예측 검증 · 검증일 2026년 10월 12일
적중 여부: 세계 선도 AI 모델들의 예측
미국과 중국의 선도 AI 모델들은 AI가 어디로 향하는지에 대해 예측을 내놓습니다. 날짜가 있는 예측은 모두 원문 그대로 옮겨 실제 결과와 비교하므로, 누구의 예측이 맞았는지 확인하실 수 있습니다. 예측은 미리 기록해 두었으며, 기한이 도래하는 대로 검증합니다.
결과가 이미 가려진 예측(적중, 부분 적중, 빗나감)만 집계합니다. «아직 판단 불가»은 제외합니다. 정확도 열에서는 «부분 적중»을 절반으로 계산합니다.
모델별 예측 정확도
| AI | 진영 | 예측 | 적중 | 부분 적중 | 빗나감 | 아직 판단 불가 | 정확도 |
|---|---|---|---|---|---|---|---|
| Claude | 미국 | 13 | 6 | 3 | 0 | 4 | 83% |
| Qwen | 중국 | 10 | 2 | 5 | 3 | 0 | 45% |
| Meta | 미국 | 8 | 4 | 0 | 0 | 4 | 100% |
| Gemini | 미국 | 2 | 1 | 1 | 0 | 0 | 75% |
| Kimi | 중국 | 3 | 1 | 0 | 0 | 2 | 100% |
| Ernie | 중국 | 4 | 0 | 2 | 1 | 1 | 33% |
| Grok | 미국 | 4 | 0 | 0 | 0 | 4 | 아직 판단 불가 |
| ChatGPT | 미국 | 2 | 0 | 0 | 0 | 2 | 아직 판단 불가 |
모든 예측
적중중국은 2026년까지 50개가 넘는 AI 표준을 제정할 것입니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: MIIT's 2024 draft guideline targeted at least 50 sets of AI standards by 2026 plus participation in 20+ international standards. By August 2026, MIIT officials said China had formulated nearly 200 key AI standards; specific releases in 2026 include GB/Z 177-2026, a national grading standard for AI-enabled consumer devices (May 8, 2026). The 'nearly 200' figure is an official cumulative count and may include standards drafted or in progress, not only formally issued ones.
China (government plan)의 예측을 인용
원문: China plans to establish 50+ AI standards by 2026
작성: Claude출처:
적중2026년에 기업의 AI 투자가 늘어날 것입니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: McKinsey's 10th annual survey, 'The State of AI in 2026: On the Road to ROI' (reported Sept 1, 2026), found 60% of respondents plan to increase AI investment over the next year (the November 2025 source cited 92%). Gartner's 2026 CIO survey reported 89% of CIOs planned to increase AI spend in 2026. Other 2026 surveys (e.g., Bain: ~75% of companies raising AI investment; reports of enterprise AI budgets roughly doubling from 0.8% to 1.7% of revenue) point the same direction; realized full-year 2026 spending totals are not yet final.
McKinsey (survey)의 예측을 인용
원문: Enterprise AI investment will increase (92% of companies surveyed by McKinsey)
작성: Claude출처:
적중OpenAI는 2026년에 GPT-6을 출시할 것입니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: OpenAI released GPT-6 in 2026: GPT-6 Astra entered limited preview on September 3, 2026 and public release on September 4, 2026; GPT-6 Sol and GPT-6 Luna followed on September 22, 2026. On October 7, 2026 OpenAI rolled GPT-6 models with 'Intelligent UI' into ChatGPT for paid tiers, expanding to Free and Go users from October 8. Before that, the GPT-5.6 family launched July 9, 2026.
원문: GPT-6 (OpenAI)
작성: Claude출처:
- GPT-6 (Wikipedia) · 2026-10
- GPT-6 and Intelligent UI for everyone (OpenAI) · 2026-10-07
- OpenAI brings GPT-6 to ChatGPT and debuts Intelligent UI · 2026-10-07
적중Anthropic은 2026년에 Claude 5를 출시할 것입니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: Anthropic released Claude Fable 5, described as its most capable model for the general public, on June 9, 2026, following the limited 'Project Glasswing' access program. Claude Opus 5 followed on July 24, 2026, priced at $5/$25 per million input/output tokens and positioned at roughly half the price of Fable 5. Neither model is branded simply 'Claude 5', but both carry the version number 5.
원문: Claude 5 (Anthropic)
작성: Claude출처:
적중Google은 2026년에 Gemini 3.5 또는 Gemini 4를 출시할 것입니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: Google announced the Gemini 3.5 family at Google I/O on May 19, 2026 and shipped Gemini 3.5 Flash that day; on July 21, 2026 it added Gemini 3.6 Flash, Gemini 3.5 Flash-Lite and Gemini 3.5 Flash Cyber. The flagship Gemini 3.5 Pro was promised for June 2026 but, per reports through September 20, 2026, remained in partner testing and listed as 'coming soon'. No Gemini 4 release was found as of October 12, 2026.
원문: Gemini 3.5/4 (Google)
작성: Claude출처:
적중DeepSeek는 2026년에 DeepSeek-V4를 출시할 것입니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: DeepSeek released DeepSeek-V4 (preview) on April 24, 2026 as open weights in two Mixture-of-Experts variants: V4-Pro (1.6T total / 49B active parameters) and V4-Flash (284B / 13B active), both with a 1M-token context window. API pricing was published at $1.74/$3.48 per million input/output tokens for V4-Pro and $0.14/$0.28 for V4-Flash. Later checkpoints (e.g. V4-Pro-0813) appeared on hosting providers.
