2026: When AI Hits Reality

What happens when AI hits physics, economics, and reality

I started the year recording the Nerd Show podcast about predictions for 2026. It was fun to catch up with the guys.

After the microphones were off, I kept thinking about what didn’t quite make it into the conversation. Not because it was controversial, but because it was harder to package.

So this is that second pass.

This isn’t forecasting. It’s more like watching a wave that’s already formed and guessing where it breaks. Anticipating those breaks is part of my role at PIRATE.global, helping position e.g. division5, Visualmakers and FunnelIn for where the market is heading.

This post became too long, so I split it in two. This first part focuses on AI.

Some of this will be wrong. The useful question isn’t whether these things happen exactly as described, but whether the underlying tensions are real. I think they are. Tell me what you think in the comments.

The context shift no one is arguing about anymore

We’re still debating how smart models are getting. That argument already feels stale.

The real shift isn’t intelligence. It’s gravity. AI has moved from experimentation to economic force. It now pulls on budgets, infrastructure, pricing, regulation, and power. Once something becomes part of the cost structure rather than a demo, different questions matter. Reliability beats cleverness. Integration beats novelty. Economics beat vibes.

That shift changes where value and risk live. When AI systems become operational, the biggest failures don’t come from models being too dumb. They come from systems not knowing what they’re supposed to know - or remembering it inconsistently.

That’s why context matters more than IQ.

Not abstract “world knowledge,” but the messy, accumulated knowledge organizations actually run on. The individual and collective knowledge people hold. And the externalized knowledge no one quite owns anymore. Documents, decisions, defaults, exceptions, half-forgotten reasoning, institutional memory. The boring stuff spread across folders, tools, emails, and old tickets.

By 2026, the winning systems won’t be the cleverest. They’ll be the ones that can reliably remember what an organization already knows, keep that memory current, and do it without collapsing under cost or complexity.

Intelligence without context is a demo. Context without runaway costs is the product.

Companies don’t want brilliance that hallucinates and forgets - especially in high-stakes workflows. They want systems that are predictably useful. Better something that’s reliably 80% right than something that swings wildly between genius and error.

The obvious shift from rankings to citations

Search behavior is changing faster than most admit. People still type queries, but increasingly they get synthesized answers instead of lists of links. The web is flooding with AI-generated content - much of it interchangeable, shallow, and forgettable. In that noise, traditional SEO is becoming outdated fast.

It was already visible in 2025. By 2026, the game isn’t about ranking high on a results page. It’s about being the trusted source that generative engines cite when they compose answers. Generative Engine Optimization - GEO - rewards clarity, authority, and signals of real expertise over keyword tricks or volume. Low-effort AI slop gets ignored. Content that demonstrates genuine experience, backed by consistent signals across platforms, gets surfaced and quoted.

Trust isn’t optional anymore. It’s the new moat.

Models keep getting cheaper. Context doesn’t.

Large language models are drifting toward commodity status. Open models are good enough for many tasks. The real competition is moving upstream, into integration, retrieval, and personalization.

This creates a different kind of lock-in. Not through proprietary models, but through accumulated context. Your workflows. Your data. Your defaults.

Privacy concerns rise alongside this shift. The more context a system holds, the more people worry about where it lives and who controls it. That tension creates space for smaller, expert systems and for tools that are opinioned about what they remember and what they forget.

Apple’s quiet advantage

Putting Gemini as the underlying LLM for Siri in 2026 seems like a capitulation. Like so often, Apple will take its time to get things right. They eventually will launch capable models directly on devices people already trust and carry.

They don’t need the smartest system in the world. They need one that feels private, predictable, and already there. Distribution and trust do a lot of work when anxiety is high. In 2026, we’ll realize that that tradeoff starts to look less like a compromise and more like a strategy. Especially as competitors grapple with regulatory scrutiny and potentially data scandals.

When progress runs into physics

Robotics demos will keep improving. Reality will keep sorting them.

Industrial robotics will see a real rise. Factories, warehouses, logistics hubs, labs, agriculture - constrained environments where tasks are repeatable and ROI is measurable - will keep adopting robotic automation. This is already happening, and it will accelerate. These systems won’t look like humans. They’ll look like arms, carts, gantries, drones, and purpose-built machines designed to do one thing extremely well. Europe will play a strong role here, because this is where precision manufacturing, industrial buyers, and integration expertise actually matter.

Consumer robotics is a different story. Outside of narrow use cases, progress will remain slow and underwhelming. General-purpose humanoids are compelling demos, but they collide head-on with safety, maintenance, cost, and edge cases the moment they leave controlled environments.

Optimus - if it launches in 2026 - will disappoint expectations. Robotaxis won’t hit the streets at scale. Others won’t do much better. Not because the models are weak, but because hardware doesn’t scale like software. Manufacturing, servicing, liability, and the real world pile up fast. Reality is messy. Hardware is hard.

