AI Predictions 2027: My 10 Trends, and What 2026 Taught Me

TL;DR: my 10 predictions for 2027

1. The IPO year makes AI’s economics public. Anthropic and OpenAI will face quarterly scrutiny, and the gap between AI spending and AI revenue becomes a public number.
2. The backlash turns into local politics. Electricity bills, water and jobs, more than “AI risk,” decide where data centers get built.
3. The junior ladder gets rebuilt. Entry-level hiring keeps narrowing, and education compresses into shorter, AI-native programs.
4. Agents become coworkers, and verification becomes the craft. Humans move from doing the work to specifying it, checking it and owning the result.
5. Personal and disposable software wins. People and teams build for their own needs. The vibe-coded app store gets crowded.
6. Memory, power and permits set the pace. The bottleneck is physical, and 2027 looks like the tightest year yet.
7. Physical AI moves from demo to deployment, slowly. World models and robots scale in narrow, structured settings first.
8. China wins on volume and openness, and Europe pushes for sovereignty. Who can switch off your model becomes a board question.
9. AI moves into the home and the wallet. Assistants, glasses and agents that buy on your behalf.
10. Agents that hack force new rules. Agent identity, access controls and AI-on-AI security become mandatory.

Searches for “AI predictions 2027” will start climbing next month, and I want to publish mine before the noise starts. This is my own analysis. I have written about AI and digital strategy all year on this site, and these ten predictions are where my reading of 2026 lands.

How I built these predictions

I went beyond media coverage this time and used three types of sources:

• The labs’ own economists. Anthropic’s Economics team published more than 15 reports in 2026, including the Economic Index, labor-market studies and a survey comparing what AI insiders and the public expect. OpenAI published enterprise usage research. These are the closest we have to real usage data.
• AI-first companies. Google, Spotify, Amazon, OpenAI, Anthropic, Shopify, Uber and others are 12 to 24 months ahead of most organizations. What they did in 2026 is what everyone else will try in 2027.
• The agenda setters. McKinsey, BCG, Bain, Gartner, PwC and Microsoft shape what boards and CEOs believe. I checked where they agree with me and where they don’t.

One warning before the list. Most of the numbers below come from companies with something to sell, including AI labs reporting on themselves. I mark where evidence is self-reported or comes from a single source.

What 2026 looked like in 12 numbers

• 89% of organizations use AI in at least one function, but only 37% report any EBIT impact (McKinsey State of AI 2026).
• 75% of Google’s new code is AI-generated, up from 25% in 2024 and 50% in fall 2025 (Sundar Pichai, April 2026).
• 6 to 1: agent hours outnumbered employee hours at Anthropic by August 2026 (Anthropic Economics).
• 8.3x: “frontier firms” generate 8.3 times the AI output per user of typical firms, up from 2.6x in January (OpenAI, August 2026).
• 15x: growth in active agents in Microsoft 365 year over year (Microsoft Work Trend Index 2026).
• August 2026: the first month GitHub saw more agent-authored pull requests than human-authored ones (GitHub data via The Pragmatic Engineer).
• 52% of Americans are more concerned than excited about AI, up from 37% in 2021 (Pew).
• 27% of US voters would support a data center in their community, less than for a nuclear plant (Echelon Insights).
• $225 billion in AI-linked bonds issued in the first half of 2026, up 974% (S&P Global).
• 95%+ of global humanoid robot shipments came from Chinese companies in H1 2026 (IDC).
• 4 labs, OpenAI, Anthropic, Meta and Google, disclosed models that escaped test environments and breached real systems.
• 18 days: Anthropic’s Fable 5 and Mythos 5 were switched off by a US export-control order in June before access was restored.

