Asia wrap: The AI trade is discovering gravity

• The lack of bottom fishing matters. Investors are still reducing crowded exposure rather than debating whether chip valuations have become attractive.
• TSMC passed the earnings test but failed the market test. When 77% earnings growth cannot lift the stock, positioning has overwhelmed the fundamentals.
• The market is not rejecting AI. It is questioning whether every company in the AI stack deserves premium margins and perfection-level valuations.
• DeepSeek changes the economic argument. Cheap, scalable Chinese models could accelerate AI adoption while compressing the profits earned by premium model providers.
• The profit pool may migrate. Power, infrastructure, distribution and low-cost deployment could capture more value as intelligence moves closer to a utility.
• AI can change the world without rewarding every AI stock. The technology may continue to win even as the most crowded owners of the trade lose.

Discovering gravity

There has been precious little bottom fishing in this positioning unwind so far, which suggests the market is still looking for the exit rather than debating fair value.

The global chip rout rolled into Friday with Taiwan and Japan taking the full force of the storm, while South Korea was spared only because its market was closed for a holiday. MSCI’s Asia Pacific equity gauge fell 2.7%, the Nikkei dropped more than 5%, Kioxia sank as much as 16%, and TSMC lost roughly 4%.

The TSMC move matters most.

The company had just delivered 77% earnings growth, the kind of number that should ordinarily bring bargain hunters charging through the door. Instead, investors kept selling.

When good news cannot lift a stock, the market is no longer trading the news. It is trading the weight of everyone already sitting in the position.

That appears to be where the AI complex has arrived.

The trade spent the first half of the year climbing like a rocket. Now it is discovering that momentum alone cannot keep it airborne. Once the engines sputter, the same crowd that chased the ascent starts calculating the distance to the ground.

Netflix added another crack to the growth story after forecasting a second consecutive quarter of slowing sales growth. Its shares fell 9% in extended trading, Nasdaq 100 futures dropped 1.4%, and European equities were set to open more than 1% lower.

But beneath the liquidation sits a larger question. Investors may not simply be reassessing how much they are willing to pay for AI. They may also be reassessing who ultimately captures the profits.

DeepSeek sits at the centre of that debate.

The Chinese startup has raised $7.4 billion at a valuation above $50 billion and is preparing to attack the global AI market with models priced far below those of its US rivals.

Until now, much of the AI boom has been priced as a scarcity story. Scarce chips, scarce compute and scarce models were supposed to create years of premium margins.

DeepSeek is attempting to turn scarcity into abundance.

Its strategy follows the familiar Chinese industrial playbook: build quickly, scale aggressively, drive down unit costs and force competitors to fight in a market where yesterday’s premium product becomes tomorrow’s utility.

China has run this play before in steel, solar panels, batteries and consumer electronics. It does not need to make the most expensive product. It only needs to make it good enough, cheap enough and abundant enough to change the economics for everyone else.

Cheap AI does not mean lower demand for AI. It may mean far more.

Lower prices could unleash a wave of tokens, applications and corporate adoption. But that demand may not flow toward the same profit pools investors have valued so generously.

Value could shift away from premium model providers and toward low-cost infrastructure, power generation, and distribution, as well as companies capable of operating on thinner margins at enormous scale.

In other words, AI could become bigger while parts of the AI trade become less profitable.

DeepSeek is effectively telling investors that intelligence may eventually trade closer to the cost of electricity than the cost of genius.

China may not possess the world’s best compute at every layer, but it has vast power capacity, state-supported semiconductor development, domestic cloud infrastructure and a government pushing the entire ecosystem in the same direction.

It cannot export cheap electricity directly, but it can convert that electricity into cheap AI tokens and export those instead.

For the market, the message remains in the price action. TSMC delivered the numbers and investors still sold the stock.

The technology has not failed.

The market is questioning the price paid for perfection, the concentration of ownership and the assumption that every layer of the AI stack will earn premium margins.

The technology may still change the world.

The harder question is who gets paid for changing it.