원문: DeepSeek-V4 (DeepSeek)
작성: Claude출처:
적중OpenAI의 AWS 인프라(380억 달러 규모 OpenAI-AWS 계약에 따른)가 2026년 1분기까지 가동에 들어갈 것입니다.예측일 2025년 11월 29일 · 기한 2026년 3월 31일
2026년 10월 12일 현재 알려진 사실: OpenAI and AWS announced the 7-year, $38B partnership on November 3, 2025, stating OpenAI would begin using AWS compute immediately, with all targeted capacity (NVIDIA GB200/GB300 on EC2 UltraServers) to be deployed before the end of 2026 and expansion into 2027+. Reports describe OpenAI as having begun using AWS right away; no public milestone found specifically declaring the full infrastructure 'live' by Q1 2026.
원문: OpenAI’s AWS infrastructure goes live (Q4 2025 / Q1 2026).
작성: Meta출처:
적중Anthropic은 텍사스와 뉴욕에 데이터센터 건설을 시작합니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: On November 12, 2025 Anthropic announced a $50 billion US data-center investment with Fluidstack, with the first custom facilities in Texas and New York coming online through 2026 (about 2,400 construction and 800 permanent jobs). An August 31, 2026 analysis reports the New York site (Lake Mariner, a repurposed coal plant) partly online, with a 42 MW building live in early July 2026 and a 168 MW building expected late September 2026, while the Texas site (Abernathy) was under construction for H2 2026 delivery. No primary-source groundbreaking ceremony announcement was found.
원문: Anthropic is breaking ground on data centers in Texas and New York.
작성: Meta출처:
- Anthropic invests $50 billion in American AI infrastructure · 2025-11-12
- Fluidstack selected by Anthropic to deliver custom data centers in New York and Texas · 2025-11
- Anthropic plans $50bn US data center spend, starting with Fluidstack sites in Texas and New York (DCD) · 2025-11
- Anthropic's 15+ GW Data Center Build: Claude's Compute, Campuses, and Chips · 2026-08-31
적중Gemini 3는 얼리 액세스 국가 밖으로 더 폭넓게 출시될 것입니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: Gemini 3 Pro launched November 18, 2025 across the Gemini app, AI Mode in Search, AI Studio and Vertex AI, and Google then expanded Gemini 3 and Nano Banana Pro in Search to more countries. During 2026 Google shipped successor models: Gemini 3.1 Pro, Gemini 3.5 Flash (Google I/O, May 19, 2026), and Gemini 3.6 Flash / 3.5 Flash-Lite (July 21, 2026); Gemini 3.5 Pro was still testing with partners as of late July 2026, and in September 2026 Google said Gemini 4 would come 'as soon as possible'. Access still varies by country, account and subscription.
원문: Wider rollout of Gemini 3 beyond early-access countries.
작성: Meta출처:
- Gemini 3: Introducing the latest Gemini AI model from Google · 2025-11-18
- Gemini 3 and Nano Banana Pro in Search are coming to more countries around the world · 2025-12
- Google releases three new Gemini models — but no 3.5 Pro · 2026-07-21
- Google says Gemini 4 release is coming 'as soon as possible' · 2026-09-24
적중GPT-4급 모델의 추론 가격은 계속 하락할 것입니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: Inference prices at fixed capability levels continued falling through 2026. Epoch AI analyses report prices for a constant performance level falling roughly 47% per quarter (median declines on the order of tens to hundreds of times per year since 2024). One example cited: an ARC-style result that cost $0.30 per question with o3 in January 2025 cost about $0.0004 with GPT-5.6 Luna in June 2026. Market trackers reported average token prices of ~$1.16-1.18 per million tokens in early August 2026, down 43% from $2.04 at end-May 2026; DeepSeek V4-Flash launched at $0.14/$0.28 per million tokens in April 2026.
원문: Continued pressure on inference costs — prices for GPT-4-class models have fallen ~90 % since 2023, making AI far more accessible.
작성: Meta출처:
적중2026년 하반기에 미국, EU, 아시아별로 뚜렷한 AI 거버넌스 모델이 자리 잡을 것입니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: By October 2026 the three regions followed visibly different tracks. The US issued a December 11, 2025 executive order ('Ensuring a National Policy Framework for AI') directing a DOJ task force to challenge state AI laws and pushing a minimally burdensome federal approach. The EU kept its risk-based AI Act but agreed (provisional deal May 7, 2026) to delay high-risk obligations to Dec 2, 2027 / Aug 2, 2028. In Asia, South Korea's AI Basic Act took effect January 22, 2026, while China continued issuing state AI standards (nearly 200 key standards per MIIT, Aug 2026).
원문: Regional divergence: Distinct AI governance models will solidify across US, EU, and Asia
작성: Qwen출처:
- Trump signs executive order to preempt state AI laws (Lawfare) · 2025-12
- EU Lawmakers Reach Provisional Agreement to Delay Key EU AI Act Obligations (Sidley) · 2026-06-22
- South Korea's AI Basic Act: Overview and Key Takeaways (Cooley) · 2026-01-27
- China expands AI standards and advances 6G development plans (OpenGov Asia) · 2026-08-26
적중2026년 하반기에 AI 칩과 모델 아키텍처의 긴밀한 통합이 경쟁 우위가 될 것입니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: Chip-model co-design was a visible competitive theme in 2026: Google sells its in-house Ironwood (TPUv7) capacity externally, with Anthropic committing to up to one million TPUs (announced Oct 2025) and Meta reported to have signed a multibillion-dollar TPU deal in February 2026. OpenAI and Broadcom (announced Oct 13, 2025) are co-designing custom accelerators, with 10 GW to be deployed starting H2 2026 and first racks targeted for late 2026. 'Competitive advantage' itself is not directly measurable; these are the closest indicators.
