AI and Economic Growth: What the 20% GDP Boost Means for Swiss Decision-Makers

The number making the rounds

Elon Musk put it bluntly at a recent conference: 20 to 30 percent additional global economic growth from AI is merely the "floor" - the digital component alone. Goldman Sachs provides concrete figures on near-term GDP effects: in the first half of 2025 alone, AI-related spending accounted for roughly one percentage point of US GDP growth - without it, the US economy would have been essentially flat. For 2026, the bank expects global AI investment to exceed one trillion dollars, with $600 billion in the US alone - close to 2 percent of US GDP.

Two speeds: software productivity vs. physical automation

Musk distinguishes two waves. The first is digital: software agents that write, answer calls, analyse data, and prepare decisions. This wave is already running - Starlink's Grok Voice reportedly handles around 15,000 calls per day, 70 percent of them fully autonomously, without human intervention. The second wave is physical: robots, autonomous vehicles, humanoids in factories and warehouses. Musk talks about one billion humanoid robots by the 2030s - equivalent to five times today's human economic output. In his view, this second wave will eventually outpace the first, but it arrives later.

• Digital AI: knowledge work, customer service, analysis, reporting - scalable within months
• Physical AI: manufacturing, logistics, care, construction - scalable within years, bound to hardware cycles

What the research actually shows (and where economists disagree)

Goldman Sachs' own original 2023 forecast was considerably more cautious than Musk's 20-30 percent: plus 7 percent global GDP over ten years, plus 1.5 percentage points of productivity. McKinsey estimates annual value creation of $2.6 to $4.4 trillion across 63 concrete use cases, with a productivity contribution of 0.5 to 3.4 percentage points through 2040. MIT economist Daron Acemoglu is considerably more sceptical: he sees only around 5 percent of all tasks as genuinely automatable within ten years - which would translate into a GDP effect closer to 1 percent, not 7. Goldman Sachs counters that Acemoglu's model underestimates how broadly the technology will eventually spread.

Even the most conservative serious estimate - one percent of additional economic growth from AI - would still translate into billions in value for an economy the size of Switzerland's.

The Swiss reality: dense talent, cautious adoption

Switzerland holds a genuine locational advantage: with 110 AI researchers per 100,000 inhabitants, no other country has a higher research density. But there is a gap between research and application across the broader economy. According to a survey by HWZ and Swisscom, 34 percent of Swiss SMEs already use AI today - yet the large majority of them, 76 percent, remain at a novice level: isolated tools, no end-to-end integration. A strategic look at how Swiss SMEs can avoid the three biggest mistakes in the global AI race explains why exactly these isolated solutions become a liability once competitors scale faster.

Why waiting costs more than an imperfect start

Accenture puts it plainly: the pilot phase is over, what matters now is scaling. Companies still experimenting with isolated test projects while competitors automate processes end-to-end are not losing ground gradually - they are losing it at an accelerating rate. Deloitte Switzerland puts the return for companies deploying AI systematically at CHF 3.70 per franc invested.

It is also worth noting that later entrants are not automatically at a disadvantage. Companies starting now benefit from more mature tools and the learning curves of early movers. The real risk lies not in the timing of entry but in the depth of integration once you begin - provided the approach is structured rather than a series of unconnected experiments.

The first step: not technology, but clarity about the lever

The 20-30 percent figure is a global order of magnitude, not a promise for your company. What matters for you is the shift that is already measurable: tasks that consume hours today can be handled by the right agents in minutes - without expanding your headcount. The biggest mistake is not starting too early - it is waiting too long for a perfect blueprint that will never exist in the first place.