24AUG

Fractile lands NATO and CIA chip cash

4 min read

16:40UTC

Oxford inference-chip startup Fractile closed a $220m Series B at roughly $1bn on 20 May with NATO Innovation Fund and In-Q-Tel on the cap table; no UK sovereign vehicle joined the round.

← Back to UKRI's sponsor department no longer exists Jump to analysis ↓

TechnologyDeveloping

Key takeaway

Allied national-security capital filled the UK chip slot the SAIU and BBB direct mandate left empty.

Oxford inference-chip startup Fractile closed a $220m Series B at roughly $1bn valuation on 20 May, led by Accel, Factorial Funds and Founders Fund ¹. The cap table is the news: NATO Innovation Fund (NATO-IF), the Alliance's €1bn multi-sovereign venture Arm, joined alongside In-Q-Tel (IQT), the CIA's non-profit strategic investor, with Oxford Science Enterprises as the university-linked tag-along. Anthropic, the AI safety company, is in early talks to buy custom inference chips from Fractile. Founded in 2022 by Walter Goodwin, Fractile develops SRAM in-memory compute silicon designed to run large language models without the memory bottleneck of conventional GPUs.

No UK sovereign vehicle participated. The Sovereign AI Unit (SAIU), DSIT's £500m AI infrastructure equity vehicle launched in April , is absent despite naming compute-hardware founders as a priority cohort . The British Business Bank (BBB) direct mandate, given a £6.6bn deployment pot to lead venture rounds at up to £60m per company , is also absent across a sequence of recent direct cheques . IQT's house style is investing where the US intelligence community wants visibility on a technology; its appearance alongside NATO-IF at Series B is structurally unusual, and reads Fractile's SRAM compute as a dual-use national-security capability rather than a civilian deep-tech bet.

Liz Kendall pre-announced an AI Hardware Plan for London Tech Week on 9-12 June at her RUSI address in late April . The Hardware Plan is supposed to identify first-customer UK chip companies that DSIT will procure from. Fractile's round closing without UK state money three weeks before that launch reframes the plan's question: not which companies London picks, but whether picking them still matters once allied state capital has already moved.

Deep Analysis

In plain English

Fractile is an Oxford University startup building a new kind of computer chip designed specifically to run AI models. Instead of the usual setup where a chip fetches data from separate memory, Fractile's chip stores and processes data in the same place, which makes it much faster and uses less power for AI tasks. This week it raised $220m; roughly £175m; from investors including the venture arms of NATO and the CIA. No UK government programme put money in, despite the UK government having mentioned Fractile by name in a ministerial speech just weeks ago. The implication: a British chip company building technology that multiple governments want access to raised its biggest round yet without British state backing. The US intelligence community effectively bought a position in a UK chip startup before the UK did.

Deep Analysis

Root Causes

Three structural conditions explain why UK state capital was absent from Fractile's $220m round.

First, the SAIU's mandate requires UK-owned companies meeting four eligibility dimensions, and its published criteria (as of 16 May) still do not specify an ownership-percentage threshold. Fractile's pre-existing Oxford Science Enterprises position and the incoming NATO-IF and In-Q-Tel stakes create a cap table that DSIT may have assessed as incompatible with the sovereignty intent of the SAIU, even though no public criterion rules it out.

Second, the BBB's direct mandate caps at £60m per company and requires it to lead or co-lead at Series B; a £220m round led by Accel, Factorial and Founders Fund leaves the BBB in a follower position it may not have been willing to take at the price implied by the ~$1bn valuation.

Third, the AI Hardware Plan's procurement pledge; naming UK chip startups as first customers; is a forward commitment timed for London Tech Week in June. UK policy is operating on a six-week lag behind the capital market: Fractile raised on commercial and allied-security terms before DSIT's programme was ready to offer a credible alternative.

Escalation

The Fractile round escalates the allied-capital-fills-UK-gap pattern from anecdote to data point. The SAIU's first three equity investments were into UK-majority-owned companies with no foreign state capital on the cap table. Fractile now demonstrates that for UK chip startups at Series B, the commercial terms and allied-security capital are sufficient without UK sovereign participation.

What could happen next?

Risk
In-Q-Tel's standard governance rights may create technology-sharing obligations with US intelligence agencies before the DSIT AI Hardware Plan has specified procurement terms, constraining Fractile's freedom to sell chips to non-allied customers.
Short term · Suggested
Precedent
If the SAIU does not join Fractile's Series C, the allied-capital pattern becomes structural: UK chip startups route to NATO-IF and In-Q-Tel at growth stage and DSIT is left with procurement guarantees rather than equity positions.
Medium term · Assessed
Opportunity
Anthropic's customer interest gives DSIT a procurement angle: naming Anthropic as a Fractile chip buyer in the AI Hardware Plan announcement would make the UK government a market-maker without requiring an equity stake.
Short term · Suggested

Source Landscape

This story draws on neutral-leaning sources

Primary parallel: ARM Holdings' 1998 capital structure placed Texas Instruments and Motorola as strategic investors alongside UK venture capital, creating exactly the governance tension that played out 22 years later when Nvidia attempted acquisition.

US corporate strategic positions in a UK chip architecture company created implicit technology-access expectations that outlasted the investors and complicated the government's 2020 blocking review. Fractile now has US intelligence-community capital in an analogous position at Series B; earlier in the company's life than TI entered Arm, and with more explicit national-security intent.

