Sustainable Data Centres, Part IV

Author: Ian Bitterlin, Portman Partners AssociateIn

This fourth and final article of the series, we move our discussion to the impact that Utility power has on data centres and, increasingly often, the impact that data centres have on national grids and their need to de-carbonise. Data centres are currently buying up all the renewable power they can get and, in some cases, investing in production. But is it another form of greenwashing?

As I write this, there is a growing list of locations that are either calling for a moratorium on new data centre deployments or considering the future ‘out loud.’ These include Amsterdam, Frankfurt, Dublin, Sydney, and Singapore.

Ireland and Singapore – record-breaking nations?

There are two interesting examples of national grids facing immediate data centre expansion problems: Ireland and Singapore

With its rapid expansion to reach the top four data centre hubs in Europe, Ireland has an operator base investing in Irish wind power to achieve their ‘green’ aspirations.
With its limited solar and wind energy opportunities, Singapore is both space-constrained and reliant on gas.

Both countries have limited capacity interconnects to import (or export) energy across their borders. Moreover, both Ireland and Singapore have immediate neighbours who will find that increasing their own renewable capacity (for electric transport and domestic heating) and de-carbonising their expanding grids will make exporting to another country far less attractive than meeting their own carbon ambitions.

Ireland

Ireland, to be precise, Dublin, has joined London, Frankfurt, Amsterdam, and Paris as a key European hub for good reasons, and a league table might show that Dublin will soon be competing for the second spot. However, compared to the population size/grid capacity, Eire has already become the densest data centre hub in EMEA.

Dublin is an attractive place to live for employees of the big tech companies, with a well-educated, English-speaking workforce, flexible labour laws, and a relatively high unemployment rate making high-quality skilled jobs easy to fill. Having said that, direct data centre jobs are low in number compared to the business size, but many support companies have sprung up and flourished to add to the jobs created, which must amount to several thousand by now. The Irish business tax regime has been very generous to data centres (although I am not sure that this is the prime mover of the data centre immigration), as is the construction planning system and, apparently, the national power company, although there are rumblings of concern in EirGrid. The fibre connections between the USA, Dublin, and Europe (traditionally via the UK and Rotterdam) are now very strong. It is not a ‘low cost’ location (both CapEx and OpEx), but I do not believe that low costs are a key driver for the type of hyperscale operators attracted to Dublin.

Electrical energy is dominated by fossil fuel, but intermittent onshore wind generation has grown dramatically. Plans to spend €1Billion on a subsea cable link direct to Brittany (the Celtic Interconnector) are in place, despite strong public objections in Cork. However, that interconnector will probably be used more to import French nuclear energy than to export Irish wind and stabilise the Irish wind in the process.

All things considered, while it is unlikely that the data centre expansion will slow down if the underlying criteria do not change, it looks like it might. Pressure is building on Irish power generation that only the consumers can fund, and there is an initiative (2021) for many countries to harmonise their tax laws on ICT ‘cloud’ operators.

Whilst researching for a Channel 4 ‘Dispatches’ documentary on video streaming and the effect of data centres on the Irish grid, I looked at the 10-minute Irish grid statistics. On 5/11/20 at 3:10 pm, Eire was consuming 3.42GW. The wind wasn’t blowing, so the grid was high in carbon at 512g CO2/kWh with the following fuel mix:

Coal: 18% (620MW of 1.2GW capacity) at 820g CO2/kWh
Wind: 1.2% (38MW of 4.2GW capacity) at 11g CO2/kWh
Hydro: 5% (173MW of 216MW capacity) at 24g CO2/kWh
Gas: 49% (1.67GW of 4.27GW capacity) at 490g CO2/kWh
Oil: 12% (394MW of 1.27GW capacity) at 650g CO2/kWh
Other: 2.6%
Importing 15% from the UK (530MW of 500MW interconnect capacity) at 313g CO2/kWh
Exporting 1.8% to Northern Ireland at 512g CO2/kWh

The grams of CO2/kWh for each energy source are the internationally agreed figures (IPCC 2014).

You may notice that the wind capacity already installed is 4.2GW, more than the entire grid load. If it were more, how would Ireland use the excess? Maybe hydrogen production.

