The artificial-intelligence (AI) revolution is underpinned by a powerful compute infrastructure that will come to define the 21st century. Compute capacity is a key driver of progress in the modern world; it is powering scientific knowledge and enabling incredible productivity gains. It is a foundation upon which leaders can reimagine the concept of the state by using AI and data to deliver better, more efficient and lower-cost services.
Building on previous interpretations by the OECD[_] and the AI Now Institute,[_] TBI defines compute as the ability to store, process and transfer data at scale. It refers to a stack of hardware such as graphics-processing units (GPUs) and central processing units, as well as software that enables the use of specialised chips and supporting infrastructure in data centres (including servers and cooling equipment). It incorporates classical and quantum computing in both the public and private domain. It is also important to note that whereas the OECD definition refers to AI compute specifically, our conception is broader.
The increasing use of technology for computational modelling (using computer science to solve scientific problems), coupled with declining unit costs, mean that compute is the modern agent of change. For example, without large-scale compute, Moorfields Eye Hospital would not have been able to produce foundation models that predict the likelihood of patients developing eye diseases[_] or cancer, or having a heart attack. And that’s not all: driverless-car companies would not be able to train their models effectively[_] and weather-forecasting technologies would be nowhere near as accurate.[_] In many sectors, compute represents the foundation of progress.
Throughout history, technological revolutions have never been driven purely by the technological breakthroughs that define them: the underlying ecosystem and enabling infrastructure have been just as crucial. For example, it was the railway system rather than the steam train that ushered in the first Industrial Revolution.
Building functional compute capacity is complex. It is a system of systems; it is a technical architecture that includes not only hardware but also networks, plus the skills and capabilities to operate everything. It also needs to be accompanied by regulatory and legislative frameworks, to allow countries to take full advantage of the infrastructure available to them.
The countries that appreciate the urgency are already forging ahead, investing heavily in the components, hardware, software, research and skills that are central to the appetite for compute power. But user demand is proving to be insatiable, leaving room for further growth (even among those countries taking the initiative). Meanwhile, those countries that are cut off from access to this infrastructure will be disadvantaged not only in terms of day-to-day productivity, but also in their potential to innovate and grow.
To gain a more comprehensive understanding of the global state of access to compute power, the Tony Blair Institute (TBI) commissioned research group Omdia to carry out an audit of 55 countries across 25 key indicators that contribute to a compute ecosystem, including countries’ access to the skills base to operate them all. This report presents our initial analysis and findings, and plots the data in an interactive data explorer.
Our work builds on existing attempts to take stock of compute capacity (Tortoise Media[_] and Oxford Insights[_] have created AI indexes that draw on data from the TOP500 supercomputer list[_]) by tracking A100 clusters and AI chip usage in research at an aggregate global level. The OECD’s AI Policy Observatory[_] is beginning to record several AI compute indicators for its partner countries, but there is a gap: our report and data explorer provide a wider picture of the state of access to compute, as well as offering targeted policy recommendations for leaders around the world. (An annex of the full methodology is available as a downloadable PDF.)
There is growing evidence that access to compute is forming the basis of a new global digital divide. To address this, countries that haven’t already need to start investing in the necessary infrastructure – but no government should feel like it is acting in isolation. TBI’s data can help countries identify how they can best leverage their resources with new partnerships and initiatives that will deliver the most impact for investment and engagement.
We have concluded from our analysis that there are several key actions that countries can take to enable more equitable access to compute and prevent the digital divide from widening further.
Countries with emerging compute ecosystems
Complete a comprehensive review of their national compute capacity to complement their AI and innovation strategy and goals.
Create a consortium or regional framework to share compute resources, modelled on the EU’s Joint Supercomputing Initiative.
Improve compute security and resilience by stabilising the grid, investing in sustainable data centres and developing comprehensive cyber-resilience legislation.
Develop a talent pool able to operate and scale a 21st-century compute ecosystem.
Advanced computing nations
Build public compute infrastructure that spans the entire country and improves domestic access for startups and the public sector.
Introduce new measures for compute governance, such as incentivising responsible practices in exchange for access to public compute infrastructure.
Establish multilateral agreements to develop shared quantum-computing infrastructure and programming best practice.