1. The IPO year makes AI’s economics public

What I saw in 2026

• Anthropic confidentially filed for an IPO on June 1, reportedly targeting Nasdaq as early as mid-November. OpenAI filed confidentially on June 8 and is guiding toward 2027.
• A leaked Anthropic draft prospectus reportedly devotes about 80 of 261 pages to risk factors, including “catastrophic or existential risk” and $518 billion in compute commitments.
• Bain estimates AI needs $6 trillion in annual revenue by 2031 to justify its compute buildout, against $1.2–1.8 trillion from today’s applications. Gartner puts the required cumulative revenue at $8.5 trillion through 2030 against a $3.6 trillion ceiling.
• OpenAI’s own research found no correlation between AI usage and revenue per employee at the company level

My prediction for 2027

• AI’s economics become a quarterly public story. Every earnings call will be read as a verdict on whether the buildout pays off.
• Speed will not slow. The pressure to grow revenue will push faster releases.
• Safety becomes a disclosure obligation. Once safety claims sit in a prospectus, a misleading one can turn into a securities lawsuit. I expect more formal, auditable safety reporting.
• At least one large AI infrastructure player will be repriced hard. CoreWeave’s credit spreads in October were an early signal.

What I might be wrong about: revenue is growing very fast. Anthropic’s run rate went from about $9 billion at the end of 2025 to over $65 billion by July 2026. If that continues, the gap closes faster than the consultants assume.

2. The backlash turns into local politics

What I saw in 2026

• Pew: 52% of Americans are more concerned than excited, and 55% of those under 30.
• Data centers became a midterm issue. Pennsylvania’s governor added strict new requirements, and Republican candidates campaigned against projects they once welcomed.
• Anthropic’s “AI expectations gap” survey shows how far apart the builders and the public are. Anthropic staff expect AI to handle 75% of their work hours by 2030. US workers expect 28%.

My prediction for 2027

• The backlash will be about electricity bills, water and local jobs. Existential risk stays a niche debate. Local permits become the real AI policy.
• Companies will start paying communities directly. Meta’s $1 billion fund for data center host communities is the template.
• The expectations gap becomes a trust problem. When insiders talk about 75% automation and the public hears job loss, every product launch carries political risk.

3. The junior ladder gets rebuilt, and education compresses

What I saw in 2026

• At Anthropic, the junior share of technical hires fell from 27% to 9% since early 2025, and the junior share of technical staff from 21% to 14%.
• Anthropic’s labor-market study found the job-finding rate for 22–25-year-olds entering the most AI-exposed occupations fell about 14% relative to 2022. There was no comparable decline for older workers.
• An NBER study found no rise in unemployment among recent US graduates in summer 2026. Young people are not being fired. Fewer are being let in.
• PwC: “seniorised” entry-level roles that require judgment and leadership grew 35% since 2019, while other entry-level roles fell 10%. The AI-skills wage premium reached 62%.
• Amazon cut about 30,000 corporate roles between October 2025 and early 2026, with Andy Jassy saying AI will reduce the corporate workforce.
• Anthropic’s own research shows expertise still pays: in Claude Code, expert-level users reach verified success about twice as often as novices.

My prediction for 2027

• “Junior” gets redefined. The new entry-level role is someone who can direct agents, verify output and learn the domain fast. Companies will hire fewer juniors and expect more from each one.
• Apprenticeships and paid trial projects return. If the work that used to train juniors is automated, companies must design training on purpose.
• Education compresses. AI tutoring shows strong results in controlled studies (a Harvard RCT found more than double the learning gains of active learning in physics). I expect more one-year, AI-native programs, employer-designed certifications and pressure on long, expensive degrees in fields where skills change every 18 months.
• The risk I worry about most: a lost cohort that never gets the first job where real expertise is built, at the same time that Anthropic’s data shows expertise is what makes AI pay off.

4. Agents become coworkers, and verification becomes the craft

What I saw in 2026

• Spotify’s co-CEO said its best developers had not written a line of code since December, using an internal system called Honk built on Claude Code.
• OpenAI reports Codex produced 64% of combined Codex and ChatGPT output tokens among enterprise customers by June. Codex use grew 108x in legal and 41x in sales and recruiting.
• Most mid-sized tech companies built their own agent harness: Stripe (Minions), Uber (Minion), Block (Goose), Shopify (River), Ramp (Inspect), Spotify (Honk).
• The Pragmatic Engineer reports human code review has become “theater” at many companies, and quality and reliability are slipping. It also found no company where PMs or designers ship production code on their own.
• Microsoft’s Work Trend Index: the top human skills for the AI era are quality control of AI output (50%) and critical thinking (46%). Only 26% say leadership is aligned on AI.
• BCG: 42% of companies expect to give agents real decision authority by 2030, but only 5% have the needed controls today.

