2026 Compute Power Siege: Major moves by NVIDIA and Microsoft on the same day signal a life-or-death battle over chips and electricity

Author | Eric

The two major events that occurred on the same day sent a clear signal:Computing power remains extremely scarce.

$NVIDIA (NVDA.US)$ It was disclosed that shares were purchased at $87.20 per share $CoreWeave (CRWV.US)$ Class A stock (approximately 23 million shares), with a total investment of $2 billion. According to Reuters calculations, this nearly doubled NVIDIA's previous ownership stake. CoreWeave's stock price surged accordingly.

On this very day, $Microsoft (MSFT.US)$ Microsoft has launched its latest self-developed AI inference chip, Maia 200. Microsoft claims that the cost-performance ratio of the new chip has increased by 30%.

If you connect these two events, you can understand the ultimate truth of 2026:Stop talking about 'cloud narratives.' This is essentially a life-and-death competition about 'who has the silicon chips, who has the electricity, and who can run faster.'

Why is NVIDIA increasing its investment in CoreWeave?

NVIDIA and CoreWeave have defined this expansion as a sprint, aiming to build more than 5GW of AI factories before 2030.

For CoreWeave investors, the key focus is onallocation rights. CoreWeave will deploy multiple generations of NVIDIA infrastructure through early adoption, explicitly including the Rubin platform, Vera CPU, and BlueField storage systems.

In an era where top-tier NVIDIA chips are subject to rationing, gaining an edge in the product cycle goes beyond the scope of marketing. It builds a defensible economic moat.

The启示of Microsoft's Maia 200: The strategic truth behind self-developed chips

Microsoft's announcement about Maia 200 is filled with classic 'cost efficiency' jargon. Anyone familiar with Microsoft's previous overpromising on Maia won't take it too seriously. However, the underlying message is more significant: hyperscale cloud providers are eager to gain control. This is essentially a 'massive insurance policy' against supply risks.

When NVIDIA's supply is tight, in-house chip development becomes the ace up the sleeve for dealing with 'nightmare scenarios.' The so-called nightmare refers to a sudden explosion in demand while you're left scrambling because Neocloud and others are aggressively hoarding resources, leaving you unable to secure enough GPUs.

Deployment speed always trumps theoretical cost.

Microsoft claims that Maia 200 is purpose-built for inference, surpassing the previously hyped $Alphabet-C (GOOG.US)$ latest TPUv7, which boasts 216GB HBM3e memory and 7TB/s bandwidth. Even if all these specs hold true, its real strategic value lies in Microsoft's ability to scale a fully controlled platform at its own pace.

But reality often proves otherwise.

The Information revealed that Anthropic’s actual per-unit inference costs on $Alphabet-C (GOOG.US)$ TPU and $Amazon (AMZN.US)$ ASIC chips, held by its two major shareholders, were about 23%higher than expected. This directly slashed Anthropic's gross margin forecast for 2025 to around 40%, a full 10 percentage points lower than internal estimates. Following this, Anthropic began seeking direct GPU collaboration with NVIDIA.

What does this prove? It proves that 'self-built alternatives' can also face expensive real-world challenges. There has never been a guarantee for low-cost paths.

So, don’t blindly believe the corporate PR pitch about how 'ASIC chips save money'.By 2026, the core purpose of self-developed chips will be just one thing: ensuring you can secure enough computing power. Whether it saves money or not is another story.

Who should we keep an eye on next?

CoreWeave is the most direct 'NVIDIA's favorite child' concept stock, but these Neocloud players are also worth watching:

– $NEBIUS (NBIS.US)$ ：NVIDIA also holds shares in it. Previously, NEBIUS signed a $17.4 billion agreement with Microsoft and later disclosed a five-year $3 billion deal with Meta. This is a classic example of giant endorsements combined with heavy asset investments.

– $Applied Digital (APLD.US)$ ：It’s more like an 'AI data center landlord.' Although not purely a software company, it still benefits from scarcity dividends. According to Reuters, a 15-year lease at its Polaris Forge 2 campus is expected to generate approximately $5 billion in revenue, with CoreWeave as the key client.

– $IREN Ltd (IREN.US)$：A typical player transitioning from mining to Neocloud. Their AI cloud scale has just doubled to 23,000 GPUs, and they’ve raised their Q1 2026 ARR revenue target to over $500 million.

Summary

The screening criteria for this industry may be dull, but extremely effective. The final winners must possess three aces:A clear ticket to the next-generation NVIDIA platform, reliable power supply pipelines, and the ability to raise funds without diluting shareholder equity during volatile periods.

210K Views