Anthropic yesterday announced a $65 billion fundraising, bringing its valuation to $965 billion. At first glance, the operation confirms the intact appetite of the markets for the champions of artificial intelligence, however, behind this record financing, Anthropic above all reveals the new economic reality of AI. The main challenge is no longer just designing the best models, but ensuring access to the physical resources to train and operate them.
For more than a decade, technology has been defined by software. Artificial intelligence reintroduces it to the world of physical constraints.
A valuation that finances global infrastructure
With annualized revenue now estimated at $47 billion, Anthropic operates in a category that no longer has much in common with that of traditional startups. Few companies in software history have reached this level of revenue so quickly. A journey which partly explains the valuation of 965 billion dollars granted by investors.
But what sets this fundraising apart from previous ones is not only its amount, but what it will be used for. Thus, the capital raised will not only be used to recruit researchers or develop new generations of models. They will be used to secure the infrastructure necessary to meet demand which is now growing faster than available capacity.
The real announcement: up to ten gigawatts of capacity
Anthropic announces that it has reached agreements representing up to five gigawatts of capacity with Amazon, to which are added five gigawatts of next-generation TPU with Google and Broadcom. The company also benefits from access to Colossus GPU infrastructures developed by SpaceX.
The appearance of the gigawatt as a strategic unit probably constitutes an indicator of the ongoing mutation. Artificial intelligence is gradually ceasing to be a primarily software activity and becoming a heavy industry. The situation is more reminiscent of the major phases of industrialization in telecommunications or cloud computing than the usual software cycles.
Semiconductors become as important as models
The other major lesson from this operation lies in the presence of industrial partners such as Micron, Samsung and SK hynix.
These groups occupy a central position in the global chain of components used for artificial intelligence. Their participation illustrates a profound evolution of the competition. For many years, competitive advantage relied primarily on algorithms. Today, the availability of chips, HBM memory and network infrastructure is becoming an equally determining factor.
The AI industry is gradually discovering the realities that industrial sectors have long faced: growth depends as much on the supply chain as on innovation itself.
This development explains why memory manufacturers, component producers and infrastructure operators now occupy a growing place in the AI laboratory ecosystem.
The rapprochement between venture capital and infrastructure capital
The list of investors also illustrates this change. Thus, alongside big names in venture capital like Sequoia, Greenoaks or General Catalyst, we now see Blackstone, Brookfield, Temasek, GIC and Capital Group.
This mix would have seemed improbable just a few years ago. The former traditionally finance technological innovation. The latter usually invest in energy networks, transport infrastructure, real estate assets or equipment essential to the global economy.
Their presence in the same round reflects the way in which the markets now perceive artificial intelligence.
Even Mistral enters the gigawatt economy
This dynamic does not only concern the American giants. In Europe too, AI players are gradually facing the same constraints.
Mistral AI reiterated yesterday its intention to deploy 200 megawatts of capacity by the end of 2027 before reaching one gigawatt by 2030. If the orders of magnitude remain far from those mentioned by Anthropic, but the strategic logic is identical.
The French company no longer only seeks to develop competitive models, but seeks to secure the resources necessary for their operation and to build part of the European AI infrastructure.
This development is particularly revealing of a change of era; until recently the debate pitted American players with considerable resources against European players forced to focus on efficiency. From now on, everyone is converging towards the same industrial reality.
An AI war that becomes an energy war
Anthropic is not alone in this race. OpenAI is increasing initiatives around Stargate in order to secure massive computing capacities. xAI continues the expansion of Colossus to support the development of Grok. Meta plans to invest several hundred billion dollars in its artificial intelligence infrastructure over the next few years.
AI enters its industrial age
Anthropic’s $65 billion fundraising marks an important step in the evolution of the sector. The first phase of AI was dominated by research, the second by the race for models, the third is now that of infrastructure.
The most advanced laboratories look less and less like startups. They secure energy capacity, negotiate long-term contracts with component manufacturers, reserve data centers and build industrial alliances on a scale rarely seen in the software industry.