I take this opportunity to point out that no fuel has ‘zero’ carbon at the point of customer connection, despite many claims to the contrary – including EDF currently advertising on UK TV a Nuclear/Wind/Solar mix as ‘zero carbon’ instead of around 12g. Yes, the EDF mix is very low in carbon emissions and is where we should be heading for in our entire grid, but it is not zero. The carbon footprint comes from the construction of the generation plant, the distribution system, and maintenance to get the energy to the customer – concrete, steel, iron, copper and carbon fibre, etc.

We can also note that, in theory, if the wind was blowing strongly and evenly (above 80% of maximum turbine rating) and the load was low (a warm summer day), then the Irish grid could run on wind alone. The same theoretical scenario applies to the UK, but in both cases, the slightest ramp in load creates a risk of the system tripping off-line. Experience from Denmark and the UK (events of 9th August 2019) points to 60-70%% as a ‘safe’ limit for intermittent wind if backed up by continuous baseload generation.

The climates and (lack of) heavy industry profiles of Ireland and the UK are similar enough to make comparisons possible – with Ireland’s GDP at 60% service-based (US$77,000 per capita) and the 28th largest economy compared to the UK at 83% service-based (US$43,000 per capita) and the 6th largest economy, pre-pandemic. Ireland has a population of 5m, so the 3.42GW was equivalent to 0.68kW per person, including the Dublin data centre estate. If we deduct 500MW (a current estimate, based to an extent on Greenpeace data) for data centre load, the ‘per person’ figure drops to 0.58kW – which we can compare to the UK below.

At the same time as the Irish grid number above, the UK was consuming 38.3GW at 313g CO2/kWh (nuclear helping, no coal or oil, low wind, and mainly dependent upon gas) with a population of 66.6 million – equivalent to 0.58kW per person. If we deduct 3-4% for UK data centres (1GW?), that drops to 0.55kW per person. So, it looks like the Irish grid supports a much higher proportion of data centres than the UK. That is in the order of 1GW in the UK compared to 0.5GW in Ireland. We can further compare the city of Dublin to the city of Bristol. Dublin has a population of 1.2 million out of the national 5 million, so Dublin’s ‘share’ of the national 3.42GW was 800MW. If we deduct 500MW for data centres (the most likely actual load of 600MW capacity given the type of data centre in Ireland), we are left with approximately 300MW for the non-data centre load and the general population. Bristol has 90% of the population of Dublin, and neither city has much manufacturing or heavy industry, but Bristol has a small data centre estate and consumes 230MW on average with a winter peak of 300MW.

Clearly, with the climate being nearly identical, it would appear from this that 500MW of data centre actual load (from an installed base of 600MW) for Ireland looks in the right area and represents around 15% of the grid demand – a huge proportion in comparison to the UK’s best-guess 3-4%. Is it already the national record holder? Schemes with existing planning permission for further Dublin data centre developments, including 200MW by AWS, could easily push the data centre share from 15% to 30% in the next five years. All highly loaded, not serving the Irish populations’ ICT needs, and all wanting and investing in renewable energy, will create more problems for the grid.

Simply put, what happens when the wind doesn’t blow a lot or blows too strongly? The net carbon emissions of the energy consumed by the Irish population could remain at their pre-data centre level whilst the data centres use all the renewable energy that they are, no doubt, investing in and paying for. The Dublin data centres may be encouraging wind investment, BUT they are also consuming all of it and more, and I do wonder who will pay for the grid reinforcement. Over the same period, the Irish grid must expand for the electrification of transport and domestic heating, probably by 5GW. All of it, existing and expansion, must be low carbon, if not all renewable, so it is possible that the data centre driven investment in the grid will be a partial enabler of the national decarbonisation plans. Still, the data centres will likely need their emergency backup generators more than they do today.

Some low-carbon source must provide the baseload safety case and substitute for the wind when it is low in strength.

Singapore

Singapore, an island state with a non-seasonal equatorial climate and a population of 5.7m, has become a major ICT hub in the entire SE Asian region. The grid demand on 4/3/21 at 10:45 am was 6.9GW (at 492g CO2/kWh) which is 1.2kW per person, and inspection of the data shows that this load is nearly constant over the year. Given the hot and humid climate, it is impossible to compare this to Ireland, despite the similar population. Still, there is no doubt a 24/7 high air-conditioning load explains some of the 100% increase over the UK and Ireland.