International development institutions
Curate a Compute Development Fund. This would be earmarked for investment in countries with developing compute capabilities and include a GPU reserve fund to act as collateral for future investment.
Launch a Digital Development Corps for compute and innovation assistance. The United Kingdom can play a key role in this endeavour, creating a model of compute diplomacy that would include a Compute Knowledge Exchange Bank. This would help other countries understand different applications of compute and best practices in building compute infrastructure.
What Determines Access to Compute?
Building a robust and fully functioning compute ecosystem relies on investment in multiple tiers of infrastructure. Each tier, if soundly constructed, enables the next, so getting the sequencing right is key. Broad investment in high technology will not translate to the cloud capability required for a roundly developed ecosystem – but direct investment in data centres might. As such, a bespoke approach to next-tier investment is crucial when striving to meet compute-capacity goals.
Uneven distribution of (and investment in) core underpinnings of the compute ecosystem will leave some regions and countries at a disadvantage, despite their hardware investment. For example, a compute ecosystem cannot exist without the people to conceptualise, operate and maintain it. That’s why the measurement framework used to assess the state of access to compute in 2023 captures the key drivers of a country’s access. In February 2023, the OECD set out a high-level framework for compute capacity,[_] specifically focused on national AI compute. Our project expands on this and other similar frameworks, visualising a broader definition that incorporates both AI and non-AI compute-specific indicators within a wider ecosystem. It encompasses 25 indicators sitting under seven categories, which together represent a holistic view of a compute ecosystem (see figure below and methodology annex at end of paper).
Click on the categories below to reveal more information
A Growing Industry – But Is It Growing Apart?
Every year more than $300 billion is invested globally in the bits and bytes required to grow compute capacity. However, these investments are not evenly distributed: the disparity in access to compute infrastructure within and between regions is becoming increasingly clear.
The global market for the supercomputers that are used to solve the most complex scientific problems was valued at around $12 billion in early 2023 and is expected to see annual growth of 9 per cent annually over the next five years. The global semiconductor market is currently valued at $1.4 trillion,[_] and the cloud computing market alone is expected to reach $1.25 trillion by 2028.[_] This is growth that shows no signs of slowing, with companies racing to assert their compute supremacy. The pace is picking up, with NVIDIA’s latest supercomputer, Israel-1, having been unveiled two months ahead of schedule due to enormous demand.[_]
Yet states generally remain sluggish. Our research found that in 2023, Meta bought 30 times more H100s (the leading AI chip) than the British government procured for the AI Research Resource, a new national facility.
Private companies are investing heavily in compute resources
That said, 60 per cent of the TOP500 list of supercomputers is made up of machines located in the United States, China and Germany.[_] There are, for example, no supercomputers in East Africa. And there are large language model (LLM) researchers who, struggling to access compute resources in their own country, find themselves having to send data abroad, to then be governed by the terms and conditions of competing tech companies. [_]
Through our analysis we discovered yet more evidence of this unequal distribution:
A small cluster of countries is forging ahead in terms of developing compute capacity, while many others are starting from scratch. The US’s advantage on data-centre builds vastly exceeds similar investment in any other country, although China, Singapore, the Netherlands and a handful of others have also built significant capacity. Still, the overwhelming majority have built fewer than 20 leading data centres in their country.
Many countries need to invest more in compute capacity
The Balance of Power
Countries making limited investments in hardware in search of compute capacity, only to be hindered by the absence of a robust baseline infrastructure, is an increasing problem. Low power-grid maturity – defined by a large number of power outages in a country on an annual basis – is inversely proportional to most drivers of compute capacity.
Directly linked to power-grid maturity is internet accessibility and the number of internet users in a country. The latter two factors are facilitated by access to subsea cables and internet exchange points, leaving vast disparities between regions. The lack of access to subsea cable landings in sub-Saharan Africa, for example, is a clear indicator of a baseline-infrastructure shortcoming that needs to be remedied as part of efforts to prevent a compute divide.
Another issue related to infrastructure is that servers running compute workloads need constant power to function; a data centre that experiences an outage will be immobilised in its entirety. These shutdowns are costly for data-centre operators because it means that their clients’ critical workloads cannot be accessed, causing possible security concerns and potential business losses.