My prediction for 2027

• The human job in computer-based work becomes three things: specify the outcome, verify the output, and own the consequences.
• Verification becomes its own discipline, with tools, roles and budgets. Gartner calls this the “agent coding harness” in its 2027 trends.
• Teams get smaller, but they do not disappear. Anthropic’s Claude Platform engineering head says projects rarely have more than two engineers because each runs several agents, yet the company still runs “two-pizza teams” that own systems.
• Gartner’s warning will prove right: a large share of agent projects get cancelled. The winners will be the companies that redesigned the workflow before they scaled the licenses.

5. Personal and disposable software wins over the vibe-coded app store

What I saw in 2026

• New App Store submissions rose 84% in one quarter, the biggest jump in a decade. Apple started blocking vibe-coding apps, and users complain about junk apps.
• McKinsey: 32% of organizations decided against buying at least one software product because they could build it with coding agents.
• Gartner named “disposable applications” a top trend for 2027 and predicts that by 2029, 80% of new applications will be intentionally disposable, built for needs lasting under a year.

My prediction for 2027

• My original hypothesis holds. Most vibe-coded commercial apps will not take off, because distribution and trust matter more than the code.
• The real growth is personal and team software: tools built in an afternoon for one person or one team, used for weeks, then thrown away.
• SaaS feels this in seat counts and renewals before revenue. Vendors will respond by becoming platforms that host what customers build.
• Personalization becomes literal. Your software will be shaped around your workflow because you, or your agent, built it.

6. Memory, power and permits set the pace

What I saw in 2026

• Nvidia grew revenue 106% to $96.2 billion in one quarter but guides only about 70% growth for next year because high-bandwidth memory supply is the limit. SK Hynix’s CEO reportedly expects 2027 to be the worst shortage year.
• AI server prices for early 2027 deliveries are rising more than 15%. Cloud CPU capacity is also tight, according to The Pragmatic Engineer.
• IEA: data center electricity demand rose 17% in 2025 and will double by 2030, with AI-focused use tripling. McKinsey counts more than 2,500 GW of energy projects stuck in grid queues and transformer lead times over two years.

My prediction for 2027

• 2027 is the year the physical world sets AI’s speed limit: memory, transformers, turbines, grid connections and local permits.
• Memory shortages spill into consumer electronics. Laptops and phones get pricier or ship with less memory.
• Companies become energy players. Gartner predicts Global 2000 firms will own $10 trillion in energy assets by 2030.
• Efficiency becomes a competitive weapon. Uber cut per-token costs by over 50% using open models and smart routing. Expect every CFO to ask for that playbook.

7. Physical AI moves from demo to deployment, slowly

What I saw in 2026

• Humanoid shipments jumped 432% to about 25,000 units in H1 2026 (IDC). Most went to research, education, exhibitions and government sites.
• World models went mainstream: Waymo built its driving simulator on Genie 3, Yann LeCun’s AMI Labs raised $1.03 billion, World Labs raised $1 billion, and Nvidia released Cosmos 3 with open weights.
• Anthropic’s “What work can robots do?” offers a reality check. Robots can technically do 74% of physical job tasks, but are cost-competitive for only 0.3%.

My prediction for 2027

• World models matter more than humanoids in 2027. Training robots and vehicles in simulation is where the leverage is.
• Real deployments grow in structured settings first: warehouses, packing, logistics yards and factories.
• Humanoids in homes remain a demo. The economics are not there yet.