Based on a recent survey of the region (CBRE, see below), the estimated data centre estate is around 600MW, a grid share of 9%. However, new developments are regularly announced, including the ’50MW’ Facebook facility that will be highly loaded. And, that proportion, even if it is accurate, will rise as the government encourages inward business investment, which will, in turn, attract more data facilities. By Q2-2022, the CBRE report suggests that Singapore’s data centre power capacity, when fully loaded, could total 1GW, nearly 15% of the grid. Similar attributes to Dublin (of connectivity, well-educated workforce, English as a first language, and an attractive location to live) make Singapore a destination of choice in SE Asia and the wider AsiaPac.

The only ‘problem’ they have (or perceive they have) is that they are >97% dependent on natural gas for electrical generation at 490g CO2/kWh. Land is expensive and in limited supply, so solar PV is constrained to a peak of 2.7%, and onshore wind suffers from similar restrictions. Offshore wind, highly preferable to onshore, is heavily constrained by some of the busiest shipping lanes in the southern hemisphere. At the time I looked, 0.66% was coming from solar PV at 45g/kWh, and 0.55% was imported from Peninsular (Malaysia) at 681g/kWh. I am unsure of the capacity of the only interconnect (with Peninsular), but it is usually only delivering low single-digit percentages – and it is always high in carbon at 680g CO2/kWh, being 60% coal, 35% gas, and 5% hydro, so not of any help in reducing Singapore’s carbon footprint. Singapore currently appears to be self-sufficient in generation capacity (using gas), and it is a matter of national security that this remains so.

But is this a problem for Singapore data centres? I do not think so. Rather it is a problem that needs fixing in the grid. I would suggest mini-nuclear from Rolls Royce (but remember what I said about opinions) because the other attractive features of the location as a data hub currently outweigh the ‘problem’ of no renewables. However, many facilities will, undoubtedly, want to advertise themselves as renewably powered. So, in the short term, Singapore facilities such as Facebook are already indulging in Virtual Power Purchase Agreements to reach their green agenda and maybe a version of sustainability, and we will review VPPAs below.

How do market commentators estimate the size of a locations’ data centre estate?

Reports quoting, with apparent precision, the total MW capacity of data centres in a variety of countries or cities regularly appear in the trade press. For just one example, CBRE reported 411.9MW for Singapore carrier-neutral colocation in Q1 2020, up 14% on the previous year, with an additional 311MW upcoming by 2022. The data comes from local CBRE knowledge of clients, managed deals, and sources such as planning applications, etc. However, the apparent accuracy is probably questionable since other facilities, in particular enterprise and hyperscale, do not fall into the description of ‘Tier 1 & 2 carrier-neutral colocation.’ Therefore, the total data centre capacity in Q1-2021 could easily be more than 600MW, rising to 1GW in 2022.

What is really needed is clarity on what these types of reports mean by ‘MW,’ and a few options for ’50MW’ can be explored:

The planning application is unlikely to include the net ICT design load as a data centre size/rating descriptor. So, we could assume that ’50MW’ is the capacity of the power supply rather than the ICT load.
The power company always delivers the grid capacity in MVA, not MW, and at a power factor of 0.8. So ’50MW’ might be a 50MVA utility feed that can only deliver 40MW. Given the ambitions of marketing material and power factor not being easy to understand, it is possible that MVA is confused with MW.
The relationship between grid capacity and the net ICT load is not the PUE1-3 but the peak PUE on the hottest day of the year. This is what the original Green Grid called PUE0, which was not annualised kWh but peak kW divided by rated ICT kW, sadly not included in ISO/IEC 30134.
And finally, there is always a difference between rated full load and the actual load – so-called partial load. This can have a minor impact for some facilities by the end of the first 12 months, such as Google or Facebook, which will run at 85-90%. Some hyperscale facilities, such as AWS, may reach the 75-85% level, but it will take longer to reach. At the other end of the scale, retail colocation may start with an anchor tenant at 15% and take four years to reach 55-60% and not go much higher. Colocation facilities running at below 40% load are too numerous to count yet are highly profitable businesses.

We can see that our reported ’50MW’ facility might represent an ICT load of as little as 3MW and as much as 17MW if it were described by grid connection OR have a 50MW rated ICT load fed by a 112MVA grid supply transformer (90MW at 0.8PF) with an unknown actual starting load and unknown growth plan.