Developing the Right Skills
North America dominates in its access to leading compute training programmes, which are provided by commercial leaders such as Amazon Web Services (AWS), Hewlett Packard Enterprise, Linux, Microsoft and Redhat. Europe, the Middle East and North Africa lag behind significantly in this respect, with less than half the number of programmes that are available in the US (not to mention Asia and Oceania).
Global availability of compute training programmes varies hugely
Divides that could lead to missed opportunities have already emerged. In Europe, the availability of compute training programmes varies enormously: at one end of the scale you have the UK, which boasts the readiest access to such resources, while at the other end Latvia finds itself significantly disadvantaged. And while countries such as Norway, Finland and Estonia are at the forefront of some elements of digitalisation, they too are struggling to access the training programmes that would help them to grow their compute ecosystem.
Training programmes across Europe also lack parity
Gaps in the Clouds
Beyond AWS and Microsoft being two of the largest providers of training programmes, they also have the resources for machine learning and other parallel computation tasks (cloud services optimised for machine learning). Indeed, AWS, Microsoft Azure and Google Cloud collectively make up 65 per cent of the global cloud market and have significant influence over many of the factors that shape digital ecosystems.
Divides in access to this infrastructure are stark. While 23 per cent of the countries we researched have in-country access to parallel computing through AWS, Google and Microsoft, 47 per cent do not have any local access in place at all. These hyper-scaling tech companies are also heavily investing in every level of the compute stack and communications ecosystem.
Private-sector investment in the layers of the compute ecosystem
As such, bridging the digital divide and developing a sustainable compute ecosystem will depend on how well leaders are able to navigate their relationships with private-sector companies, which will be key partners when it comes to enabling growth. Essentially, those countries that have the strongest relationships with large technology providers will be best placed to provide effective public and private computing infrastructure.
The Global Compute Picture
Each country is on its own compute journey and while there may be clear leaders, the global picture viewed as a whole is much more nuanced. Some countries are clustered around their strengths, characteristics and potential trajectories, with each cluster demonstrating that building access to compute does not have to be an absolute endeavour: there is value in building competitive advantage in one part of the compute ecosystem and sharing it. Here are some examples of the countries making up each category.
Global Leaders: China and the US
Both countries have unmatched raw compute capacity. The US has 29.3 million servers installed, with continued server revenue growth at 4.6 per cent in 2023, high levels of mobile-computing access, significant local supercomputer capacity and access to parallel computing. China has 15.8 million servers installed, with revenue growth at 8.1 per cent in 2023. Its mobile-computing access is slightly less than half that of the US, as measured by growth in mobile-phone acquisition. China matches the US on local supercomputer capacity and access to parallel computing in terms of investment, with the latter a good proxy for future performance. The US has higher data-centre investment and R&D incentives but China is the frontrunner when it comes to the percentage of GDP spent on investment in high technology, at 27 per cent versus 11 per cent.
Established Powers: Canada, France, Germany, Japan and the UK
These countries have a strong track record of pre-existing investment in traditional computing and supercomputing capacity, which is complemented by reliable infrastructure and good access that in turn ensures high societal participation. They all invest significantly in R&D, which will help ensure their long-term leadership. For example, the UK is the second-highest investor in quantum computing in the world, second only to the US and followed by Canada, France and Germany (a trio that have relatively high R&D incentives). Germany and Japan look set to lead the group as they have long committed to high R&D expenditure that has only increased over the past few years, while that of Canada, France and the UK has flatlined.
Accelerators: Ireland, Luxembourg, the Netherlands, Singapore and Switzerland
There are many countries that are punching above their weight with higher compute capacity per capita than those mentioned above. These countries tend to have a mature infrastructure and ready availability of cloud and colocation data centres, as well as plans to invest in more. These countries have the highest number of servers installed per capita of all 55 countries we have focused on, while Ireland, the Netherlands and Singapore are global leaders in cloud computing.
These are countries that have a high number of software engineers per capita and are regarded as innovators (Switzerland, for example, is currently leading the Global Innovation Index). Ireland and Singapore, meanwhile, should be considered silicon islands, having made themselves attractive destinations for international tech firms such as Amazon and Microsoft. Google too: Ireland’s low corporate tax rates helped to make it first choice to host the firm’s European headquarters, while Singapore is its largest Asian hub.