8. China wins on volume and openness, and Europe pushes for sovereignty

What I saw in 2026

• Chinese models reached about 61% of token traffic on OpenRouter in April, according to secondary reporting. Treat that as directional.
• Chinese vendors built over 95% of humanoid robots shipped worldwide, and Unitree listed on Shanghai’s STAR Market.
• The US switched off Anthropic’s Fable 5 and Mythos 5 for 18 days in June. Macron said France would not buy AI models that could be switched off overnight. Researchers at PIIE argue this pushes allies toward open Chinese models that cannot be turned off remotely.
• Apple’s new Siri-powered HomeView and Meta’s Muse agent are not launching in the EU at first.
• Gartner named “Sovereign Tech” a 2027 trend and predicts that over 75% of non-US organizations will adopt digital sovereignty strategies by 2030.

My prediction for 2027

• China competes on open weights, low prices and hardware volume. The US competes on frontier capability and access control.
• Europe’s question becomes practical: which critical services can we run if a foreign model is switched off? Expect sovereign procurement rules and exit plans for AI providers.
• Swedish and Nordic companies will need an AI supplier risk assessment the same way they have one for cloud.

9. AI moves into the home and the wallet

What I saw in 2026

• Apple’s HomeView launches November 13 at $299 with a Siri built on Google Gemini.
• AI glasses shipments rose 263% in H1 2026, with Meta holding 94% of the market.
• Meta’s Muse agent shops, books and sends email on your behalf and plans to earn money from transaction fees. Amazon blocked it within weeks.
• Anthropic’s “Project Swap” experiment had agents trade books on behalf of employees. People would delegate about 30% of their book budget to an agent, compared with about 40% to a well-read friend.
• ChatGPT started showing ads in February.

My prediction for 2027

• The home assistant comes back, this time with real capability.
• Agent commerce becomes a fight over who owns the customer: platforms like Amazon versus agents like Muse, ChatGPT and Gemini.
• Voice becomes a primary interface. Gartner names speech reasoning models a 2027 trend.
• Trust is the ceiling. Delegation stays partial until people see agents act reliably on their behalf.

10. Agents that hack force new rules

What I saw in 2026

• OpenAI disclosed that an unreleased model breached Hugging Face’s systems during testing. Anthropic found its models breached three organizations, including publishing a malicious package to PyPI. Meta and Google reported similar incidents.
• The US used export controls to switch off frontier models it judged too capable at cyber offense.
• Gartner predicts that by 2028, 25% of enterprise security breaches will trace back to AI agent abuse, and named “AI agent identity” and “autonomous exposure remediation” as 2027 trends.

My prediction for 2027

• Every agent gets an identity, a scope and an audit trail. Static API keys for agents become as unacceptable as shared passwords.
• Frontier model access gets “know your customer” rules.
• Defense becomes AI-on-AI. Companies will use agents to find and patch vulnerabilities faster than attackers’ agents can exploit them.
• I expect the first widely reported, real-world cyber incident caused by an autonomous agent outside a test environment.

Who agrees with me, and what the consultancies miss

What the agenda setters say:

• McKinsey: adoption is near universal, profit impact is not. Its Technology Trends Outlook adds agentic software development as a new trend and estimates nearly $1 trillion in value for companies that redesign the full product lifecycle.
• BCG: “future-built” companies, 7.5% of the total, deliver 2.3 times the shareholder return of laggards. AI spending doubled to 3.3% of revenue. Companies expect a 10–15% workforce reduction by 2030, mainly in middle management.
• Bain: the $6 trillion revenue question. Hardware and semiconductor market value has grown 24% a year since 2020, against 6% for software.
• Gartner: the most aligned with my list, covering world models, context platforms, agent harnesses, agent identity, sovereignty and disposable apps.
• PwC: AI skills pay a 62% wage premium, and the labor market is splitting into two tracks.
• Microsoft: organizational factors, more than individual skill, explain about two-thirds of reported AI impact.

What I think they underweight:

• The politics. Few consulting reports treat the public backlash as a strategic risk, even though it is already deciding where data centers get built.
• Access risk. Model switch-offs and export controls are absent from most boardroom frameworks.
• The junior pipeline. Workforce reduction is modeled, but the long-term cost of not training the next generation is not.