원문: Hardware-software co-design: Tight integration between AI chips and model architectures will become competitive advantage
작성: Qwen출처:
- OpenAI taps Broadcom to build its first AI processor in latest chip deal (Reuters via Yahoo Finance) · 2025-10-13
- OpenAI and Broadcom to co-design custom AI chips, deploy 10 GW by 2029 · 2025-10
- Anthropic expands use of Google Cloud TPUs targeting one million units · 2025-10
- Google TPU chips and Nvidia: the AI chip war (GO Markets) · 2026
적중주 모델과 별도로 존재하는 이미지 생성(DALL-E 3, Imagen, Midjourney)은 2026년에 주 모델에 통합될 것으로 예상됩니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: Image generation is now natively integrated into the main assistants: OpenAI released gpt-image-2 on April 21, 2026 and 'ChatGPT Images 2.0' to all ChatGPT plans on April 22, 2026 (native, with reasoning before generation), succeeding the 2025 GPT-4o native image generation that replaced DALL-E 3 in ChatGPT. Google's 'Nano Banana' line is Gemini's native image generation (Gemini 2.5 Flash Image, Gemini 3 Pro Image, Gemini 3.1 Flash Image). Midjourney remains a separate product.
원문: Image generation still separate (DALL-E 3, Imagen, Midjourney) but integration expected in 2026.
작성: Kimi출처:
부분 적중2026년에 최첨단 모델의 API 가격이 추가로 50% 이상 인하될 것입니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: Price moves in 2026 were mixed rather than uniform. OpenAI cut GPT-5.6 Sol (its frontier model) from $5/$30 to $4/$20 per 1M input/output tokens in August 2026 for a three-month period (~20-33%), after cutting mid-tier Terra by 20% and the small Luna model by 80% on July 30, 2026. Industry trackers describe a bifurcated market: sharp deflation in cheaper tiers (a token-expenditure index fell ~43% from late May to early August 2026) while top-tier frontier list prices rose relative to early 2026 (e.g., GPT-5.6 Sol listed at $5/$30 before the cut, vs GPT-5 at $1.25/$10 in 2025). No single authoritative index shows a 50%+ cut in frontier API list prices during 2026.
원문: Additional 50%+ price cuts on frontier model APIs
작성: Claude출처:
- OpenAI cuts developer pricing for frontier GPT-5.6 Sol model by more than 20% · 2026-08-22
- OpenAI cuts frontier model pricing · 2026-08-24
- Frontier AI price war · 2026-08
- Gartner Predicts That by 2030 Performing Inference on an LLM With 1 Trillion Parameters Will Cost GenAI Providers Over 90% Less Than in 2025 · 2026-03-25
부분 적중2026년에 Anthropic은 기업 시장에서 입지를 넓히고, Google은 멀티모달 분야에서 세를 키울 것이며, 헬스케어·법률·금융 등 특화된 버티컬 모델이 탄력을 받을 것입니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: Menlo Ventures' December 2025 report put Anthropic at 40% of enterprise LLM API spend vs OpenAI 27% and Google 21% (Google up from 20% in July 2025). Ramp's AI Index (card/bill-pay data from 50,000+ US businesses) showed Anthropic at 34.4% business adoption in April 2026, overtaking OpenAI's 32.3% for the first time. Google's Gemini 3.1 Pro was reported to top most major benchmarks in spring 2026; no single authoritative 2026 measure of 'multimodal leadership' or of vertical-model momentum (healthcare/legal/finance) was found.
원문: Gaining: Anthropic (enterprise dominance), Google (multimodal leadership), specialized vertical models (healthcare, legal, finance)
작성: Claude출처:
부분 적중2026년에 OpenAI는 기업 시장 점유율을 계속 잃을 것이며, 순수 오픈소스 도입은 정체될 것입니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: Menlo Ventures' late-2025 data put Anthropic at about 40% of US enterprise LLM API spend versus OpenAI at 27% (OpenAI had ~50% in 2023). In the Ramp AI Index for August/September 2026, 43.8% of Ramp business customers paid Anthropic vs 39.8% for OpenAI; however, a WSJ analysis of OpenRouter data (reported Oct 9, 2026) found OpenAI's share of coding/productivity traffic rose from under 25% at the start of 2026 to almost 50% by September. On open source, Constellation Research (Aug 24, 2026) cited Vercel AI Gateway data showing open-weight models rising from 11% of token volume in April 2026 to 62% by August 22, 2026, though under 4% of spend.
원문: Facing Pressure: OpenAI (enterprise share erosion), pure open-source plays (adoption stalling)
작성: Claude출처:
- Ramp AI Index September 2026: Anthropic vs OpenAI spend · 2026-09
- OpenAI Nearly Doubled Its Share Against Anthropic In 2026, According To New Report (citing WSJ/OpenRouter) · 2026-10-09
- Open-Weight Models Are Gaining Ground in Enterprise AI (Constellation Research) · 2026-08-24
- Enterprises prefer Anthropic's AI models over anyone else's, including OpenAI's (TechCrunch) · 2025-07-31
부분 적중Anthropic은 2026년에 Claude의 멀티모달 기능을 확장할 수 있습니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: On April 17, 2026 Anthropic launched Claude Design (Anthropic Labs, powered by Claude Opus 4.7), which generates visual prototypes, slides and one-pagers on a live canvas for paid subscribers. Claude also has a voice mode (launched in beta May 2025) and image/document input; as of 2026 reporting, Claude still offers no native image, audio or video generation. Claude Fable 5 (June 2026) and Opus 5 (July 2026) were positioned mainly on long-running agentic, coding and knowledge-work tasks.
원문: May expand multimodal capabilities in 2026 to counter Google’s advantage, but likely to maintain focus on high-value professional applications.