Specialists: Australia, Finland, Italy and South Korea
These countries serve as specialised computing hubs with good supercomputing and parallel computing availability. Australia and South Korea are the only two countries globally where each of the major international cloud-service providers offer services that are optimised for AI. Finland, Italy and South Korea, in particular, have world-leading supercomputing capacity, with Finland having the highest number of supercomputer FLOPs (floating-point operations per second, indicating performance capability) per capita of any country.
Rising Giants: Brazil, India, Indonesia, Mexico, Nigeria, Saudi Arabia and South Africa
These countries have several characteristics in common: very high rates of server growth; moderate supercomputing capabilities; reasonable levels of quantum computing investment; and an expectation that more cloud and colocation data centres will be built in the coming years. There is also a common theme: the state has played a central role in the progress of each of these countries.
For example, Brazil’s largest supercomputers are developed by state-owned oil and gas company Petrobras;[_] similarly, Saudi Arabia’s King Abdullah University of Science and Technology built supercomputer Shaheen III.[_] The state is an important vehicle for any country wishing to leverage compute, especially those that do not have traditionally strong relationships with large technology giants.
Nascent: Greece, Poland, Rwanda and Thailand
These countries are in the very early stages of compute development. However, this is slowly changing, with increasing investment in the building of data centres (using both local companies and international cloud and colocation service providers). Greece and Poland, for example, have begun to invest in quantum computing in the same way as more advanced countries. However, the countries in this bracket still don’t have enough human capital and will need to improve access to compute training, as well as boost their infrastructure capabilities, if they are to realise their potential.
Emerging: Ethiopia, Ghana, Hungary and Romania
These countries, at an even earlier stage of their compute development, possess relatively high growth rates in infrastructure (albeit starting from a lower base). Critically, countries that are "waking up" have very unstable power grids, which is a barrier to scaling any supercomputing ambitions.
The countries included in our report
Regional Snapshot: Current Status
Asia and Oceania
The percentage of GDP that the Asia and Oceania region spends on advanced technology is more than twice that of any other region. While China is the leading powerhouse in the area, countries such as Indonesia benefit from numerous subsea cable landings and internet exchange points, helping to increase server growth.
Strength: High R&D spend on advanced technology.
Weakness: Regional inconsistency in server installations and availability of cloud services.
Europe’s reputation for strong governance of digital issues is a keystone of its compute ecosystem, but it is also upping its investment in quantum computing. Meanwhile, the prevalence of open-data practices across most European countries, particularly Denmark, Finland and Poland, is cementing the continent’s leadership potential. In addition, the health of the innovation ecosystem is reflected in the number of compute patents coming out of Austria, Germany, Sweden and Switzerland, all sitting just behind Japan.
Strength: Access to quantum computing.
Weakness: Disparities in prevalence of training programmes.
Countries in Latin America have been proactively setting up the governance to underpin their AI systems, deploying the technology across justice administration, procurement and fraud detection. To develop the necessary compute capacity there has been collaboration with the EU on the RISC2 project, which facilitates the sharing of best practices between the two continents and, most importantly, speeds up capacity-building and high-performance compute research. This shared infrastructure will be a significant asset in supporting emerging digital economies on both sides.
Strength: The rise of Brazil’s compute ecosystem.
Weakness: Low investment in servers.
Middle East and North Africa
Investment in core capacity has been key to the development of the region’s compute ecosystem. The UAE has emerged as a regional leader, placing science and tech at the forefront of its political mission; it already has cloud-service deals in place with AWS and Microsoft, and recently signed a deal with Cerebras to build nine additional supercomputers. Meanwhile the Abu Dhabi-based G42 recently announced the open-source release of Arabic LLM Jais, helping to ensure that the non-English speaking world is not left behind. The UAE is also working to close its talent gap and incentivising sustainable energy projects.
Strength: Local supercomputer capacity.
Weakness: Poor investment in quantum computing.
The US government has invested billions of dollars in compute infrastructure in recent decades. In 2023, the government earmarked a historic $210 billion for federal R&D, including more than $100 billion for the National Science Foundation. This is bolstered by the private sector and universities investing in data centres and high-performance computing systems. Major cloud providers such as AWS, Google Cloud Platform and Microsoft Azure have built more than 100 large data centres across the country.