There is also the case where some facilities, whose business does not involve HD images/video and that refresh their ICT hardware every 24 months, witness a shrinking power demand of their data centre, despite a growing ICT usage. An example in Singapore is a facility built in 2003 to house an 8MW IT load (about a 20MVA supply) that was loaded to 2MW by 2017. As they refreshed the ICT hardware on a 36-month cycle, its load was shrinking by 5-10% each year. If it appeared in a report, what would its entry say? 8MW, 20MVA, 2MW, or 5MVA? Perhaps the reader can see why I said that no one really knows, as information given for the situation to be clear is insufficient.

Hyperscale is probably the easiest to predict. The PUE is usually low because they quickly fill up and need to be heavily loaded to maximise investment return. They also refresh hardware to slow down the rate of new facility construction to meet demand.

Renewable power

We have, finally, reached the third step towards true sustainability, which our industry has decided is the first, and that is to power our optimised data centre with renewable energy, or, at the very least, low carbon.

The sources of renewable power in the grid are also intermittent and variable or predictably time-phased; wind, solar-PV, solar-thermal, sustainable wood, geothermal, hydro, tidal, wave, etc. These can be used to create generation fuels such as green-hydrogen, bio-diesel, and ethanol. Other sources are low carbon and will play their part in combatting climate change, but not strictly renewable, such as municipal waste combustion. My favourite candidate for delaying climate change is nuclear fission.

All sources, renewable or not, have a carbon content associated with the plant construction, installation, distribution system construction, maintenance, decommissioning, and recycling. For example, solar PV has a carbon content of 45gCO2/kWh, 400% that of wind, and 5% that of coal. It will probably take more than 100 years to get most of the carbon content out of the power system after all the construction, operation, and material production processes (metals, carbon fibre, ceramics, concrete, and electronics) are renewably powered, and the old plant is replaced. Therefore, we should be careful with terms like ‘zero carbon’ in our thoughts and language. Zero is a good target, but ‘as low as is possible’ is more accurate, albeit less sensational. After all, we are a carbon-based lifeform living on a carbon-based planet that depends on nuclear energy (the sun) for our very existence, and where would trees be without CO2?

These sources are available (depending upon location and resources) to be delivered to the data canter via the utility connection, either as part of the national fuel mix or in a specific blend via a Power Purchase Agreement (PPA) which we will consider below. It is possible and often done to specially purchase non-renewable energy, such as nuclear, but this is rare for data centres so far. In some locations, the grid is already (through geography) 100% renewable, e.g., Norway (hydro) and Iceland (geothermal), and specific contracts have no use or meaning, but they are small grids.

It is probably worth pointing out that power in the grid does not flow from one specific source to one specific consumer. For example, a London facility may contract and pay for hydro-power from Scotland, where most UK hydro is located. Still, it will consume power that is a blend of all the sources connected to the grid at any given moment in time. This is especially important to understand when contracting to purchase enough intermittent renewable energy to power your facility over one full seasonal year. For example, with wind power, there will be times when there won’t be enough wind blowing to turn a blade and even more times when there’s more wind blowing than you can consume. You will pay for your annualised consumption of 100% wind, obtain the energy source certificate, and claim to be 100% renewably powered – yet consume energy from the grid that doesn’t come from anywhere in particular and is composed of the national fuel mix. It is a paper exercise that needs transparency.

If the reader has not thought about the concept before, the easiest way to visualise the flow of energy in a national grid is to imagine an inflated balloon with many valves, like a musical bagpipe. The generation sources (numbered in hundreds, highly redundant, and concurrently maintainable) are all connected to the balloon and managed in harmony to keep the pressure up to a precise value. The pressure is the system voltage. The consumers, many thousands, are connected to the balloon and draw off the pressure to drive their load, at which point the balloon pressure drops slightly, and the generators increase their feed to maintain the system voltage. The balance is a matter of constant control to ensure that generators push in and consumers take out, with no ‘reverse flow’ happening. In your data centre facility, you are usually connected at Medium Voltage (such as 11kV or 22kV in the UK) with a tolerance that meets the regulatory, operational licence. The voltage in the grid is also alternating (AC), also at a precisely controlled frequency (50Hz or 60Hz by country). If the system frequency rises or falls too much, the safety systems disconnect the consumers for plant protection and human safety. However, at the low voltage level in the facility, we get 8V/Hz (400V/50Hz or 480/60Hz by country) in a sinusoidal waveform generated by the rotating machines in the grid. So, importantly, we need to understand that the electrons in the cables do not flow in one direction but oscillate forward/backwards at the (50Hz or 60Hz) system frequency. We can imagine the power transfer from the grid to the facility as a continuous push-pull action, with the push and pull coming from the entire grid at one end of a broom handle and you at the other, absorbing the energy. The energy does not come from any individual generation source, and renewable energy electrons are not coloured green.

