The former victim in the AI race, Google, which only months ago looked like it could be eaten alive by OpenAI’s next-gen AI chatbot, is suddenly storming back and, in a twist, pushing Nvidia dangerously close to the cliff as it takes the lead in a way few saw coming.

To put things rapidly into context: the past two years haven’t been rosy for Google. Gemini took time to lift off, hallucinated and became a punchline in its early days. The model quietly improved its way through end users until Gemini 3 hit hard last week. Google eventually got its AI model right and moved aggressively into 3D reasoning, agentic coding and “vibe coding” — the kinds of end-products that could mint billions in revenue.

But that’s not where the story ends.

It’s where it begins.

As AI chatbots seep into everyday life, demand for inference is exploding. Inference is when AI takes your request and figures out an answer. And with it, the cost of inference — the cost of running a trained model every time someone queries it — is exploding too. For OpenAI, that bill for 2024 is projected to hit around $2.3 billion, roughly 15× its training costs.

And here comes the plot twist: Meta and OpenAI are reportedly moving toward Google’s TPUs — Google’s homegrown chips — to run their own models because they are cheaper to operate while offering comparable performance. Both Meta and OpenAI are said to be seeking up to 4× better performance-per-dollar on inference workloads.

And inference is the next big thing because it never stops: every time you chat with a bot, the cost accumulates. Inference costs are projected to make up almost three-quarters of total AI computing costs by 2030.

So the world’s biggest AI players could be shifting toward Google’s TPUs — cheaper, more tailored to AI workloads — and potentially replacing Nvidia.

Read that again.

That’s a real risk for Nvidia, whose client base is nearly half made up of these same Big Tech giants. This is why — on top of the accounting drama that hit the company last week — the stock shed another 2.60% yesterday, while Google rallied to a fresh ATH.

In the meantime, Meta boosted its ad revenue thanks to AI, but its long-term business model is unclear. Meta is spending billions to transform its social media platforms into AI-content platform — a direction that risks disengaging users. Its Llama model is rarely mentioned in enterprise-grade discussions, and its oversized compute spending could backfire. Unlike Google, which can simply rent excess compute through its existing cloud offering, Meta must actually build that business from scratch.

Outside the US, Alibaba’s AI efforts may be paying off. The company announced a stronger-than-expected 34% growth in its cloud business, that helped counterweigh their spending on consumer subsidies and AI investments. But the numbers couldn’t bring investors on board. The share price is struggling to a reverse October – November softness.

In summary, Nvidia is being broadly questioned, Meta may be hitting its potential, while Amazon is the one Big Tech name that could benefit meaningfully from robotics when the time is right.

But right now, Google suddenly seems to have it all: the data, the data centres, the chips, the AI model and the interface. It might well be the next $5 trillion beast. And if you think about it, Alibaba also has many of these assets. It’s got the data, the data centres, its own chips, its AI model, its e-commerce empire, and incredible reach within China and beyond. So if you believe the future is “everything under one roof,” Alibaba is - has always been - a strong candidate.

What about Nvidia? Nvidia has been struggling since its latest earnings blew up in its hands as investors focused on swelling inventories and deferred payments. The company has been compared to Enron, booed because of the Google-TPU news, and are now defending themselves by saying “we’re not Enron” and “we are happy for Google.” Their main argument is that Google’s TPUs are designed for one specific function, whereas Nvidia’s GPUs are compatible with every AI model. But will that matter if companies simply want chips that do the job cheaply and efficiently?

So, the moment has come ladies and gentlemen: competition for Nvidia is arriving from an unexpected direction. That could eat into its revenue potential and market share. Everyone is waiting to see how Nvidia will respond — by expanding customers beyond Big Tech, rolling out more inference-friendly GPUs, or pushing deeper into cost-competitive partnerships. We’ll soon find out.

Meanwhile, US consumer sentiment is waning. More than half of the strong US GDP this year came from massive AI investment. Yesterday’s retail sales and PPI came in soft — softer than expected — although major retailers upgraded their annual forecasts and said the holiday season should look fine.

And if not, the Fed will be there to save the day. The probability of a 25bp cut rose to around 85% after the latest data. The US dollar slipped below its 200-DMA, helping the EURUSD break above the September–November bearish consolidation trend.

Cable also extended gains into today’s Budget announcement— an announcement that might bite. There have been plenty of leaks about where Rachel Reeves will squeeze out £30bn to get the numbers right and keep both markets and households happy. Ultimately, no one will be fully satisfied.

The good news is that stress in gilt markets has been contained over the past few days. The bad news is that yields are near the levels reached during the Liz Truss mini-budget crisis three years ago, and Reeves has the smallest fiscal headroom on record — giving her zero margin for error. After today’s Budget, we’ll have a clearer view on whether the measures will be enough to keep gilt markets tidy and whether they are deflationary enough to convince the Bank of England to cut rates in December — which I think they will be. If so, current levels look appetizing for GBP sellers.