March 19th was a good day for Inflection – but perhaps a better one for Microsoft. That day it was revealed that Redmond had agreed to pay £495m to license and host the enterprise AI startup’s powerful large language model (LLM), a deal that would not only repay Inflection’s investors but also, the startup claimed, help to “put cutting-edge AI capabilities in the hands of thousands of developers.”
Meanwhile, 72 members of Inflection itself – including its two co-founders, Mustafa Suleyman and Karén Simonyan – would find themselves newly employed in Microsoft’s new AI division. A statement from Redmond’s chief executive Satya Nadella was positively euphoric. “There is no franchise value in our industry,” said Nadella. “Let us use this opportunity to build world-class AI products, like Copilot, that are loved by end-users!”
Despite the strangely Marxist phrasing, the underlying message was aggressively capitalist: a big tech whale had effectively swallowed the resources and talent of a smaller, nimbler contemporary. Such partnerships are viewed with intense suspicion by the UK’s Competition and Markets Authority (CMA). In a speech delivered to competition experts in Washington back in April, its chief executive Sarah Cardwell said that the watchdog had identified an “interconnected web” of 90 such deals involving the same six firms: Google, Apple, Microsoft, Meta, Amazon and Nvidia.
“We believe it is important to act now to ensure that a small number of firms with unprecedented market power don’t end up in a position to control not just how the most powerful models are designed and built,” said Cardwell, “but also how they are embedded and used across all parts of our economy and our lives.”
Probes by the CMA were subsequently launched into deals between Microsoft and the startups Inflection, Mistral and Open AI and two separate arrangements struck by Google and Amazon with Anthropic. And yet, despite its chief executive’s thunderous invective against the power of big tech over AI, the CMA repeatedly came up short in its merger investigations. In five out of six of its probes, the watchdog found that the deal under scrutiny did not, in fact, ‘qualify for investigation under the merger provisions of the Enterprise Act 2002.’
That included the Inflection deal. The CMA’s executive director for mergers, Joel Bamford, put a positive spin on the regulator’s decision to abandon its inquiry. “In this case we were able to complete our review at pace, clarifying the jurisdictional reach of UK merger control but confirming that these particular arrangements pose no competitive threat,” Bamford wrote on LinkedIn (the CMA chose not to respond on the record to Tech Monitor‘s requests for comment.) “This is, in my view, the UK merger regime working exactly as intended by parliament.”
This begs an important question – if a deal like that reached between Microsoft and Inflection does not meet the CMA’s criteria of an alliance that violates the principles of free competition in tech, then what kind of partnership would? Does the UK’s competition watchdog even have the teeth or expertise needed to guarantee its mission?
Simon Heath doesn’t think so. “The CMA is not really fit for purpose in the fast-moving nascent technologies space that we have today,” says the former KPMG M&A veteran-turned-COO of Heligan Group. “Historically, it was there to review market size, market share, revenue and number of products a company has. The challenge with these technology companies is [that] they’re pan-global and operate above nation-state jurisdictions – which makes it harder for single countries to put restrictions in place.”
Threats to competition
There’s no common rationale behind the CMA’s decisions in these five merger probes. In the case of its investigation into the $2bn partnership inked between Alphabet and Anthropic – wherein the latter also gained prodigious compute from the former – the regulator concluded, after reviewing reams of internal documentation, that it was unlikely the Google creator had any material influence over the startup through its board or stake. The fact that Anthropic’s annual turnover didn’t exceed £70m also discouraged the watchdog from proceeding further with its inquiries (the same test also put paid to its inquiry into the startup’s relationship with Amazon.)
Microsoft’s $650m deal with Inflection, meanwhile, was deemed a relevant merger situation by the CMA. Despite the deal involving the very public absorption of the startup’s senior leadership team into Redmond’s ranks, however, the watchdog ultimately determined that the deal did not “give rise to a realistic prospect of a substantial lessening of competition,” mainly because Inflection’s ‘Pi’ chatbot proved a puny competitor to OpenAI’s unassailable ChatGPT.
In that case, is the CMA’s playbook for these investigations fit for purpose? According to Heath, many of the well-worn criteria the regulator uses to define anti-competitive practices in other markets do not translate well to the world of AI. For one thing, the investment structures of businesses in the sector make it devilishly hard to prove whether or not a big tech firm has ‘material influence’ over the decisions made by the likes of OpenAI or Anthropic.