We can rule out most electrical energy sources as suitable for being sited in/beside the data centre site due to lack of space, such as on-site wind, where the land could be used to leverage more data centre space. Another reason could be due to low energy density and intermittence, such as roof-mounted solar-PV generating no more than 1-2% of the site demand, but all can be imported from the grid.

If on-site wind is contemplated, there is also a limitation on the turbine capacity, where it cannot exceed the facility load plus the facility supply rating, or the turbine rating can never be reached, and the annual energy supply of the turbine will not be maximised. For example, if you wanted to build a 100% wind-powered data centre with its own turbines on site, then your grid connection would have to be 3x your data centre load (not ICT load), and you would need a very large energy storage system, plus emergency generation.

The primary energy source out in the wider grid that data centres look toward to lower their carbon footprint is wind, which we will examine below, but solar PV is another option. Solar is predictable, unlike wind, and reassuringly quotidian, but takes a lot of land space (550kWh/m2/year), and it needs replacement every 12-15 years. Solar produces energy at 45gCO2/kWh, compared to 11g for wind.

Wind power in the grid can be located on or offshore, and the output depends upon the wind speed profile and turbulence, especially ground topology and obstructions onshore. There are offshore locations where it is too windy too often, and the turbine must be stopped for safety. But, assuming that the location is ‘windy’ in general:

Wind turbine-rated power is stated at its maximum operational wind speed, just before it shuts down. Turbines of the type used in national power grids are now rated in the 5-10MW range, with 10MW requiring rotor blades of 200m diameter and a nacelle height of 150m.
The chosen size is influenced by the location, not only the wind speed profile but also visual amenity. And, for example, the larger the turbine, the higher the wind speed required to start accelerating the blades with some turbines using grid power to get the blades up to stable power generation speed.
Even in usually windy locations there are often weather conditions that produce no/light wind that prohibit any generation for several days at a time, e.g., doldrums. Fortunately, in temperate climates that don’t have a high air-conditioning load, these periods most occur in the summer when grid demand is at its lowest.
Onshore turbines, if sited correctly, can be relied upon to generate the equivalent of 35-40% of the peak rating, averaged over a full seasonal year. So, a 4MW turbine might produce 13GWh annually from an annualised average of 1.5MW, minimum 0MW, and maximum 4MW.
Offshore turbines can be relied upon to generate the equivalent of 45-50% of the peak rating, averaged over a full seasonal year. So, a 4MW turbine might produce 15GWh annually from an average of 1.7MW, minimum 0MW, and maximum 4MW. However, the increased production of an offshore location is always offset by being more expensive to install, connect to the grid, and maintain.

It is important to note that as power output rises almost linearly with wind speed, the annualised output of wind turbines can drop significantly if the peak energy can’t be consumed when it is available. This becomes an increasing problem as the proportion of wind in the grid increases beyond a certain point if the grid is to avoid stability problems with the risk of loss of supply. Denmark found it to be around 60%, and the UK had related issues in 2019. The solution is to have sufficient baseload generation to cope with variable wind output and/or fast variations in the connected load.

Purchasing renewable energy

The last step towards sustainability is to power the optimised load from sustainable sources. As the national fuel mix of all countries has increasing levels of renewable generation, the consumer gradually reduces their carbon footprint.

For data centres ISO/IEC 30134 (‘Metrics for resource-effective data centres’) includes Carbon Usage Effectiveness, CUE, which describes the annualised total carbon in the energy systems per kWh of ICT load, i.e., gCO2/kWh. The total carbon is the annual attribution/consumption in the electrical utility, the carbon content of renewable energy generated on-site, the embodied carbon in potable water used for adiabatic or evaporative cooling, and the carbon released by diesel generator operations. For example, a UK data centre that connects to the grid without any special purchase agreements can rightly claim a Renewable Energy Factor (REF) of 0.22 because, for a full year, the UK’s renewable energy content is 22% of the grid. But, when the wind doesn’t blow, we will be burning gas.