In contrast, Canada’s government has not publicly disclosed any dedicated vehicles for AI infrastructure investment, and has so far shown little inclination to finance high technology, data centres, R&D incentives and sustainability incentives.
Strength: Access to machine-learning talent in the US.
Weakness: Apart from Italy, Canada trails all other G7 nations on compute investment.
Server investment growth in sub-Saharan Africa is about 13 per cent higher than the global average, and the likes of Nigeria, Rwanda and Senegal have been attracting significant local and foreign investment in data centres. The region is also developing a base of skilled technical talent, implying a readiness to enhance capacity. There are other building blocks in place too: for example, the region’s rate of contribution to the open-source community is much higher than that of Asia, Oceania, North America and Europe. Moreover, the continent has a higher rate of STEM graduates than Europe, Oceania and South America, plus a greater availability of training programmes than Europe, the Middle East, North Africa and South America.
However, sub-Saharan Africa continues to face obstacles in its efforts to fully develop regional compute ecosystems due to a lack of cloud and colocation data centres. Cloud computing in the region has a penetration rate of 15 per cent (compared with 71 per cent in Europe), predominantly run by foreign service providers with data centres overseas.
Strength: Favourable tax and tariff environment.
Weakness: Lack of stable power.
How countries go about accessing compute should be completely reimagined. Instead of fighting for hardware or investing in partial systems that never quite deliver, they should combine their competitive strengths as part of a multi-state effort. Where values and legislation align, countries can pool their resources to create regional compute capacity.
The European High Performance Computing Joint Undertaking[_] (EuroHPC JU) is a prime example: it aims to support the development of industry and digital transformation while protecting Europe’s technological sovereignty. This type of model allows for more equitable and sustainable access to compute across the continent, as well as putting mechanisms in place to minimise AI safety risks.
Regions that pool their resources to build capacity together will be best placed to pull ahead and compete globally. Progress will vary depending on a country’s political aims and geopolitical relationships, but it requires vision, infrastructure and coordination.
Vision Whether countries want to participate in an effort to cultivate regional capacity or develop sovereign capacity, leaders need to identify their best areas of contribution and advantage to build their compute vision and strategy.
Infrastructure Our 25 indicators represent a set of potential levers for the growth of the 21st-century state. They can inform domestic policy but also set off alarm bells for the international policy community as to new inequalities that might be emerging in the global tech ecosystem.
Coordination By measuring the core drivers of compute power it’s possible to identify partnerships that can be the basis of a new collaborative model of overseas digital development.
Partnering for Capacity Growth
Bridging the access-to-compute divide will depend on how well leaders are able to navigate relationships with private-sector companies. Large cloud-compute providers are key stakeholders and partners for enabling growth: AWS, Google Cloud and Microsoft Azure together make up 65 per cent of the global cloud market and have significant influence over many of the factors that shape digital ecosystems.
These partnerships are becoming even more integral because tech companies have diversified (and thereby consolidated) their positions within the communications infrastructure. Google, for example, is a major funder of subsea cables, such as the recently announced Nuvem cable connecting the US, Bermuda and Portugal.[_] Most top providers also pay for dedicated bandwidth on subsea cables and Tier 1 networks. Additionally, many internet exchange points tend to be in close proximity to the large data centres of major tech companies.
Regional Snapshot: Spotting Potential
Countries display strengths in different elements of the compute ecosystem, pointing to opportunities for the co-development of infrastructure, skills and regulatory capacity. By visualising regions as a whole, key areas for regional investment and opportunities for shared approaches emerge.
Asia and Oceania
India is emerging as a critical player in the region. It has all the conditions to lead in compute on the world stage: some of the best access to parallel computing; massive investment in servers and quantum computing; and a highly developed ecosystem of software engineers and training programmes. That said, Singapore has 100 times as many servers per capita as India and the largest concentration of data centres in the world.
Based on our data, Indonesia is the highest-ranking country on number of subsea cable landings and internet exchange points, and the seventh highest for server-investment growth. With Indonesia and India having significant populations, the rapidly rising demand for computing resources stands to make the region a hub of innovation; the political direction of both will be crucial for the global compute ecosystem in the coming years.
Meanwhile, with the latest round of US export controls sweeping away the ability for companies such as NVIDIA to export their AI chips to China, there is a risk that Asia’s compute ecosystems will diverge from the West. It could result in countries looking to develop their own domestic chip capacity, which may not be well integrated with Western software, resulting in a fragmented ecosystem.
