’s consensus.

The method’s character shows once more inside the 12 trends. Each trend is described with two key predictions, as noted, and the two differ in texture. Prediction 1 is generally a tidy statement of an already-observable flow (chip development, digital employees, drug-discovery gains), while the more speculative claims attach as you move to prediction 2. Lines like “AI-Native Network rising,” “quantum + AI solving hard problems,” “privacy tech becomes essential,” “watermarking mandated internationally.” The first prediction is an extension of the present; the second is closer to a bet on the future. The two-prediction structure ended up placing an easy-to-verify observation and a hard-to-verify forecast side by side in the same cell. The scoring in later parts is likeliest to split on that second prediction.

Closing - what it means to reopen the snapshot every half-year

To sum up, NIA’s 2026 trend report is a diligently built snapshot. It gathered 50 reports through a transparent procedure, narrowed them to 12 trends, and its broad direction largely matches industry consensus. The problem is not the report’s quality but that the form of a once-a-year snapshot is structurally late in front of a subject that refreshes weekly.

From the next part on, each trend is held against six months of what actually happened. In the order the research finished first: AI infrastructure (1), AI agents (2), tech sovereignty (6), physical AI (3), and 6G/satellite (4). Each part carries the trend’s verdict in one line at the top.

• Blue: ahead - realized faster or stronger than predicted
• Green: on-track - unfolding in the predicted direction
• Yellow: lagging - right direction, short on speed or scale
• Red: overhyped - under-realized or exaggerated
• White: missed - not in the 12, but surfaced within the half-year

The labels stay only in the tables and the one-line verdict atop each post. The body should read not as a report card, but as a record of what happened over six months.

One thing to lay down in advance: the real question of this work is not “who forecast better.” It is by what cadence, by whom, and by what method the genre of trend forecasting should be refreshed in an era where technology moves this fast. Reopening the snapshot once every half-year, this series is itself one experiment in answering that.

---

The source this series holds as its reference is NIA, “NIA’s Outlook: 2026 Top 12 AI & Digital Trends” (IT & Future Strategy, Volume 6, December 31, 2025). The derivation method, core technologies, and trend definitions are all cited on the report’s own terms.