Another example for Carbon Usage Effectiveness (CUE), is France where CUE = 0.16 – because 85% of the French grid is nuclear, making Paris a much ‘greener’ place to run a data centre if ‘carbon’ is your poison. However, the larger data centres, especially hyperscale, are moving forward in their effort to use renewable energy and are using renewables in their publicity and CSRs to highlight their ‘sustainability.’ Some of these hyperscale facilities also go much further in terms of improving the process (high utilisation and low PUE). Still, none try to reduce the actual ICT load. Their business is to sell digital services, and if the customer wants to buy them in increasing quantity, the seller is unlikely to try to sell them less.

Before we proceed, one question arises that I would like the reader to consider: Why should data centres take more than their share of the renewable sources in the grid? Why not prioritise hospitals or care homes, or schools, or government tax offices, etc.? With a significant proportion of the global data traffic being pornography and YouTube (60%), why should that be renewably powered? I don’t know.

The large data centre and ICT businesses directly invest in renewable generation projects or become a major, or even the sole, consumer of new wind/solar farms. This encourages the projects to be built, and many commentators make this point into a reinforcement of the idea that such data centre developments are playing a pivotal role in the decarbonisation of the grids in which they sit.

There is no doubt that when the national/regional data centre load is <5% of the total grid capacity and the renewable content in the national fuel mix averages out at <25%, the grid can safely cope with the variability factors in both sources and load. But as the data centre load increases out of proportion as it has in the hubs of Virginia, Dublin, and Singapore, to >20-30% of grid capacity, another dynamic comes into play, one that becomes a negative consequence of data centres. That factor is that new data centres are responsible for the new renewable sources to be built and then consume all of it (annualised), leaving all the remaining consumers (existing and new) with the original grid profile. The grid also has a few issues to overcome:

Technically, it must deliver energy to those large data centres 24/7, regardless of the intermittence and variability of the renewable sources. For Europe, data centre loads do drop slightly at 3-4 am GMT but not by more than 20%, and the peak is always at 8-9 pm GMT even if the wind isn’t blowing.
To capture the peak output of the renewables, it must find another load (or energy storage, but that is unlikely at grid-scale) or export it to another country. This can require high investment, but it is not clear who pays for it and ignores the fact that other countries will be trying to raise their own renewables at the same time. The interconnects between countries will have to be several times the capacity of the existing typical 2GW (one large power station).
Suppose the proportion of renewables in the grid rises too far without the grid having a low-carbon non-intermittent capacity to fall back upon. In that case, the usually expected high-reliability supply (>99.99%) becomes an issue. That fall-back position could end up needing to be nuclear, and who will foot the bill for that? It is unlikely to be the data centre. This raises issues of data centre emergency power generation being needed far more often, and for longer, than the present average of 0.15% of the year, or 12 hours, & using which fuel?
Commercially, when the grid is deregulated, it appears the considerable opportunity a hyperscale data centre puts onto the market (e.g. 200MW, see the example at the end of this paper) makes it very attractive for investors in the renewable project.

Usually, it results in a low energy price and a long contract. It would be preferable that sources such as wind attract a premium kWh price (rather than any government subsidy) so that investors in new resources built them for all the consumers.

To illustrate the point, consider the situation of a data centre hub with a peak grid capacity of 5GW plus 2GW of heavily loaded hyperscale data centres that have all contracted for 100% wind energy. The grid had a 5GW of fossil-fuelled power capacity, and it has been expanded by 6GW of wind that generates between zero and 18GW depending upon wind speed. In the summer, the normal load drops to 3.5GW, but the data centres continue to draw 1GW, so the load is 4.5GW. When the wind doesn’t blow, the data centres are fed by fossil fuel or at least 1GW of renewable energy must be imported by an interconnector rated at 1GW to another grid with renewable power to spare (over and above meeting its targets and supplying its own customers). When the wind blows strongly (but not higher than the turbine safety limit), the grid should have a use for 13.5GW or will have to install more turbine capacity to meet the contracted data centre clients 100% renewable demand. In any case, the grid must be expanded to cope with 18GW of distribution, must keep its 5GW of fossil-fuelled generation to reliably feed its 5GW of normal load (including domestic). The ‘net’ grid has not de-carbonised, the national emissions are the same, yet the total new grid looks like it has made tremendous progress. Does anyone see the similarity in this scenario that EirGrid might be contemplating?