Opportunity: Computing innovation coming out of India and Indonesia.
Threat: Export controls inhibiting market growth.
France, Germany, the Netherlands and the UK invest heavily in quantum companies, which could be further boosted by greater engagement with major digital players such as Amazon, Google and IBM.
Europe as a whole should maximise its infrastructure and hardware investments, increase the availability of compute training programmes and ensure consistent power access in eastern Europe. The continent should also prioritise the training of more software engineers.
Opportunity: Quantum leadership potential.
Threat: Lagging investment in data centres.
To take local and regional compute capacity to the next level, Latin America needs to increase investment in servers (it is particularly low in some key countries) as well as upgrading baseline infrastructure in anticipation of this acceleration. This will provide the foundation for financing high technology, to then deliver the value needed on the compute journey.
Brazil is leading Latin America in terms of compute capacity thanks to its influence in software development. It is a leading player in the open-source community, ranking third globally after Switzerland and Belgium in terms of GitHub contributions, and boasts a significant software-developer ecosystem, with approximately 200,000 such professionals. With robust access to parallel computing and supercomputing, the country is seizing opportunities to rapidly develop its ecosystem.
Opportunity: Potential for collaboration with Europe.
Threat: Political instability.
Middle East and North Africa
While many countries in the region have invested in leading-edge AI capacity, further investment in other parts of the compute ecosystem is still required. Saudi Arabia would strengthen its position with further public and private financing of its national quantum-computing ecosystem; Morocco would progress more quickly if it matched its investment in hardware with baseline infrastructure such as data centres and co-location centres. Essentially, this region is a good example of why failure to invest in all parts of the compute ecosystem can limit countries wanting to scale their science and technology ambitions.
Opportunity: Improving commercial training.
Threat: Infrastructure instability.
While Canada’s compute capability is among the best in the world, it jeopardises reaping the full economic and social dividends of its financial commitment to AI – especially given its lead in AI policy – without further funding for compute infrastructure. Without this investment it risks losing its strong talent pool to the US, where demand for software talent outstrips supply and robust R&D incentives spur a dynamic AI start-up ecosystem.
Opportunity: Engagement with private sector.
Threat: Shifts in semiconductor supply chain could impact US leadership.
The region should empower its growing developer talent and mobile-first markets by building research capacity into distributed-edge computing (where tasks take place entirely on devices such as phones and laptops), thereby reducing reliance on high-powered data centres. A good example of this is Google Ghana’s research programme into deploying efficient AI models via mobile devices.
This is a region primed for investment in compute infrastructure in the next decade. Although it scores lowest within our data explorer on critical AI infrastructure resources (such as supercomputers and access to parallel computing), its youth demographic is driving an intense demand for digital services. In Nigeria in particular, a bourgeoning pool of talent alongside a favourable tax and tariff environment suggest a growing appetite for compute supply.
Countries in sub-Saharan Africa have the opportunity to develop direct strategic partnerships and investments with leading technology companies, to not only build the necessary infrastructure but also develop next-generation sustainable compute technologies. Based on TBI’s research, these countries are less fossil-fuel dependent than many others around the world, meaning they could stand at the forefront of championing innovative, green data-centre infrastructure and pioneer new approaches to clean computing.
Opportunity: Access to renewable energy.
Threat: Talent migration from underdeveloped IT ecosystem.
We’ve taken a closer look at the steps that are required across the compute spectrum to both maintain progress and begin to establish a more level playing field. We’ve also considered the part that governance has to play in these proceedings.
Emerging Compute Ecosystems
The demands of digital development and providing baseline infrastructure are still key challenges in many countries – and the means of securing access to compute has not yet been calculated as part of their economic plans. However, as this report has established, many of the steps that will increase digital connectivity and the provision of digitally enabled public services are reliant upon increased compute capacity. Targeted investment in the key drivers of compute will not only deliver short-term benefits but also smooth the path to increased access. This requires a three-tiered approach, depending on the compute-ecosystem maturity of a particular country.