작성: Qwen출처:
부분 적중2026년 상반기에 전 세계적인 칩 부족으로 AI 역량 발전이 제약될 수 있습니다.예측일 2025년 11월 29일 · 기한 2026년 6월 30일
2026년 10월 12일 현재 알려진 사실: In H1 2026 the main hardware bottleneck was memory and packaging rather than GPU dies: SK Hynix, Samsung and Micron reported 2026 HBM output essentially sold out, DRAM contract prices rose ~93-98% in Q1 2026, and data-center GPU lead times were reported at 9-12 months. Compute scarcity hit AI services: Anthropic tightened Claude peak-hour usage limits in late March 2026 and in May 2026 contracted all of SpaceX/xAI's Colossus 1 (~220,000 Nvidia GPUs); GitHub paused new Copilot sign-ups in April 2026 citing compute demand. At the same time frontier releases continued (DeepSeek V4 in April, GPT-5.6 in July, GPT-6 in September) and Nvidia reported record Q2 FY27 revenue of $96.2B (data center $89.0B).
원문: Hardware constraints: Global chip shortages may limit AI capability advancement despite algorithmic improvements
작성: Qwen출처:
부분 적중2026년 상반기에 조직들은 AI 실험에서 대규모 실제 운영 배포로 옮겨갈 것입니다.예측일 2025년 11월 29일 · 기한 2026년 6월 30일
2026년 10월 12일 현재 알려진 사실: McKinsey's State of AI 2026 (reported Sept 1, 2026) found 44% of organizations scaling AI across the enterprise, up from 38% a year earlier (54% of $1B+ companies, about one-third of smaller firms), with only ~6% qualifying as high performers with measurable bottom-line impact. Gartner-cited 2026 data showed only 17% of organizations had deployed AI agents, and Gartner (Sept 29, 2026) predicted many vendor-built agentic deployments will be abandoned by 2028. No single direct measure of a 'shift to production at scale' in H1 2026 exists; these are the closest proxies.
원문: Enterprise maturity: Organizations will shift from AI experimentation to production deployment at scale
작성: Qwen출처:
부분 적중2026년 하반기에 멀티모달 상호작용 프로토콜을 위한 업계 표준이 등장할 것입니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: Agent-interoperability standards consolidated under the Linux Foundation: Anthropic donated MCP to the Agentic AI Foundation (AAIF, founded December 2025 with OpenAI, Anthropic, Google among members), and Google's A2A protocol is also hosted by the Linux Foundation. MCP Apps (January 26, 2026) added interactive UI returns, and the MCP 2026 roadmap lists first-class audio/video modalities. No dedicated, broadly adopted standard specifically for multimodal (voice/vision) interaction protocols was found as of October 2026.
원문: Multimodal standardization: Industry standards will emerge for multimodal interaction protocols
작성: Qwen출처:
부분 적중2026년 하반기에 AI 에이전트와 워크플로를 거래하는 마켓플레이스가 주류가 될 것입니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: Several large platform-run AI agent marketplaces launched or expanded in 2026: Google Cloud launched an Agent Marketplace in April 2026, Microsoft an Agent Store inside Microsoft 365 Copilot in May 2026, and AWS launched an AI agent marketplace on July 15, 2026 (AWS Summit New York) with Anthropic as a launch partner. AWS Marketplace's agents category (opened July 2025) grew from ~800 to over 2,100 listings by December 2025, and Salesforce AgentExchange launched with ~200 partners. 'Mainstream' has no standard metric; there is no independent measurement of enterprise purchasing share via such marketplaces.
원문: Agent economy: Marketplaces for AI agents and workflows will become mainstream
작성: Qwen출처:
부분 적중2026년에는 AI 에이전트가 사람의 입력 없이 독자적으로 행동할 것입니다. 예컨대 휴가를 예약하고 자산을 관리하게 됩니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: Agentic transactions began appearing in 2026 but largely with human confirmation and in pilots: Visa launched its 'Agentic Ready' program on March 17, 2026 with 21 European issuers, an agent acting for a Santander customer completed a Visa card purchase in March 2026, and Mastercard took Agent Pay live in Portugal and the Netherlands in May 2026. In travel, Google ran a limited US test of agentic hotel booking in AI Mode while agentic flight booking had not launched; Priceline called its agent 'fully agentic' in June 2026. OpenAI discontinued ChatGPT's Instant Checkout on March 4, 2026, redirecting purchases to merchant sites. No broad consumer use of agents managing finances without human input was documented.
원문: AI agents will act independently (e.g., booking vacations, managing finances without human input).
작성: Ernie출처:
부분 적중2026년 양자 AI 실험이 신약 개발과 기후 모델링의 돌파구로 이어질 수 있습니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: 2026 saw research collaborations and demonstrations rather than a widely recognized drug or climate breakthrough attributable to quantum AI: Cleveland Clinic, RIKEN and IBM simulated protein complexes up to 12,635 atoms with hybrid quantum-classical methods; IonQ with AstraZeneca, AWS and NVIDIA reported a 20x speedup on a Suzuki-Miyaura reaction workflow; Roche-Quantinuum and Boehringer Ingelheim-Google partnerships target early-stage drug discovery; IBM and AMD announced quantum-centric supercomputing aimed at drug discovery and climate modeling. No quantum-derived drug candidate or climate-model result in production use was found.
원문: Quantum AI experiments may lead to breakthroughs in drug discovery and climate modeling.