Power Purchase Agreements (PPAs)

At some time in the future (and ‘now’ in places like Norway & Iceland), there will be no advantage or need in trying to be sustainable by purchasing renewable power. That ‘future’ is a 100% low-carbon utility (that can power the traditional loads plus domestic heating and 100% electric transport) that must be brought forward in time. It is a huge task we will struggle to reach – enlarging the grid capacity by 2-2.5x and making it all low-carbon (never ‘zero’) before 2050? In the meantime, all consumers can put commercial pressure on the grid by opting to pay for low carbon. The price may be an issue, but fossil-fuelled energy must be raised in price (a carbon tax) to make low-carbon (including all renewables and nuclear) look attractive, and we have a mechanism to do that now – a PPA, for all types of consumers, including domestic & data centres.

Inside one grid system, it is simple to regulate by not selling more renewables or low carbon than each grid produces or directly importing (with its renewable energy certificate) from a neighbour via an interconnector. Unfortunately, in a democracy, change in the electrical utility is not a fast-moving process. Regulatory planning for new renewables and low carbon has its objectors, including visual or amenity loss, and NIMBY, extremely so for nuclear power. Wind and solar also use land area, often on the coast or in shallow water near the coast, and will require many more EHV power distribution pylons, so expansion of low carbon capacity will not appear overnight.

Virtual Power Purchase Agreements (VPPAs)

The most complex PPA to understand is ‘Virtual’, the so-called VPPA. It’s perfectly legal and appears to be commercially acceptable in our deregulated markets, as a legitimate way for a data centre in one grid (country, or regional grid within a country, that may, or may not, have an interconnector of sufficient capacity) to pay for renewable energy in another grid, gain the renewable energy certificate, and claim that a proportion (or 100%) of the data centre is renewably powered. The concept of VPPAs is hardly discussed in public, and the public probably takes the sustainability statements at face value.

Some of the largest data centre operators aggregate their global demand and sign PPAs and VPPAs worldwide to cover 100% of their consumption with renewable energy certificates.

VPPAs offer the appeal that businesses can have data centres built wherever they want them to be, e.g., Dublin for tax, Singapore for a regional hub, Virginia for bulk availability of connectivity, and London for the financial markets. At the same time, they can claim the benefit of renewable power from somewhere where it is abundant and low cost. The source sells the certificate to the data centre, and the actual energy is pushed into the source’s local grid for any local consumer to buy. The local power provider to the data centre is paid a capacity charge for the delivery and what is sometimes called a ‘sleeving through’ fee. As they pay the remote supplier for the energy, they can also apply a separate mark-up on each kWh.

With a VPPA in place, the data centre continues to draw energy from the local grid at the fuel mix of that grid but offsets its carbon consumption, albeit in another grid. It should be noted that the principle of carbon offsetting has long been used in the air travel industry by promising to plant trees etc.

An example was illustrated in a Channel 4 (UK) Dispatches TV program making two statements about the data centre hub in Loudoun County, Virginia (USA), one of the largest concentrations in the world. First, the sole power provider, Dominion, currently with no renewable energy in their output and based on burning coal, stated that they aim to be ‘carbon and methane free’ by 2050. However, one of their biggest clients, AWS, said that they would be 100% renewably powered by 2030 – some 20 years earlier. Second, AWS, which fulfils all the requirements of the second step of sustainability, will not move their numerous facilities (several hundred MW) out of Loudoun County. Instead, they sign PPAs as Dominion slowly de-carbonises and VPPAs in other grids, probably outside of the USA, to reach their 2030 target.

In general, for VPPAs, because of the large-scale and low variability of the annual energy purchase load, the contract price (with all fees) is usually lower than that in the general market, but, in some instances and locations, not as low as the bulk supply of coal-fired power. An example of 440GWh/year (50MW continually) of coal-fired energy was reported a few years ago from Prineville, Oregon, USA, of $USD 3.5c/kWh, for a 25-year contract, which was about 25% of the cost of commercial power in the UK at the time. The energy is still supplied from the contracted source and will be for another 15 years. Still, the data centre is now ‘powered by 100% renewable energy from Texas – a separate and distant grid.

