AI capex is starting to dwarf the rest of tech. SpaceX’s IPO disclosures suggest its AI infrastructure spending now exceeds its rocket spending, and the four U.S. hyperscalers are projected to spend roughly $725 billion on capex in 2026 alone. China is deploying tens of billions through industrial policy, the Middle East is funding sovereign AI infrastructure, and Canada has committed about CAD $2 billion to a sovereign compute strategy. The frontier AI race is no longer software-shaped, it is starting to look like a national-infrastructure arms race.
For years, the technology industry rewarded speed.
The startup playbook was relatively straightforward:
- move fast
- ship product
- grow users
- scale software cheaply in the cloud
AI is changing that equation.
The recent SpaceX IPO disclosures may have revealed one of the clearest signals yet that the AI race is no longer behaving like traditional software.
According to reports tied to the filing:
AI infrastructure spending exceeded rocket spending.
That is an astonishing statement. (reuters.com)
A company famous for reusable rockets and satellite launches is now spending more aggressively on AI infrastructure than aerospace infrastructure.
That is not just a financial detail.
It is a signal that AI capex is becoming the new global technology arms race.
AI Is Becoming Industrial Infrastructure
For the past two years, much of the AI startup world has focused on:
- prompts
- wrappers
- workflows
- copilots
- agents
- UX layers
Those areas still matter.
But the largest players in AI increasingly appear focused on something else entirely:
- compute
- energy
- networking
- chips
- data centres
- cooling
- infrastructure ownership
The economics are shifting from:
“Who has the best software?”
to:
“Who can afford to build and operate industrial-scale AI infrastructure?”
That is a very different kind of competition.
The Scale of AI Spending Is Becoming Extraordinary
The hyperscalers are now spending at levels rarely seen in the history of technology infrastructure.
Recent reporting suggests:
- Amazon spent roughly $44 billion in quarterly capex
- Google spent over $35 billion
- Microsoft spent over $30 billion
- Meta raised its 2026 AI capex guidance to as much as $145 billion. (finance.yahoo.com)
Collectively, Google, Amazon, Microsoft, and Meta are now projected to spend roughly:
$725 billion on capex in 2026 alone. (finance.yahoo.com)
That number is staggering.
This no longer resembles traditional SaaS economics.
It increasingly resembles:
- telecom infrastructure
- national utilities
- semiconductor fabrication
- industrial manufacturing buildouts
NVIDIA Is Powering the Entire Arms Race
No company better represents the AI capex boom than NVIDIA.
Its latest earnings reflected:
- explosive datacenter growth
- a 92% increase in datacenter revenue
- and Jensen Huang describing AI factories as:
“the largest infrastructure expansion in human history.” (theguardian.com)
The market cap of NVIDIA now reflects something profound. GPUs are no longer just components.
They are strategic infrastructure.
Countries are discussing GPU sovereignty. Startups are raising billions primarily to secure compute access. Entire data centre projects are being designed around AI chip supply.
In some cases, access to GPUs may matter more than product differentiation itself.
OpenAI, Anthropic, and xAI Are All Signaling the Same Trend
At the same time:
- OpenAI is reportedly preparing confidential IPO filings (axios.com)
- Anthropic continues reporting explosive enterprise growth and improving margins
- SpaceX/xAI is aggressively expanding compute infrastructure
One of the most interesting disclosures from the SpaceX filing was that:
Anthropic reportedly agreed to pay SpaceX approximately $1.25 billion per month for AI compute capacity through 2029. (reuters.com)
That single agreement reveals how expensive frontier AI infrastructure has become.
Even highly successful AI companies increasingly depend on industrial-scale compute providers.
Anthropic’s Revenue Growth Changes the Conversation
One of the most important developments in AI may be how quickly enterprise revenue is scaling.
Recent reporting suggests Anthropic could exceed:
- $10 billion in quarterly revenue
- while approaching operating profitability. (reuters.com)
That matters because it validates something important. Enterprises are willing to spend enormous amounts on AI productivity.
But it also exposes the economic tension inside the industry.
The more successful AI becomes:
- the more inference costs grow
- the more GPUs are required
- the more power consumption expands
- the more capital is needed simply to keep up
Success itself creates infrastructure pressure.
China Is Treating AI as National Infrastructure
The AI arms race is not limited to Silicon Valley.
China is aggressively investing across the entire AI stack:
- chips
- semiconductor manufacturing
- data centres
- infrastructure
- domestic AI ecosystems
According to Stanford’s 2026 AI Index, Chinese government guidance funds alone are estimated to have deployed roughly:
$184 billion into AI firms and infrastructure initiatives. (hai.stanford.edu)
China is also pursuing what analysts increasingly describe as a:
“full-stack AI strategy.” (brookings.edu)
That includes:
- domestic chip production
- local AI models
- sovereign cloud infrastructure
- semiconductor self-sufficiency
Chinese chipmakers like SMIC are rapidly expanding manufacturing capacity as global AI demand strains supply chains. (reuters.com)
Meanwhile, China has reportedly set targets to integrate AI into:
90% of its economy within five years. (abc.net.au)
That is not a typical software market strategy.
That is industrial policy.
Canada Is Entering the AI Infrastructure Race
Canada is also beginning to recognize that AI competitiveness increasingly depends on physical infrastructure.