작성: Ernie출처:
- IBM-AMD Alliance Targets Drug Discovery, Climate Modeling With Quantum Supercomputers · 2026
- 'Can it actually deliver?': Why Big Pharma is making the leap to the quantum realm · 2026
- IonQ teams up with AstraZeneca, AWS and NVIDIA to speed up drug development with quantum computing · 2025-06
- Bayer taps Google's quantum power for drug discovery · 2026
빗나감EU AI 법(AI Act)의 전면 시행이 2026년 1월에 시작되며, 고위험 AI 시스템에 대한 규정 준수 문서가 요구됩니다.예측일 2025년 11월 29일 · 기한 2026년 1월 31일
2026년 10월 12일 현재 알려진 사실: The EU AI Act did not reach 'full enforcement' in January 2026; its original schedule set August 2, 2026 for most high-risk (Annex III) obligations. The Digital Omnibus on AI, proposed in November 2025, was published in the Official Journal on July 24, 2026 and entered into force July 27, 2026, deferring Annex III high-risk obligations to December 2, 2027 and Annex I (product-embedded) high-risk obligations to August 2, 2028. Prohibited practices (Feb 2025), AI literacy, GPAI obligations (Aug 2025) and Article 50 transparency duties stay on their original schedule.
원문: EU AI Act implementation: Full enforcement begins January 2026, requiring compliance documentation for high-risk AI systems
작성: Qwen출처:
빗나감2026년 상반기에 EU AI 법 준수 기한으로 인해 모델 배포 전략에 상당한 변화가 불가피할 것입니다.예측일 2025년 11월 29일 · 기한 2026년 6월 30일
2026년 10월 12일 현재 알려진 사실: Instead of new obligations biting in H1 2026, the EU adopted the 'Digital Omnibus on AI': provisional agreement on May 7, 2026, Parliament vote June 16, 2026, publication in the Official Journal July 24 and entry into force July 27, 2026. It deferred stand-alone high-risk (Annex III) obligations from August 2, 2026 to December 2, 2027, and product-embedded high-risk (Annex I) from August 2027 to August 2028, and gave existing systems until December 2, 2026 for AI-content labelling. GPAI model obligations had applied since August 2, 2025.
원문: Regulatory enforcement: EU AI Act compliance deadlines will force significant changes in model deployment strategies
작성: Qwen출처:
빗나감2026년 상반기에 인수합병과 폐업을 통해 독립 AI 스타트업 수가 15~20% 줄어들 것입니다.예측일 2025년 11월 29일 · 기한 2026년 6월 30일
2026년 10월 12일 현재 알려진 사실: No census of the number of independent AI startups exists, so a 15-20% reduction cannot be measured directly. Acquisition activity rose: Crunchbase counted 195 AI startup acquisitions early in 2026 (already 14% above all of 2025, per reporting), OpenAI made 8 acquisitions in Jan-Apr 2026, and SpaceX's reported ~$60B purchase of Cursor/Anysphere closed in Q2 2026. SimpleClosure's H1 2026 report said AI companies were only 14.4% of its shutdowns and it saw its fewest AI closures ever, while AI took $355.9B of $412.7B US startup investment in H1 2026.
원문: Consolidation wave: Expect 15–20% reduction in independent AI startups through acquisitions and closures
작성: Qwen출처:
빗나감2026년에는 AI 규제가 강화되며, EU AI 법과 미국의 AI 권리장전이 세계 표준을 정하게 될 것입니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: The 'AI Bill of Rights' was a non-binding 2022 White House blueprint; it was removed from active White House policy after January 2025, and Executive Order 14110 was rescinded, replaced by a deregulatory approach (AI Action Plan, July 2025; December 11, 2025 order targeting state AI laws). In the EU, the AI Act remains in force but the May 2026 Digital Omnibus deal delayed high-risk obligations to 2027-2028. South Korea's AI Basic Act took effect January 22, 2026.
원문: Regulation will tighten, with EU’s AI Act and U.S. AI Bill of Rights setting global standards.
작성: Ernie출처:
아직 판단 불가2026년까지 기업 애플리케이션의 40%가 과제 특화형 AI 에이전트를 사용하게 될 것입니다(2025년 5%에서 증가).예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: Gartner's August 26, 2025 press release forecast that 40% of enterprise applications would feature task-specific AI agents by end-2026, up from under 5% in 2025; no Gartner measurement of the actual 2026 share of enterprise applications with embedded agents was found as of October 12, 2026. Closest proxies: Gartner survey data cited in 2026 that only 17% of organizations had deployed AI agents (60%+ planning to within two years), and McKinsey's State of AI 2026 (Sept 1, 2026) reporting AI-agent deployment at 40% of large ($1B+) companies (up from 27%) and 22% of smaller ones. Gartner on Sept 29, 2026 also predicted 70% of enterprises will abandon agentic AI built by vendor forward-deployed engineering by 2028.
Gartner의 예측을 인용
원문: Gartner predicts 40% of enterprise applications will leverage task-specific AI agents by 2026, up from 5% in 2025.
작성: Claude출처:
- Gartner Predicts 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026, Up from Less Than 5% in 2025 · 2025-08-26
- Companies' financial value from AI holds firm in 2026 (FM Magazine on McKinsey State of AI 2026) · 2026-09-01
- Gartner: Agentic AI to Define Next Wave of Enterprise ROI (Technology Magazine) · 2026
- Gartner Predicts 70% of Enterprises Will Abandon Agentic AI Built by Vendor Forward-Deployed Engineering by 2028 · 2026-09-29
아직 판단 불가OpenAI의 매출은 2027년까지 1,000억 달러에 이를 것으로 전망됩니다.예측일 2025년 11월 29일 · 기한 2027년 12월 31일
2026년 10월 12일 현재 알려진 사실: OpenAI's annualized revenue run rate was reported at about $25B at end of February 2026 (vs $21.4B at end-2025), and Reuters/Bloomberg-sourced reports in late September/early October 2026 put it near $70B, up >70% since early July (unaudited, anonymous-source figures). Full-year 2026 recognized revenue is projected far lower than the run rate (one cited internal projection ~$29.4B). Sam Altman has reportedly floated $100B revenue by 2027; the deadline (end-2027) has not passed and no 2027 actuals exist.
