Newbird AI was, until earlier this month, a struggling sustainable trainer brand called Allbirds. Its market value had fallen from roughly $4bn at IPO to $22m. Then it sold the shoes for $39m, picked a new name, declared it would build silicon chips and data centres, and saw its share price rise nearly six-fold despite zero relevant experience. The Telegraph’s Russ Mould flagged this as the kind of late-cycle behaviour that defined the “onics and tronics” boom of the 1960s, the dotcom bubble of the 1990s, and the rash of blockchain-themed rebrandings earlier this decade. Whether AI itself is in a bubble is a question for equity strategists. The operational question for UK leaders, procurement teams and investors is more immediate: when an AI claim arrives at your desk, how do you tell genuine capability from a polished name change?
The signal Mould is sending
Strategists at investment banks watch micro-cap reinventions because they often arrive earlier than the obvious crash signals. By the time IPO pipelines, margin debt, and leveraged ETF flows show clear distress, the move to cash is usually already late. The Newbird AI episode matters not because one struggling trainer firm renamed itself, but because its share price moved nearly six-fold on a press release alone. That gap — between zero demonstrated capability and a multi-fold valuation jump — is the bubble-layer signal.
Critical Context: AI is the technology. The bubble layer sits on top of it — the fund flows, the rebrands, the “AI-powered” prefixes attached to ordinary software. UK leaders need to engage with the technology and stay sceptical of the layer simultaneously. They are not the same thing.
For UK procurement teams, boards and investors, this is less a market-timing question than a counterparty-quality question. The volume of AI claims arriving in commercial contexts has risen sharply. Some are genuine — a supplier has built transformer-based extraction into its document workflow with measurable accuracy gains. Some are hollow — the same supplier has bolted a chatbot onto its login page and rebranded as “AI-first”. Both walk through the procurement door with similar marketing pages.
Critical numbers — Allbirds to Newbird AI
| Metric | Value |
|---|---|
| Valuation at Nasdaq IPO (late 2021) | ~$4bn |
| Valuation by start of April 2026 | $22m |
| Cumulative after-tax losses | ~$450m |
| Sale price of the shoe business | $39m |
| Share-price reaction to AI rebrand | ~6x on announcement |
| Valuation post-retreat | ~$75m |
| Demonstrated experience in chips, data centres or AI infrastructure | None |
What separates real AI from a rebrand
Three things compounded into the Newbird AI move. First, a distressed micro-cap with little to lose attempted a strategic reset — unremarkable on its own. Second, the chosen reset sector was the one where capital is currently easiest to attract, and where a credible operating history is, for the moment, only weakly required by some retail buyers. Third, public discourse has been primed for months by genuinely large AI capital deployments at the hyperscalers, blurring the line between “AI is being heavily invested in” and “this specific company will benefit”.
The combination produces what looks like a rational market response — six-fold appreciation — to what is, on examination, a press release without a product, a team without a track record, and a balance sheet without the capital intensity that real AI infrastructure requires. UK readers should note that the same pattern is starting to appear in domestic listings, particularly on AIM and at the smaller end of the main market, where the cost of issuing a strategic update is low and the upside of catching an AI tailwind is high.
Strategic Insight: A pivot is not the same as a capability. The presence of an “AI strategy” in a release tells you the company has noticed AI; it tells you nothing about whether it can build, integrate or operate it. Treat strategy announcements and capability claims as separate evidence categories.
Genuine AI adoption inside an organisation almost always carries observable footprints. Cosmetic adoption rarely does, because producing the footprints is what costs money.
- Capital and talent commitments that match the claimed ambition. Real AI infrastructure is capital-intensive and people-intensive. A firm claiming to provide silicon chips and data centres needs balance-sheet capacity and named hires from the relevant industries — not aspiration.
- Measurable internal use before external claims. Companies running real AI internally usually have concrete metrics: hours saved, accuracy lifts, defect reductions, customer-resolution times. The absence of any internal proof point alongside an external claim is informative.
