Google Just Reported Its First Negative Cash Flow Quarter.
China's AI Labs? They Are Thriving.

On July 22, Alphabet shared its Q2 2026 results. Revenue hit $119.8 billion. That is up 24% from last year. Google Cloud revenue jumped 82% to $24.8 billion. Operating profit went from $2.8 billion to $8.8 billion. By most old measures, this was a strong quarter.

But the market did not care about revenue. It cared about cash.

Alphabet spent $44.9 billion on AI gear — chips, servers, data centers — in just three months. That is about $490 million per day.

For the first time since going public in August 2004, Google's free cash flow was in the red.

The company also raised its full-year 2026 spending target. Up to $195 billion to $205 billion. That is up from $180 billion to $195 billion earlier. The stock dropped over 4% in after-hours trade. By Thursday it fell over 7%. That wiped out $293 billion in market value in one day.

The Chinese Side: A Totally Different Cost Story

Alphabet is burning cash to build AI gear. Chinese AI labs are playing a different game.

The cost gap is not small. It is built into how each side works.

DeepSeek V4 Flash is one of China's top models. It charges $0.14 per million input tokens. That is cache miss price. Output tokens run $0.28 per million.

DeepSeek V4 Pro runs at $0.435 input and $0.87 output. Cache hits cut the price even more.

Now compare that to Anthropic's Claude Opus 4.8 at $1.80 per task. DeepSeek V4 Pro costs $0.04 per task. That is a 45x gap.

Chinese models are not just cheaper. They are cheaper by whole orders of size.

Morgan Stanley says the memory market will jump. From $220 billion in 2025 to $890 billion in 2026.

Goldman Sachs expects Samsung's regular DRAM price to rise 326% year-over-year in 2026.

The big US buyers are spending billions on hardware. The Chinese labs are spending pennies on software.

The Core Difference

Google's negative cash flow is not a one-time blip. It is a built-in feature of the US AI business model.

US labs build the gear — the chips, the servers, the data centers. They spend billions doing it. Chinese labs build on the gear that already exists. They optimize the software instead.

UBS analysts say the training cost for China's top models is about one-tenth of US rivals. Inference costs sit at 10% to 20%.

The result is two very different money pictures:

- Google: negative cash flow, $205 billion yearly capex, burning $490 million per day

- DeepSeek: $0.28 per million output tokens, no new gear to build, positive per-unit profit

One side has a capital spending headache. The other has an operational spending fix.

The Market's Take

The market has been blunt. Google beat estimates on revenue and profit. But the negative cash flow sparked a sell-off.

Moody's and Goldman Sachs both warned about AI gear spending and its toll on cash flow. Goldman warned of a 'triple negative' — rising costs, falling margins, and a capital spending race.

The big question is no longer whether AI is the future. It is whether today's spending can last.

Google says it has to spend because rivals are spending. 'AI capital spending is a race that cannot stop,' one review put it.

But the math only works if the cash comes in before the money runs out.

What This Means

Google's negative cash flow quarter is a landmark. It shows when the AI spending boom hit the books of the world's richest company.

But it also shows a deep split in how the US and China are playing AI.

The US is building the gear. China is using it.

One side burns cash. The other side burns tokens.

The question is which model will hold up better over the next five years.