원문: | OpenAI | ~$13B ARR | ~$500B | $100B projected by 2027 |
작성: Claude출처:
- OpenAI hits $25 billion in annualized revenue while Anthropic surges to $19 billion · 2026-03
- OpenAI's revenue run rate approaches $70 billion but cash burn and reporting gaps cloud the picture · 2026-10-08
- OpenAI sees 2026 run rate revenue reaching or topping USD 70bln, according to Bloomberg sources · 2026
- OpenAI's trillion dollar question · 2026
아직 판단 불가2026년까지 조직의 70%가 자율형 AI 에이전트를 사용하게 될 것입니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: There is no direct measurement of '70% of organizations using autonomous AI agents' in 2026. Gartner's 2026 CIO survey (reported mid-2026) found only 17% of organizations had deployed AI agents, with 42% planning to within 12 months and a further 22% within 24 months; Gartner places agentic AI at the Peak of Inflated Expectations. A separate Gartner survey (Sept-Oct 2025) found only 15% of IT application leaders were considering, piloting or deploying fully autonomous agents. Gartner also forecasts 40% of enterprise applications will include task-specific agents by end-2026.
Gartner의 예측을 인용
원문: Gartner forecasts 70% of organizations will use autonomous AI agents by 2026
작성: Claude출처:
아직 판단 불가Meta는 2026년에 Llama 5를 출시할 것입니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: No model branded 'Llama 5' confirmed by Meta was found as of October 12, 2026; mainstream coverage (TechCrunch's Llama explainer, CNBC) still lists Llama 4 (April 2025) as the latest Llama generation. On April 8, 2026 Meta Superintelligence Labs launched Muse Spark, a proprietary model, and in August 2026 Meta released an open-weight model called Muse Glimmer and said an open-weight version of Muse Spark 1.2 would follow in coming weeks. Several low-credibility aggregator sites claim a '600B Llama 5' release in April 2026, but these are not corroborated by Meta or major outlets.
원문: Llama 5 (Meta)
작성: Claude출처:
- Muse Spark: everything you need to know · 2026-04-08
- Meta and Nvidia plant 'very firm flag' in open-weight AI race led by Chinese Labs · 2026-08-12
- Meta Llama: Everything you need to know about the open generative AI model · 2025-10-06
- Meta Hasn't Given Up on Open Source: Muse Spark Launches as Open-Weight Plans Continue · 2026-04
아직 판단 불가멀티모달 AI 시장은 2025년 33억 달러에서 2035년 약 940억 달러로 성장할 것입니다.예측일 2025년 11월 29일 · 기한 2035년 12월 31일
2026년 10월 12일 현재 알려진 사실: The 2035 target cannot be measured as of October 2026. Analyst estimates of the 2025 multimodal AI market base vary: about $2.2B (The Business Research Company, $2.83B in 2026), $2.6B (P&S Intelligence, to $23.3B by 2032 at 36.8% CAGR), and $3.09B (to $52.87B by 2034 at 37.1% CAGR). These current forecasts are lower than the ~$94B-by-2035 trajectory cited in the prediction.
원문: the whole sector is projected to surge from $3.3 billion in 2025 to nearly $94 billion by 2035.
작성: Meta출처:
아직 판단 불가기업 앱의 30% 이상이 2026년까지 멀티모달 에이전트를 내장할 것입니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: No direct 2026 measurement of the share of enterprise applications embedding multimodal agents was found as of October 12, 2026. The closest benchmark is Gartner's forecast (Aug 26, 2025) that 40% of enterprise apps would feature task-specific AI agents by end-2026 (from under 5% in 2025), which does not specify multimodality and has not yet been measured. Proxies: McKinsey State of AI 2026 (Sept 1, 2026) found agent deployment at 40% of large companies and 22% of smaller ones; Gartner-cited 2026 data put organizations with deployed agents at 17%.
원문: with more than 30 % of enterprise apps expected to embed multimodal agents by 2026.
작성: Meta출처:
아직 판단 불가전 세계 AI 지출은 2025년 3,070억 달러에서 2028년 6,320억 달러로 성장할 것입니다.예측일 2025년 11월 29일 · 기한 2028년 12월 31일
2026년 10월 12일 현재 알려진 사실: Gartner (January 15, 2026) forecast worldwide AI spending of $2.52 trillion in 2026 (+44% YoY), raised in May 2026 to $2.59 trillion (+47%), with infrastructure over 45% of the total. These figures use a much broader definition than the $307B (2025) baseline cited by VC Cafe (an IDC-style AI solutions figure), so they are not directly comparable; no 2028 actuals exist yet.
VC Cafe의 예측을 인용
원문: VC Cafe notes $307 billion in global AI spending this year, projected to balloon to $632 billion by 2028.
작성: Meta출처:
아직 판단 불가기업의 82%가 2026년까지 AI 에이전트를 도입할 계획입니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: No survey directly measures the '82% of firms embedding agents by 2026' figure (which was a plan/intention statistic). Closest 2026 measures: KPMG's Q2 2026 AI Pulse reported enterprise AI agent deployment rising from 23% to 56% in one quarter; Gartner's 2026 CIO survey found 17% of organizations had deployed AI agents and 64% planned to within two years; McKinsey's State of AI (Nov 2025) found 62% experimenting with agents but only 23% scaling them in at least one function. Results vary widely by definition of 'agent' and 'deployed'.
원문: Agentic AI – 82 % of firms plan to embed agents by 2026;
작성: Meta출처:
아직 판단 불가AGI는 2026년에 도래할 것입니다(Altman과 Musk의 주장).예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: There is no agreed definition or independent test of AGI, so its arrival cannot be measured directly. After OpenAI released GPT-6 Astra (public September 4, 2026), OpenAI President Greg Brockman said it was 'not unreasonable to feel that we are now in the AGI era' while describing AGI as a 'mission concept' rather than a contractual trigger; on August 27, 2026 Sam Altman said 'by the end of 2026, we will have an internal system that we call AGI'. Other leaders disagree: Demis Hassabis (Google DeepMind) has said AGI is still years away, while Musk and Dario Amodei have predicted systems by end-2026.