- Architecture-level specificity rather than vendor-mediated description. A team with genuine AI capability can describe its model choices, its retrieval and grounding approach, its evaluation harness, and its failure modes. A rebrand cannot.
- A roadmap that survives questioning. Plans built on real capability respond coherently to “what would have to be true for this to fail?” Plans built on positioning tend to deflect the question.
For most UK organisations, the practical task is not to spot bubble-layer fraud at the listed-company level — that is a job for regulators and equity analysts. It is to tell the difference, week to week, between suppliers, prospective hires, partners and acquisition targets that have built something and those that have rebranded. The same diligence stance protects internal projects too: a programme inside your own organisation can fall into the same trap, accumulating slide decks and “AI” labels without producing a model, an integration, or a measured outcome.
Where AI-washing actually hurts UK organisations
The pressure on executives to be visibly “doing AI” is structural. Boards ask, analysts ask, journalists ask, and competitors signal. Under that pressure, the path of least resistance is to relabel rather than rebuild. Naming an internal team “AI Centre of Excellence”, appending “AI-powered” to existing products, or commissioning a high-visibility pilot can satisfy the optics question for several quarters before producing any operational outcome. Newbird AI is the publicly-traded version of a pattern that already runs through corporate UK at lower volume.
Procurement and investment teams sit at the choke point. They see the AI claims first and have the formal authority to require evidence. The diligence stance below is designed to be applied without specialist data-science staffing — although bringing one or two technical assessors into the room for material commitments is a sensible upgrade.
| Stakeholder | What “AI-washing” costs them |
|---|---|
| Boards and investment committees | Capital allocated against capabilities that do not exist; later write-downs once measurement begins |
| Procurement teams | Vendor lock-in to a brand-led “AI” product whose underlying engine is a rules engine or a thin GPT wrapper |
| Operational managers | Project failure when the promised capability does not arrive, and reputational damage from over-promised internal rollouts |
| Investor relations and equity holders | Re-rating risk if the market shifts from rewarding AI claims to penalising unsupported ones |
| Regulators and auditors | Growing pressure to define AI-claim disclosure standards as the volume of claims rises |
A diligence outcome is useful when it produces one of three clear answers: capability is real, capability is partial-but-trajectory-credible, or capability is absent. Ambiguity is itself a signal — if a counterparty cannot be classified after a structured exercise, the default should be to assume absence, not presence.
Reality Check: AI-washing at a counterparty is not a moral failing you need to call out. It is a state of the world you need to price into the deal — through warranties, milestone-based payments, walk-away triggers, or a smaller initial commitment with the option to scale.
A four-question diligence test
Apply this in any procurement, investment, partnership, or acquisition conversation where AI capability is material to the decision.
- What problem is the AI solving, and for whom? A coherent answer names a user, a workflow, and a measurable outcome. An incoherent answer references “the customer journey” or “operational efficiency” without specifics.
- Where is the model, and where is the data? Real AI capability has model architecture (own, fine-tuned, or via a named provider) and a defined data flow. If the answer is “we use AI” with no follow-up, capability is unverified.
- What does the capability replace or augment, and by how much? Genuine adoption produces displacement or augmentation that can be measured. Without a baseline and a delta, the claim is positioning.
- What evidence has the team produced internally before this claim? Most real AI capability arrives by being used internally first, then exposed. External-only AI claims with no internal track record are weakest.
Take Action: Add the four-question test to your standard procurement and partnership diligence packs. Refusing to ask the questions is the choice — once they are on the form, suppliers either answer them or visibly fail to.
Early-stage AI engagement (no material AI in operations yet). Apply the framework defensively — focus on procurement and supplier claims. Decline the rebrand premium: do not pay more for software because it acquired an “AI-powered” prefix in the last release notes. Build a simple internal register of AI claims encountered and what evidence supported them.
Mid-stage (one or two production AI use cases). Apply the framework to your own programmes as well as to vendors. Insist on capability metrics inside steering committees — accuracy, throughput, cost-per-task — alongside narrative updates. Actively distinguish between “we use a vendor’s AI feature” and “we have built AI capability”.