Over the past two years, the federal government has launched a series of major initiatives focused on:
- sovereign compute
- domestic GPU capacity
- Canadian-hosted AI infrastructure
- large-scale national AI data centres
The centrepiece is Canada’s Sovereign AI Compute Strategy, a federal initiative backed by more than:
CAD $2 billion in planned investment. (oecd.ai)
The strategy includes:
- public AI supercomputing infrastructure
- incentives for private-sector compute investment
- programs designed to help Canadian companies access scarce AI compute resources
Budget 2025 expanded those commitments significantly, including:
$925.6 million over five years dedicated to sovereign AI compute infrastructure. (budget.canada.ca)
Canada has also launched:
- the AI Sovereign Compute Infrastructure Program
- the AI Compute Access Fund
- a national process to identify and support large-scale sovereign AI data centre projects. (ised-isde.canada.ca)
Federal discussions around sovereign AI infrastructure now include:
- commercial-scale AI data centres exceeding 100 megawatts
- Canadian-owned compute systems
- large-scale public AI supercomputers intended to support researchers, startups, enterprises, and government. (dlapiper.com)
The Scale Gap Is Still Enormous
Canada’s sovereign AI strategy is ambitious by public-sector standards.
But the global AI infrastructure race is moving at extraordinary speed.
One recent analysis noted:
- Canada committed roughly $2 billion toward sovereign AI compute
- while individual hyperscalers are now spending tens or even hundreds of billions annually on AI infrastructure. (washingtonpost.com)
That creates a real tension. Even national governments may struggle to keep pace with hyperscale AI infrastructure economics.
And that may become one of the defining questions of the AI era:
Can nations maintain meaningful AI sovereignty when frontier infrastructure increasingly requires hyperscale capital?
The Middle East Is Emerging as a Major AI Infrastructure Player
Another major development in the global AI race is the rise of sovereign AI investment from the Middle East.
Countries including:
- Saudi Arabia
- the UAE
- Qatar
are investing heavily in:
- AI data centres
- sovereign AI funds
- semiconductor partnerships
- national AI platforms
The UAE-backed investment firm MGX has already committed billions toward frontier AI infrastructure and model development partnerships. (ft.com)
Saudi Arabia has also announced plans for massive AI infrastructure investments as part of its Vision 2030 transformation strategy. (arabnews.com)
These nations increasingly view AI infrastructure as:
- economic diversification
- geopolitical influence
- future industrial capacity
That reinforces the broader trend. AI is no longer being treated like a software sector alone.
It is increasingly being treated like strategic national infrastructure.
Energy May Become the Ultimate Bottleneck
One of the biggest changes in AI is that energy is suddenly central to the discussion.
That was not true during previous software eras.
Now:
- data centres require enormous electrical capacity
- hyperscalers are exploring nuclear partnerships
- cooling infrastructure is becoming critical
- AI campuses are reshaping regional power planning
Some SpaceX discussions even referenced:
- orbital compute infrastructure
- solar-powered AI systems
- globally distributed AI infrastructure connected through Starlink
Whether those systems become commercially viable or not, the direction is revealing.
The industry is beginning to think about AI infrastructure the same way it thinks about:
- power grids
- transportation systems
- national telecom infrastructure
Sovereign Compute Is No Longer a Fringe Idea
This is why countries like Canada are increasingly discussing:
- sovereign AI
- domestic GPU capacity
- local inference infrastructure
- regional compute ecosystems
If AI becomes foundational economic infrastructure, then dependence on foreign compute providers becomes a strategic concern.
The SpaceX disclosures reinforce that reality:
AI leadership increasingly depends on physical infrastructure control.
That includes:
- energy
- chips
- networking
- land
- cooling
- manufacturing
- connectivity
The Next AI Winners May Look Very Different
Many startups still assume the winners in AI will be:
- the best UX
- the best prompts
- the smartest agents
- the most polished workflows
Those will matter.
But the emerging infrastructure race suggests the long-term winners may instead be organizations that control:
- compute
- power
- distribution
- networking
- semiconductor supply
- capital access
That is why the SpaceX IPO disclosures matter so much.
They reveal that the frontier AI race is no longer behaving like software.
It is beginning to behave like industrial infrastructure competition.
And that may reshape the global technology industry itself.
Frequently Asked Questions
What is AI capex and why does it matter?
AI capex is the capital expenditure on the physical infrastructure that AI runs on, data centres, GPUs, power, cooling, networking, and fibre. It matters because frontier AI is no longer cheap to operate. The four U.S. hyperscalers are projected to spend roughly $725 billion in 2026 alone, and that level of capital is what now decides who can build and run frontier AI at all.
Why are hyperscalers spending so much on AI infrastructure?
Because demand has outrun supply on multiple physical inputs at once. GPUs are scarce, electrical capacity is constrained, cooling needs are enormous, and frontier model training and serving require industrial-scale facilities. Building or controlling that physical capacity is now the bottleneck, not writing better software.
What does sovereign AI mean for a country like Canada?
It means meaningful domestic capacity to train and run AI on national soil, on hardware the country controls, under its own data residency rules. Canada has committed roughly CAD $2 billion across the Sovereign AI Compute Strategy, the AI Sovereign Compute Infrastructure Program, and the AI Compute Access Fund. The intent is real and the policy direction is set, but the gap to hyperscaler annual spending is still very large.
How is China approaching AI as infrastructure?
As industrial policy. Stanford’s 2026 AI Index estimates Chinese government guidance funds have deployed around $184 billion across chips, manufacturing, data centres, and domestic AI ecosystems. Domestic chipmakers like SMIC are scaling capacity, and China has reportedly set targets to integrate AI into 90% of its economy within five years, treating it as full-stack national infrastructure rather than a software market.
Can Canada and similar countries actually compete on AI infrastructure?
Not on hyperscaler spend alone. The realistic path is targeted advantage: clean energy availability, public AI supercomputing for research and startups, domestic inference for residency-sensitive workloads, and clear data-sovereignty policy. The open question is whether national strategies can keep pace with hyperscalers and Middle East sovereign funds that are now spending at industrial-buildout scale.