Sam Altman, Elon Musk의 예측을 인용
원문: AGI Timing: Claims of "AGI by 2026" (Altman, Musk) vs. more pessimistic views. If AGI emerges, all rankings become obsolete overnight.
작성: Kimi출처:
- OpenAI AGI statements (GIGAZINE) · 2026-08-27
- GPT-6 (Wikipedia) · 2026-10
- AI industry leaders diverge on AGI timelines · 2026
아직 판단 불가2026년 EU AI 법 시행은 AI 모델 시장을 재편할 수 있습니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: EU AI Act general-purpose AI obligations have applied since August 2, 2025; Amazon, Anthropic, Google, IBM, Microsoft, OpenAI, Mistral and Aleph Alpha signed the GPAI Code of Practice, xAI signed only the safety chapter, and Meta declined. Under the Digital Omnibus, EU institutions provisionally agreed on May 7, 2026 to postpone high-risk system obligations from August 2026 to December 2, 2027 (standalone) and August 2, 2028 (embedded). No quantitative study found measuring a 2026 reshaping of the AI model market attributable to the Act.
원문: EU AI Act implementation in 2026 may reshape market.
작성: Kimi출처:
아직 판단 불가2026년에는 전 세계 AI 규제가 강화되며, 미국, EU, 중국이 기본 AI 윤리 기준에 합의할 것입니다.예측일 2025년 11월 29일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: No trilateral US-EU-China agreement on AI ethical standards has been found as of October 2026. US and Chinese officials discussed frontier-AI guardrails at the May 14-15, 2026 Beijing summit and agreed to keep talking about a best-practices protocol against misuse by non-state actors, without a binding agreement. Meanwhile regulation moved in different directions: the EU agreed on May 7, 2026 (Digital Omnibus) to postpone AI Act high-risk obligations to December 2027/August 2028, and the US administration pursued federal preemption of state AI laws (EO 14365, Dec 11, 2025; National AI Policy Framework, Mar 20, 2026).
원문: Global AI regulation will tighten, with U.S., EU, and China agreeing on basic ethical standards.
작성: Ernie출처:
적중2026년에는 모든 과제에서 최고인 단일 모델은 없을 것입니다.예측일 2025년 11월 28일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: Leaderboards in 2026 show a crowded frontier: in the July-August 2026 LMArena snapshots, Claude Opus 4.8 led text and coding, with Gemini 3.1 Pro and Opus 4.7 close behind, and the Elo gap between #1 and #10 was about 28 points, largely within noise. Separate arenas (text, coding, vision, etc.) and cost/latency trade-offs lead commentators to recommend choosing models by task (e.g. Claude for coding, ChatGPT for voice/image, Gemini for multimodal, open-weight models for self-hosting). 'Best for every task' is not formally measured; the evidence is leaderboard dispersion.
원문: Looking ahead into 2026, the rational strategy is model-agnosticism. There is no single “best” model for everything:
작성: Gemini출처:
부분 적중기업들은 2026년에도 GPT-4급 API에서 오픈 웨이트 모델인 Llama 4 및 Qwen 3으로 전환을 계속할 것입니다.예측일 2025년 11월 28일 · 기한 2026년 12월 31일
2026년 10월 12일 현재 알려진 사실: An Enterprise Technology Research survey (September 2026, ~200 respondents) found open-weight models handle 34% of enterprise AI token usage, up from 23% a year earlier; 42% of respondents run open-weight models in production, and 60% of those have moved some workloads off proprietary APIs, while 93% say proprietary usage keeps growing. Google's Gemma overtook Meta's Llama as the most deployed open-weight family in that survey (Llama production use fell from 55% to 35% since July), and Qwen has the largest derivative ecosystem (>151,000 derived repos). So migration toward open weights is measured, but the leading families shifted away from Llama 4 specifically.
원문: More and more companies are migrating from GPT-4-class APIs to Llama 4 and Qwen 3 to gain:
작성: Gemini출처:
아직 판단 불가AI 관련 자본 지출이 2029년까지 2조 8,000억 달러에 이를 것입니다.예측일 2025년 11월 28일 · 기한 2029년 12월 31일
2026년 10월 12일 현재 알려진 사실: Citigroup raised its forecast in late September 2025 to more than $2.8 trillion of cumulative Big Tech AI infrastructure spending through 2029 (from $2.3T), with hyperscaler AI capex of about $490B in 2026. In 2026 the four largest hyperscalers (Amazon ~$200B, Microsoft ~$190B, Alphabet $180-190B, Meta $125-145B) guided to roughly $700-725B of combined 2026 capex, up ~77% from ~$410B in 2025, and Wall Street expects 2027 capex above $1 trillion. The 2029 cumulative total cannot be measured until 2029; current run-rates are above Citi's 2026 interim estimate.
Citigroup의 예측을 인용
원문: Citigroup estimating $2.8 trillion in AI-related capex by 2029.
작성: ChatGPT출처:
아직 판단 불가AI는 향후 10년 동안 미국 경제에 최대 20조 달러의 추가 가치를 더할 것입니다.예측일 2025년 11월 28일 · 기한 2035년 12월 31일
2026년 10월 12일 현재 알려진 사실: The $20 trillion figure comes from a Goldman Sachs analysis reported in November 2025, estimating widespread AI adoption could add about $20 trillion of value to the US economy (including ~$8 trillion of capital income) and lift US labor productivity ~15% over roughly a decade. The window runs to about 2035, so the outcome cannot be measured as of October 2026; only early indicators exist (e.g., heavy AI capex such as Amazon's ~$98B H1 2026 capex).