Mature (multiple production deployments, internal capability). Use the framework as a counterparty filter at deal flow. Where a target or partner cannot answer the four questions credibly, structure the commercial terms to reflect uncertainty — staged payments, capability milestones, walk-away clauses. This is the stage at which AI-washing risk becomes monetisable in deal terms.
What the framework misses
Four challenges sit beneath the surface of the diligence framework. None of them is fatal, but each rewards being named explicitly.
The framework can be gamed. Once the four questions are well-known, sophisticated counterparties will rehearse credible-sounding answers. Mitigation: ask for artifacts, not narratives. Model evaluation logs, internal usage telemetry, named hires with verifiable backgrounds, and reference customers willing to take a follow-up call. Artifacts are harder to fabricate than answers.
The technology is moving faster than diligence frameworks. A 2026 diligence question may be partly obsolete by 2027. Mitigation: review the framework annually with someone holding current technical context, and treat the four questions as a structure to adapt rather than a fixed checklist.
False negatives are also a cost. Over-strict diligence will reject some genuine but immature capability — particularly from smaller suppliers and earlier-stage targets that have built something real but lack the evidence trail of an established firm. Mitigation: distinguish “no capability” from “early capability with credible trajectory”, and use staged commercial structures for the latter rather than rejection.
Internal politics make calling out internal AI-washing the hardest case. Diligence applied to suppliers is uncontroversial. Applied to a colleague’s flagship internal initiative it can be career-affecting. Mitigation: position the framework as a standard governance mechanism applied uniformly, owned by a function (audit, risk, finance) rather than by an individual challenger.
Hidden Cost: The most expensive AI-washing exposure in most UK organisations is not from external vendors. It is from internal programmes that have absorbed budget, leadership attention, and headcount on the strength of a label rather than a capability. These are politically costly to confront and economically costly to ignore.
What this means for your next AI decision
The Newbird AI episode is, on its own, a small story — one micro-cap, one rebrand, one short-lived rally. Its value is as a calibration point. If a market can six-fold a company on the basis of a press release with no operational substance, the same dynamics are at work, in quieter form, throughout the supply chains, partnership markets and internal programmes that UK organisations participate in every week.
Genuine AI adoption produces evidence: capital, hires, internal use, metrics, named architecture, willingness to be questioned. Cosmetic AI adoption produces marketing: positioning, prefixes, slide decks, press releases. The difference is observable if you ask, and is hidden only if you do not. A diligence framework does not eliminate AI-washing risk — it prices the risk into commercial decisions instead of letting it accrue silently.
Three success factors carry most of the value:
- Treat strategy announcements and capability claims as separate evidence categories. A firm having an AI strategy is not evidence of AI capability.
- Insist on artifacts over narratives. Logs, telemetry, named hires, reference calls. Anyone can rehearse the right answers; few can fabricate the right artifacts.
- Apply the framework internally, not only externally. The most expensive AI-washing in any organisation is usually its own.
A short next-step checklist:
- Add the four-question test to procurement and partnership diligence templates.
- Identify the top three AI-related commitments your organisation has in flight (vendor, partner, or internal) and apply the test retrospectively to each.
- Where any of those three fails, define the smallest credible commercial intervention — a milestone, a clause, or a halt-and-review — rather than escalating to confrontation.
- Schedule an annual review of the framework with a technical assessor, recognising that the questions will need to evolve.
- Hold the line on the rebrand premium: do not pay more for the same product because its name now ends in “AI”.
Mould’s call is for prudence at the market level. The same posture, applied at the organisational level, is what will distinguish UK firms that benefit from AI in 2027 from those that spend the year explaining write-downs.
Source
Mould, R. (27 April 2026). “An ailing shoe company could be about to burst the AI bubble.” The Telegraph. Available at: https://www.telegraph.co.uk/money/investing/stocks-shares/allbirds-failing-shoe-company-signal-end-ai-boom/
Resultsense provides UK-focused AI news, analysis and strategic insights for professionals and businesses making decisions about AI adoption.