Bolster baseline critical infrastructure: This means shoring up internet connectivity by leveraging public-private partnerships for subsidised access, as well as reducing power outages and increasing grid supply and cyber resilience. Supporting R&D through science, technology and climate funding, as well as funding energy audits across data centres to improve efficiency, will provide further initiatives for long-term infrastructure resilience. This is an ongoing endeavour that requires consistent assessment of growing infrastructure needs as demand increases.
Build institutional capacity: A strong ecosystem that can leverage the power of AI requires a fresh approach; this starts with assigning a government agency to conduct an assessment of, or synthesise data on, a country’s compute needs. Then it’s a case of providing the roadmap for targeted investment in infrastructure that balances funding of short-term compute assets (cloud computing) and longer-term resources (quantum computing and sustainable compute). Leaders will need to ensure greater coordination and alignment between government agencies working on compute-related activities, as well as initiating regional and international collaboration. A good example of this is the expanded Memorandum of Understanding on Cyber and Emerging Cyber Technology Cooperation, signed by Indonesia and Australia in 2021.[_]
Increase opportunities for funding and resource sharing: For nascent compute ecosystems, collaboration is key to unlocking resources. Governments should create consortiums or regional frameworks for the sharing of compute, modelled on the EuroHPC JU; they also need to actively demonstrate their compute demand through investor events to attract local and foreign investment. This should be coupled with an enabling regulatory environment of tax incentives and subsidies to drive meaningful investment in key areas of growth, such as boosting local data-centre capacity[_] and improving access to cloud services. And to ensure the constant flow of investment there should be a function on the level of the Intergovernmental Panel on Climate Change (IPCC) to monitor and improve regional and global compute access. This recognised body would then be able to unlock funding to facilitate a country’s key targets.
Advanced Compute Ecosystems
Although many countries have made great strides to build out their compute capacity, not all researchers and businesses have access. Establishing an equitable approach is essential for maintaining a thriving innovation ecosystem. Meanwhile, most countries that are in a strong position when it comes to compute are nevertheless encountering challenges in maintaining the talent levels required to expand and benefit from large-scale compute resources. As such, these countries need to work towards a vision of a sustainable compute ecosystem capable of long-term growth, through three key steps.
Generate public AI infrastructure: Governments should partner with AI companies to set up national compute research clusters. The size of these clusters will vary depending on the economic situation in any given country, as well as broader compute availability. In our recent paper A New National Purpose: AI Promises a World-Leading Future of Britain we called for the UK to build a cluster of 30,000 GPUs. This would need to be underpinned by training and the recruitment of technical talent, to maintain clusters and support the fine tuning of models. It would also be wise to facilitate continued innovation through preferential access for domestic venture-capital firms, which can use the resource as part of their term deals with domestic startups.
Create institutional mechanisms for compute governance: There are risks to rapid expansion of compute without the appropriate oversight and governance. Increased computational power can be misused: it could result in malicious cyber activity, the creation of biological agents, disinformation at scale and misalignment risks (whereby systems do not do what users and developers intended). Governments need to create the legal and regulatory mechanisms to ensure the responsible use of public AI compute infrastructure.
Support access to the post-silicon ecosystem: Governments need to be flexible. Advances in areas such as quantum and neuromorphic computing (an emerging form of hyper-efficient computing modelled on the brain) could upend entire digital ecosystems, be it the result of improved material modelling, a reduction in the amount of energy needed for computing or some other leap forward. Governments need to stay at the forefront of growth by bringing together supercomputing and quantum-computing teams under shared bodies, away from their current siloes, as well as embedding quantum computing within the more traditional high-performance computing (HPC) exascales. Also required are new models to ensure cloud access for academics and to develop bilateral agreements with other emerging quantum giants. A EuroHPC JU-style model for quantum could be the future of the industry.
International Development Institutions
Global tech-governance conversations are currently dominated by the challenge of regulating AI. As the technology outpaces governance, policymakers are more focused on the outputs of front-running countries than democratising access to the infrastructure. Further, countries without the infrastructure find themselves without a seat at the table, thereby unable to shape the long-term future of AI and compute governance. As a result, conversations around international development must consider an expanded vision of access to compute as a global public good. A new model of 21st-century digital-development initiatives is required to empower all countries to reap the benefits equally. In this regard there are three key steps that can be taken by governments and the international community.