Goldman Sachs의 예측을 인용
원문: Goldman Sachs argues that the AI boom is “barely begun,” projecting up to $20 trillion of additional value for the U.S. economy alone over the next decade.
작성: ChatGPT출처:
아직 판단 불가AI 시장은 2031년경 약 1조 달러에 이르며, 연평균 26.6% 성장할 것입니다.예측일 2025년 11월 28일 · 기한 2031년 12월 31일
2026년 10월 12일 현재 알려진 사실: The target year is 2031, so the outcome cannot be measured yet. Market-research estimates diverge widely by definition: Fortune Business Insights sizes the AI market at ~$376B in 2026 growing 26.6% CAGR to ~$2.48T by 2034; MarketsandMarkets (June 2026) estimates ~$602B in 2026 rising to ~$3.64T by 2033 (29.3% CAGR); SkyQuest projects ~$2.36T by 2031. Several current forecasts therefore put the market above $1T before 2031.
원문: 26.6% annual growth pushing AI toward a $1T market by 2031
작성: Grok출처:
아직 판단 불가GPT-5.1/OpenAI의 상대적 지배력은 개인정보 보호 규제와 강력한 대안 오픈소스의 등장으로 1~2년 안에 약화될 수 있습니다.예측일 2025년 11월 28일 · 기한 2027년 11월 30일
2026년 10월 12일 현재 알려진 사실: As of October 2026 the evidence on OpenAI's relative position is mixed: Anthropic leads in several enterprise measures (Ramp AI Index Aug/Sep 2026: 43.8% of businesses paying Anthropic vs 39.8% OpenAI; Menlo late-2025: ~40% vs 27% of enterprise API spend), while WSJ-reported OpenRouter data show OpenAI's coding/productivity traffic share rising from under 25% to almost 50% during 2026. Open-weight models grew strongly in usage (Vercel AI Gateway: 11% to 62% of token volume April–August 2026). On regulation, the EU's Digital Omnibus (in force July 27, 2026) deferred AI Act high-risk obligations to Dec 2027/Aug 2028; no regulatory action specifically curbing OpenAI's market position was found. The prediction window runs to November 2027.
원문: However, regulatory scrutiny on data privacy and a surge of powerful open alternatives could weaken its relative dominance within 1–2 years.
작성: Grok출처:
- Ramp AI Index September 2026: Anthropic vs OpenAI spend · 2026-09
- OpenAI Nearly Doubled Its Share Against Anthropic In 2026 (citing WSJ/OpenRouter) · 2026-10-09
- Open-Weight Models Are Gaining Ground in Enterprise AI · 2026-08-24
- The AI Act Implementation Timeline: What Changes Under the AI Omnibus? (FPF) · 2026
아직 판단 불가Gemini의 OS 통합 기능은 향후 1~2년 안에 주류 사용자 사이에서 지배력을 공고히 할 수 있습니다.예측일 2025년 11월 28일 · 기한 2027년 11월 30일
2026년 10월 12일 현재 알려진 사실: The Gemini app grew from >750M monthly active users at the start of 2026 to >900M at Google I/O (May 2026) and passed 1 billion MAU in August 2026, per Google. In January 2026 Apple agreed to license a custom Gemini model to power a rebuilt Siri, and at WWDC (June 8, 2026) Apple presented Apple Foundation Models built with Gemini for iOS 27. Third-party traffic data show ChatGPT still leads but its share fell (Similarweb: ~61.7% of AI web traffic in Feb 2026 vs 75.7% a year earlier; Gemini ~24%); the 1–2 year window runs to late 2027.
원문: Conversely, OS integrations may solidify dominance in the next 1–2 years for mainstream users.
작성: Grok출처:
- Gemini app: one billion monthly users · 2026-08
- Google Gemini App Surpasses 1 Billion Monthly Active Users · 2026-08-12
- ChatGPT still leads the chatbot market but its dominance is slipping as Google's Gemini gains ground · 2026
- Apple WWDC26: Siri AI and Gemini-backed foundation models on device and Private Cloud Compute · 2026-06
아직 판단 불가Llama 4는 향후 1~2년 안에 다양한 워크로드의 기본 효율성 표준이 될 수 있습니다.예측일 2025년 11월 28일 · 기한 2027년 11월 30일
2026년 10월 12일 현재 알려진 사실: Llama 4 (released April 2025) received mixed-to-poor reviews and did not gain broad enterprise adoption. On Hugging Face, Alibaba's Qwen surpassed Llama in cumulative downloads by September 2025, and by March 2026 Qwen had ~942.1M downloads versus Llama's ~476.0M. Reports in late 2025 said Meta was using Qwen models in training its next model ('Avocado'), and Chinese open-weight models (Qwen, DeepSeek V4) dominate open-weight usage discussions in 2026. The deadline (Nov 2027) has not passed.
원문: Could become the default efficiency standard for many workloads in the next 1–2 years.
작성: Grok출처:
조건에 맞는 예측이 없습니다.
집계 방법
- 날짜나 분명한 연도가 있는 예측만 원문 그대로 채택합니다. 날짜 없는 막연한 바람은 제외하며, 선거, 전쟁, 제재, 국가 간 경쟁에 관한 예측도 제외합니다.
- «아직 판단 불가» 또는 «부분 적중»인 예측은 한 달에 한 번, 그리고 기한 직후에 다시 확인합니다. 모든 확인은 날짜와 함께 보관하며 지우지 않습니다.
- 일부 AI는 다른 이들의 예측(Gartner, Altman, 각국 정부)을 되풀이했습니다. 카드에 누구의 예측인지 밝혀 두었으며, 이는 그것을 인용하기로 한 AI의 몫으로 집계합니다. 기한이 2026년 이후인 예측은 «아직 판단 불가»로 남습니다.