Robotics will grow. It just won’t grow where the hype points.

The SaaS pricing story starts to break

Per-seat pricing made sense when software was passive and humans did the work. That logic breaks when AI does half the job.

Companies are already paying for users who barely touch the tools they’re licensed for. At the same time, real work is happening elsewhere, inside automations, scripts, and background systems no one counts as a “seat.”

That tension won’t resolve quietly. Companies have tried to price AI features on top of per seat pricing. That will only be temporary. Pricing will shift toward outcomes, usage, and completed work. Not everywhere, not all at once, but visibly.

You won’t pay for access. You’ll pay for invoices processed, tickets resolved, reports delivered.

Procurement will resist. Finance will push back. But the direction is set.

Single choke points start to feel reckless

Right now, much of the AI industry depends on NVIDIA as the main supplier for foundational infrastructure. That concentration worked when speed mattered more than resilience. It’s also a vulnerability that won’t last.

This isn’t about nationalism or grand strategy. It’s operational realism. When a single supplier, a single energy source, or a single regulatory regime can bottleneck your entire system, sovereignty stops being a slogan and becomes an engineering requirement. By 2026, diversification won’t be framed as optional resilience. It will be baseline competence.

This doesn’t end in a sudden collapse. It ends in diversification. Custom chips. Second-best options that are good enough. Margins will get compressed. Power spreads out, slowly and unevenly.

Reliance on a single choke point is a risk no serious operator can justify.

The uncomfortable economics underneath the hype

The era of AI theater is ending.

For the last few years, it was enough to show impressive demos, ship fast, and promise scale later. By 2026, revenue becomes the filter. Not growth in abstract usage, not model benchmarks, but money that comes in from customers who keep paying without subsidies or accounting gymnastics.

The big incumbent tech companies like Google and Microsoft fund their AI work from cash flow. OpenAI doesn’t. They keep raising at higher valuations, but the economics look strained.

Compute is expensive. Training is expensive. Inference is expensive. When companies lend each other money to buy each other’s services, revenue can look healthy longer than it really is. Money goes out as credit, comes back as revenue. It shows up in earnings as “growth.” This works until it doesn’t.

OpenAI isn’t fully vertically integrated and doesn’t have a strong moat besides brand recognition. That makes them vulnerable. Google - for example - has what OpenAI lacks: deep infrastructure, control over their entire stack. They have the ability to absorb losses without existential risk and can outlast a pricing war. OpenAI can’t. Even xAI will be better positioned than OpenAI because of the proprietary data from X and the real-world data experience of Tesla.

By 2026, the economics start to matter more than the narratives.

Physical limits reassert themselves

None of this is abstract. Data centers consume power and water. Communities notice. Grids strain. Costs rise. Projects will get delayed or blocked entirely.

For a long time, these were background concerns. Externalities you could smooth over with enough capital.

By 2026, energy and water constraints stop being background issues and start shaping what’s viable. AI doesn’t escape physics. It runs straight into it.

Distribution starts to creak

App stores still matter. Copycats and low-effort clones will flood the landscape, and centralized distribution still helps in a noisy world.

But the business model starts to strain. When people generate tools or games on demand, publish as web app, run apps locally, or deploy internal software without ever “shipping” it, browsing a store to find software starts to feel outdated.

The pressure won’t break the app store overnight. But the shift will be visible.

What holds across all of this

A few patterns cut across everything here.

Creation gets cheaper, maintenance gets more expensive. That gap produces both opportunity and fragility. Speed alone doesn’t guarantee success; how you sustain, integrate, and steward what’s built matters just as much.

Physical constraints still matter. Energy, water, raw materials, and logistics don’t care about code. They set hard boundaries on growth and ambition.

Strategic leverage matters more than size. Companies that survive these pressures won’t just be big - they’ll control critical context, infrastructure, and relationships with regulators and distributors. Optionality comes from control - over context, infrastructure, and relationships - not just size or liquidity.

Leverage replaces headcount. Creation keeps getting cheaper, but judgment, integration, and maintenance remain scarce. The winners won’t be the smallest teams or the largest, but the ones where every hire multiplies output rather than adding coordination drag. This principle has always held - but AI makes its impact far more visible.

Trust becomes scarce. Not because people are cynical, but because volume overwhelms evaluation. Familiarity becomes the shortcut for decisions, partnerships, and adoption.

Technical possibility far outpaces economic value. The gap between what can be done and what actually works - and what pays - remains wide.

None of this means we should slow down or speed up.

It means 2026 is the year these tensions stop being theoretical. The next layer of impact isn’t technical. It’s human and institutional. What happens when systems accelerate faster than organizations, labor markets, and leaders can adapt is a different problem entirely. That’s where the real friction shows up.

Would love to hear your take.

🙏

Be kind,

Manuel

PS: In part two, I’ll look at what this shift does to trust, talent, leadership, geopolitics, and individual agency - once AI stops being the interesting part.