Curate a compute development fund: As technology companies consolidate throughout the communications stack they are crucial partners in supporting the advancement of foundational digital infrastructure. As such, international development can no longer be driven by government alone. A joint public-private philanthropic fund would provide subsidised access to AI servers and act as a multiplier for countries’ individual investment in data centres. It could also create a system of GPU debt funds that would catalyse further investment in broader compute infrastructure; a GPU reserve fund to act as collateral for future funding would also be beneficial.
Build a Digital Development Corps: This would be a Foreign, Commonwealth & Development Office-supported initiative facilitating students, recent graduates and retirees (as well as tech workers on sabbatical) to volunteer in emerging digital economies as part of a Global Compute Capacity Initiative. As a digital version of VSO or Habitat for Humanity, this would build compute capacity, skills and knowledge in emerging digital economies and create vital links between research and policy communities globally.
Create a new model of compute diplomacy: The UK, in particular, can play a key role in correcting the imbalance of power in global digital development. From the outset, this should focus on three key areas:
- Establish a compute-knowledge exchange bank: The UK is the only country that conducts a comprehensive review of its compute capacity. Like the Education Endowment Fund’s Teaching and Learning Toolkit[_] and the OECD’s AI Policy Observatory,[_] it should provide a platform to help other countries understand the different applications of compute and best practices when building infrastructure.
- Ensure that the proposed IPCC model for AI monitors accessibility: TBI has previously recognised this challenge in the internet-policy arena, proposing a Multi-Stakeholder Panel on Internet Policy modelled on the IPCC in our paper The Open Internet on the Brink: Recommendations for a Future Model. The recommended body would be charged with producing data on the state of the internet divide, evaluating policy proposals and improving the representation of the global internet community. This should be expanded to include key indicators of compute capacity, such as creating a global register for leading-edge chips. It would also map global access to compute, with the ability to predict regions where digital divides look set to widen and act accordingly.
- Support other nations in creating their own AI Taskforce-level of state capacity: Very few governments have made substantial efforts to develop institutional structures to govern AI capacity. The UK has developed state capacity to understand, test and deploy safe high-computational systems, and so should advise other countries on doing the same. This would help governments learn how to recruit top talent, develop evaluation suites and integrate safety testing into policy and delivery.
We are entering an era that will be defined by the rapid progress of AI and, as a result, the rise of new applications beyond the realms of imagination. Being part of harnessing and building that progress is predicated on a thriving and functional compute ecosystem. As governments drive towards their vision of the 21st-century reimagined state, in which tech sits at the heart of supercharging the provision of public services and stability, they should also ensure that they zoom out to accommodate the need for access to compute as part of their journey.
With countries taking decisions to invest in key parts of their digital architecture, understanding the impact that this investment can have on their role within a compute ecosystem will be key to ensuring access. While investing in a supercomputer may provide short-term reputational benefit, taking steps to stabilise power infrastructure instead will provide greater returns in the long term. This report and accompanying data explorer represent the first step in providing leaders with a tangible understanding of the potential pay-offs presented by different levels of investment, as well as guiding the sequencing of a future-proofed digital architecture.
New models of collaboration, both regionally and internationally, offer countries the opportunity to be key actors within larger, more cutting-edge compute ecosystems. Competitive advantage can be leveraged to contribute to multi-state compute clusters, resulting in a new model of compute diplomacy where countries share their strengths in exchange for access to compute. Key to facilitating this is a long-overdue overhaul of international donor initiatives, making innovative structures for sharing best practice (and future-proofing digital infrastructure development) central to the aim of preventing a new digital divide. The train has left the station – it’s time to get on board.
We would like to extend our thanks to a steering committee of experts who offered their advice and guidance in the development of this report and data explorer:
Dr Andrea Calderaro, University of Cardiff
Bev Crair, Oracle
Vladimir Galabov, Omdia
Lennart Heim, Centre for the Governance of AI
Elaine Hutton, Omdia
Alison Kennedy, Science and Technology Facilities Council
Pascal Korte, Omdia
Aaron Lewis, Omdia
Adu Odeli, Omdia
Dr David Snelling, Fujitsu
Nigel Toon, Graphcore
Dr Claudio Torres, Universidad Técnica Federico Santa María
Thanks also go to the following TBI staff for providing additional research:
An annex of the full methodology behind this report and data explorer is available as a downloadable PDF.