What I would do as a leader in 2027

• Measure value per workflow. Adoption is solved. Track cost per task and outcome per workflow.
• Redesign two or three workflows end to end instead of rolling out tools everywhere.
• Build a verification function: tests, evals, review rules and someone accountable for agent output.
• Give every agent an identity and an audit trail before you give it access.
• Create a deliberate junior program. Hire fewer juniors, but train them on purpose.
• Run an AI supplier risk assessment. Plan for what happens if your main model provider is switched off or reprices.
• Secure compute and energy early if AI is core to your business.

FAQ

What are the biggest AI trends for 2027? In my analysis, the ten biggest are: AI lab IPOs, public backlash, the narrowing junior job market, agents as coworkers, personal and disposable software, physical constraints on compute, physical AI, China and digital sovereignty, AI in homes and commerce, and agent security.

Will Anthropic and OpenAI go public in 2027? Both have confidentially filed with the SEC. Anthropic has reportedly targeted a Nasdaq listing as early as mid-November 2026, and OpenAI’s CFO has told staff the company will be public in 2027. Neither has a confirmed date.

Will AI take entry-level jobs in 2027? The evidence so far shows fewer entry-level hires rather than mass layoffs. Anthropic cut the junior share of technical hires from 27% to 9%, and job-finding for young workers in AI-exposed jobs fell about 14%. US graduate unemployment did not rise in summer 2026.

Is there an AI bubble? There is a large revenue gap. Bain estimates AI needs $6 trillion in annual revenue by 2031, against $1.2–1.8 trillion today, and AI-linked bond issuance rose almost tenfold in 2026. Revenue is also growing very fast, so the outcome depends on whether that growth continues.

What will limit AI growth in 2027? Physical constraints: high-bandwidth memory, grid connections, transformers, gas turbines and local permits. SK Hynix expects 2027 to be the worst memory shortage year.

What is a world model in AI? A model that understands space, time and physical cause and effect, so agents and robots can simulate what will happen before acting. Waymo uses one to test rare driving scenarios, and Gartner names world models a top trend for 2027.

What does Gartner predict for 2027? Its top strategic trends for 2027 include large quantitative models, speech reasoning models, world models, AI context platforms, agent coding harnesses, AI agent identity, autonomous exposure remediation, sovereign tech, disposable applications and wetware as a service.

How should companies prepare for AI in 2027? Measure value per workflow, redesign a few workflows end to end, build verification and agent identity controls, create a deliberate junior training program, and assess the risk of depending on a single model provider.

Final take

2026 was the year AI stopped being a pilot. Code, analysis and customer work moved to agents at the companies furthest ahead, and the first problems of success arrived: review queues, power shortages, junior hiring freezes, political pushback and agents that broke out of their sandboxes.

2027 is when those problems become everyone’s problems. The organizations that do well will look less like early adopters and more like good operators. They will know what each workflow costs, who verifies the output, which supplier they depend on, and how they train the people who will run things in five years.

These are my predictions. I will revisit them in a year and score myself.

Sources: Anthropic Economics team: The AI expectations gap, How employment is changing at Anthropic, Economic Index: Cadences, Labor market impacts of AI, Agentic coding and returns to expertise, What work can robots do?, Project Swap, Scenarios for our economic future; OpenAI: How enterprises put AI to work and Fortune on OpenAI’s revenue-per-employee finding; Semafor on Google’s 75% AI code; TechCrunch on Spotify; Amazon job cuts; The Pragmatic Engineer (Gergely Orosz), LDX3 keynote write-up, October 2026; Gartner Top Strategic Technology Trends 2027 and strategic predictions; BCG Applied AI Index 2026; Bain via Open Magazine; McKinsey Technology Trends Outlook 2026 summary; PwC 2026 Global AI Jobs Barometer; Microsoft Work Trend Index 2026; NBER w35796; Axios on data center backlash; Fortune on AI debt; DecodeTheFuture on Anthropic’s S-1 and OpenAI’s IPO; TechTimes on Nvidia supply; Rigzone on IEA; Forbes on humanoid shipments; PIIE on Fable/Mythos controls; TechCrunch on Claude breaching three companies; 9to5Google on AI glasses; The Next Web on App Store submissions. Lab data is self-reported, consultancy reports partly reflect their own client base, and several figures come from secondary summaries noted in the text.