General examples of problems of large scale PPAs

The national power system operators, owners and investors are also finding problems of pricing electrical energy with high levels of intermittence, and it may be more economical to discard power than attempt to invest in energy storage at grid-scale. Below is an example from Ireland:

Cornwall Insight, Nov 4th, 2020

“We see cracks in the wholesale market approach to pricing electricity. Traditional approaches are resulting in more frequent negative LMP results, which are simply unsustainable. We need to find a better way to value electricity.”

This message came through loud and clear during a meeting that day discussing the future of Ireland’s Electricity Markets hosted by Cornwall Insight and within the article linked below. There was some talk of possibly fixing reference prices, but there is no clear answer on fixing the existing wholesale market pricing problems. The problems with Wholesale Capacity and Energy markets are symptoms of the same problem; how to value electricity, both capacity and energy, in a world dominated by an abundance of electricity coming from increasing amounts of renewable resources. Those resources have no inherent variable fuel charges that compete with other generators that rely on fuel to generate electricity. This is a big issue to net-zero it’s getting bigger, as the Cornwall Insight white paper ‘The net zero paradox: Challenges of designing markets to bring forward low marginal cost resources'” indicated.

This news item from the Netherlands published on the internet by Datacenter Forum, updated 25-02-2021, highlighted another example of the grid’s problem: “In five to seven years’ time, five enormous halls must be built on the outskirts of the municipality, packed with computer servers from an as of yet unnamed American tech giant. The municipality promises that the complex will generate 410 jobs, which will largely benefit the local population.

When the plans were presented, the municipality did not want to disclose who owns the enormous data complex. Although the municipality acknowledges that 410 jobs are limited for an area of 166 hectares, Zeewolde is nevertheless pleased with the arrival of the unknown tech company. “Data centres like this help to create the jobs of the future in a growing industry,” said De Jonge.

And the data centre industry is growing. Data centres are sprouting up in the Netherlands. There are now 189 scattered throughout the country. It makes the Netherlands a European ‘mainport’ for so-called cloud computing, and it takes up space relatively limited with comparable main ports such as Schiphol and the port of Rotterdam. However, there are not so many complexes the size of Zeewolde. Microsoft has one in Middenmeer and another under construction, and Google has such a giant centre in Eemshaven.

The arrival of these ‘computer farms’ leads to employment but also to grumbling among the local population who, in addition to the arrival of data centres, see windmills, whose green electricity would disappear directly into the computer parks, rising in the area too. In the Netherlands, more than 2 percent of all electricity goes to data centres, which indeed primarily run on green electricity. The complex in Zeewolde will also become a major consumer with an estimated capacity of 200 megawatts, enough to power 200,000 households simultaneously.

Adapting the Power Grid

There is already a shortage on the power grid in the region, but grid operator Tennet does not see the newcomer as a problem because due to the expected development time of five to seven years, there is sufficient time for connection to the high-voltage grid and the construction of a switching station, said a spokesman.

The re-use of Waste Heat

The municipality of Zeewolde has big plans to make the computer complex “one of the most efficient in the world”. For example, a steering group, including neighbouring Harderwijk, is investigating the possibilities of using residual heat – cooling water of about thirty degrees Celsius – for heating homes. However, in concrete terms, nothing has been agreed upon for an inter-municipal heat network with the international tech giant yet. Still, according to Alderman for Spatial Planning, Wim van der Es, that is not yet necessary: “The residual heat project does not have to be ready tomorrow.”

So, I come to the end of this series of articles and hope that you enjoyed them – but certainly not to the end of the discussion. We face interesting ethical times in the data centre industry.

About the Author:

Ian Bitterlin is a Portman Partners Associate and a Chartered Engineer with more than 32 years’ experience in data centre power and cooling following 21 years in rotating electrical machines and systems. Formerly CTO for Emerson Network Power Systems in EMEA, and Visiting Professor to the University of Leeds, School of Mechanical Engineering, Ian is now principal consultant at Critical Facilities Consulting. Having filled senior posts in major UPS and data centre OEMs — Anton Piller, Liebert, Emerson Network Power, Active Power & Chloride — Ian has been instrumental in key product innovations and continues to consult on new data centre design and M&E products.

Ian has been awarded ‘Outstanding Contribution to the Data Centre Industry’ twice, by Data Centres Europe in 2009 and by DataCenterDynamics EMEA in 2015.