“Regulators would typically look at who has the controlling share in a company,” says Heath. “But with these deals, we have companies putting billions into relatively early-stage businesses but not taking a majority equity share as would be expected.”
For example, Microsoft invested $11bn in OpenAI but has only a minority shareholding in the latter (49%). Similarly, in the Microsoft-Mistral AI investigation, where Microsoft has a minority shareholding position of just 1% and no board representation, the CMA determined Microsoft did not possess enough influence over Mistral to warrant further scrutiny.
It’s also challenging, says Heath, to even quantify the size of the market for chatbot-derived services, let alone determine the slice of that overall pie occupied by a given company. “Defining the market size is very challenging because it’s nascent,” he argues. “What part of AI are you looking at? Is it from a consumer perspective? Is it where it’s connected to machines or manufacturing processes?”
But what about the extraordinary computing power promised by big tech firms to startups in these deals – surely those are self-evidently anti-competitive? Not necessarily, says Tom Sheridan, a vice president with RTP Global specialising in early-stage AI investments. “Very few companies in the world can fuel foundational models” alone, he explains. The reality is that most AI startups will find it impossible to scale without access to cloud computing, and they usually do — with many paying for the privilege, like almost every other company in the market. Others, like the firms singled out in the CMA’s recent merger probes, are promised discounted rates or free access in exchange for non-voting equity, a commitment to long-term collaboration and access to their models on certain platforms, such as Anthropic on Amazon Bedrock.
Outwardly that might seem anti-competitive, but these kinds of arrangements have ultimately failed to meet the CMA’s own threshold for such behaviour. In the case of Microsoft’s alliance with Mistral, for example, the watchdog somewhat embarrassingly discovered that the French startup’s commitment to housing its data on Redmond’s servers was non-exclusive, with its foundation models also available on Amazon Bedrock, Snowflake and Perplexity Pro. The fact that Anthropic was being investigated for deep ties to two different big tech firms, too, suggests that the startup was and remains a more substantive partner in these deals than the CMA first suspected.
“The CMA needs to consider whether either of these investments could cause monopolistic characteristics,” says Heath, “but ultimately, and by definition, having two investors will mitigate CMA concerns.”
What’s next for regulation in this space?
The CMA has yet to publish a decision on whether it will mount a full investigation into the close ties between Microsoft and OpenAI. It, too, could be dropped. Not only does Microsoft maintain a minority stake in OpenAI, but the software giant – seemingly aware that it was courting antitrust scrutiny with its multi-billion dollar investments in the startup – surrendered its board seat in July of this year.
The CMA’s attempt to cut through its infamous ‘interconnected web,’ then, appears to have faltered – for now, at least. According to Nick Owers, the coming months may see the regulator gifted powers it’s long craved to curb big tech’s spidery influence in AI. “There’s much to look out for that will set the tone on the regulation of foundation models in coming months,” says Owers, a technology partner at Keystone Law – not least, he adds, an “update from the CMA on its approach to such models.”
The CMA will also acquire new powers from the Digital Markets, Competition and Consumers Bill, set to come into force next year. Described as the most consequential consumer protection law in decades, the DMCC Act will not only make merger reporting obligations mandatory for firms designated as having ‘strategic market status’, but also allow the CMA to work with overseas regulators to share and acquire information about the activities of companies it’s investigating.
“What’s clear is that the CMA now views the DMCC Act as its most flexible tool for delivering on its overall duty to promote competition for the benefit of consumers,” says Owers. “The CMA was recently explicit that the Act may provide for a firm’s conduct in digital markets to be investigated in a more timely and holistic manner than a traditional Competition Act investigation.”
Heath points out that less than 1% of 913 CMA-referred transactions from 2023 to 2024 had prohibitions or cancellations. But he does think the DMCC will provide the greater powers the CMA needs.
“The CMA often works in a partial vacuum with imperfect information,” says Heath. Increased powers “to improve its data gathering capabilities will enable it to have greater depth of understanding in the future – which, ultimately, must be good for consumer choice and fairness in the market.”
Read more: The case for a ransomware payment ban