Counter-parallel: Cambridge Display Technology (CDT), the Cambridge University spinout, raised successive rounds from US strategic investors including Dow Chemical in the late 1990s without triggering sovereignty concerns; it was eventually acquired by Sumitomo Chemical in 2007 with no UK government intervention.

CDT's experience suggests strategic investor positions do not automatically translate into blocked exits or constrained roadmaps, provided the company's IP is cleanly ring-fenced in UK entities.

Fractile's $220m at ~$1bn values the company at roughly 4.4x the £162.8m post-money from its prior round, a valuation step-up that is consistent with Series B hardware deeptech multiples in 2026 when there is a plausible near-term customer. Anthropic's early talks to buy custom inference chips is the customer signal that justifies the multiple: if Anthropic contracts at even $50m annually, Fractile's revenue multiple collapses to a reasonable range for a capital-intensive chipmaker.

The cap table carries an asymmetric risk for UK sovereign ambitions. NATO-IF and In-Q-Tel positions are typically 1-5% stakes at Series B, implying combined investment of $11m-$55m. Their economic exposure is small; their governance and technology-access rights are disproportionate to economic interest.

Fractile's next round, likely at Series C to fund tape-out and production, will need to price the discount for having allied-government investors with non-standard information rights; a cost UK state capital could have avoided if it had entered this round.

Consensus view: RUSI's Justin Bronk and Chatham House's Pia Hüsch both assess In-Q-Tel's participation at Series B in a UK chip startup as structurally unusual. In-Q-Tel typically enters at seed or early Series A, when technology is immature enough to shape. A Series B entry suggests Fractile's SRAM in-memory compute architecture has already passed internal US intelligence technology-readiness gates and is being treated as a near-term procurement target rather than a speculative bet.

Counter-view: The Ada Lovelace Institute's Kira Allmann argues the dual-use framing overstates the security implications. NATO-IF and In-Q-Tel invest in hundreds of companies annually; their presence on a cap table does not create binding procurement obligations or technology-transfer rights, and UK sovereignty over Fractile's roadmap remains intact as long as the company is UK-registered and the core IP sits in UK-incorporated entities.

Key tension: Whether In-Q-Tel's standard governance rights, which typically include technology-sharing clauses with US intelligence agencies, constitute a de facto prior claim on Fractile's inference architecture before the DSIT AI Hardware Plan even names Fractile as a first customer.

First Reported In

Update #5 · State capital splits, allied money fills gap

datacenterdynamics.com· 21 May 2026

Read original →

Causes and effects

Caused by

Sovereign AI Unit to launch with £500m

The SAIU's absence from Fractile's cap table is significant precisely because the £500m SAIU exists as the intended UK sovereign chip-investment vehicle.

Occurred 16 Apr 2026

Read story →

This Event

Fractile lands NATO and CIA chip cash

When allied national-security capital arrives before any UK state vehicle on a British chip deal, the absence is the signal.

Led to

CuspAI closes at $2.6bn, EU fund circles

Fractile's NATO/CIA-backed round with no UK co-investor set the precedent of allied capital capturing UK deep-tech upside, now repeated with EQT and CuspAI.

Occurred 2 Jul 2026

Read story →

StirlingX raises GBP15m for defence data

Fractile's allied-capital cap table established the same national-security investor appetite that funded StirlingX's Series A.

Occurred 30 Jun 2026

Read story →

Helsing picks Plymouth for £350m plant

Fractile's NATO/CIA-backed round without UK capital set the pattern Helsing's Plymouth plant follows.

Occurred 13 Jul 2026

Read story →

Different Perspectives

Highland Europe

Highland Europe, the growth-equity firm behind a €1bn-plus fund, took €65m from the British Business Bank into its Technology Growth Fund VI on 30 July via British Patient Capital. For a Geneva-based growth investor, the Bank's cheque is routine cornerstone capital, unrelated to which Whitehall department currently claims to sponsor UKRI.

Integrity Growth Partners

The Los Angeles firm put £16.1m into Prevalent AI on 24 August, the London GCHQ-alumni company's first outside capital, specifically to fund its US expansion. Its stake is a single commercial bet on one UK deep-tech founder's American growth, not a comment on which department UKRI answers to this month.

Temasek

Temasek led PhysicsX's Series C on 30 July, into which the British Business Bank put $25m through British Growth Partnership Fund I. State-backed pension capital rides behind a foreign-led growth round while UK weekly tech funding fell 72% to £102.95m across nine rounds three weeks later.

Innovate UK's Women in Innovation cohort

Sixty-one founders won £75,000 grants from Innovate UK on 5 August, the programme's largest-ever cohort, with a further 39 highly commended founders taking support without cash. For these founders the sponsorship dispute over UKRI's parent department is academic: the money and mentoring arrived exactly as scheduled.

Financial Conduct Authority

The FCA added five fintechs, ClearScore, Modulr, Teya, Urban Jungle and Zilch, to its Scale-up Unit on 7 August, the first cohort it regulates solely rather than jointly with the Prudential Regulation Authority. The unit's expansion is a routine regulatory build-out running on its own timetable, unconnected to the sponsorship dispute over its sister department UKRI.

Cabinet Office

The Cabinet Office ran the DSIT-to-BIST transfer as a standard cross-government machinery change, the kind gov.uk pages and framework documents routinely take months to catch up with after any department is abolished or renamed. Officials treat the paperwork lag as administrative sequencing, not dysfunction: funding and grant-approval layers moved first because they had to keep working.