Over the three days from 5–7 December 2025, a flurry of news and analysis has reshaped how investors, policymakers and engineers view Scale AI, the San Francisco‑based data and infrastructure company behind many of today’s leading AI models.
A rival startup has beaten Scale AI to a $708 million Pentagon contract, analysts are re‑pricing the company’s strategic value inside Meta’s “superintelligence” push, competitors are raising mega‑rounds off the back of a Scale AI customer exodus, and private‑market platforms have updated their pre‑IPO valuation data. At the same time, founder Alexandr Wang is back in the spotlight as profiles explain how a 14‑billion‑dollar deal propelled him to the top of Meta’s AI organisation. [1]
This article pulls together the latest news, forecasts and commentary dated 5–7 December 2025, and sets them in context of Scale AI’s turbulent year.
1. The big December headline: Scale AI loses a $708 million Pentagon contract
On 6 December 2025, financial news outlet AInvest reported that Enabled Intelligence, a US startup where around 60% of employees are on the autism spectrum, has secured a $708 million Pentagon AI contract, beating Scale AI in the process. [2]
According to that report:
- The deal is with the National Geospatial‑Intelligence Agency (NGA).
- It covers data labeling for geospatial‑intelligence AI and machine‑learning programs across multiple defense initiatives.
- Enabled Intelligence’s autism‑inclusive hiring model is framed as a strength for intensive pattern‑recognition and detail‑oriented defense data work. [3]
For Scale AI, this is notable because the company has spent years building its reputation as the Pentagon’s go‑to data partner. In 2022, Scale secured a $250 million Blanket Purchase Agreement with the US Department of Defense to “AI‑ready” defense systems — especially across the Air Force and other branches. [4]
What this loss signals
This single contract does not mean Scale AI is being pushed out of government work. The company still operates:
- A St. Louis AI Center dedicated to US government geospatial and LLM data needs. [5]
- Long‑term contracts that fall under its earlier DoD blanket agreement. [6]
But the Pentagon decision carries several implications:
- Procurement diversification
The NGA’s choice suggests US defense agencies want multiple specialist suppliers rather than relying primarily on Scale AI for sensitive data labeling work. - Mission‑fit beats brand‑name
Enabled Intelligence differentiated itself with an autism‑inclusive workforce tailored to high‑precision, repetitive pattern‑recognition tasks. In areas like geospatial intelligence, that specialization can outweigh Scale AI’s brand and scale. - Margin pressure in defense
If more high‑value defense contracts become competitive rather than default renewals, Scale AI may need to invest more aggressively in domain‑specific teams, security clearances and on‑prem offerings to defend its position.
In short: the Pentagon news is not an existential threat, but it is a clear warning shot that Scale AI’s once‑dominant position in government AI infrastructure is now under serious competitive pressure.
2. Meta’s $14.3B bet on Scale AI is still the centre of the story
Almost every fresh piece of December commentary still traces back to the June 2025 deal where Meta agreed to acquire a 49% non‑voting stake in Scale AI for roughly $14.3–14.8 billion, valuing the company at about $29 billion. [7]
Key points confirmed by Scale, Meta and multiple outlets:
- Meta’s investment values Scale AI at over $29 billion. [8]
- The deal includes a 49% stake but leaves Scale AI legally independent. [9]
- Founder and CEO Alexandr Wang stepped down from the chief‑executive role to join Meta as the leader of its new Superintelligence Labs division, while remaining on Scale AI’s board. [10]
Scale described the transaction as “the next phase of the company’s evolution”, highlighting deeper commercial integration with Meta’s AI efforts and reiterating its mission to “develop reliable AI systems for the world’s most important decisions.” [11]
December 7: New analysis ties Scale AI directly to Meta’s 2030 AI ambitions
On 7 December 2025, AInvest published a forecast titled “The Next AI Giants: Why Meta and Amazon Could Surpass Nvidia and Palantir in 5 Years”. The analysis argues that Meta and Amazon could rival or exceed Nvidia and Palantir in market value by 2030, largely due to their AI infrastructure bets. [12]
For Meta, Scale AI is singled out as a core pillar of that strategy:
- Meta has earmarked $64–72 billion in capital expenditure for 2025, a 128% jump from 2023, much of it for AI infrastructure. [13]
- The company has allocated $14.3 billion specifically to acquire Scale AI, which AInvest describes as a data‑labeling and model‑training specialist that shores up Meta’s “superintelligence” ambitions. [14]
- Scale’s pipelines are portrayed as essential to Meta’s effort to unify its ad models and foundation models under a single “Lattice” architecture, supporting a $60 billion annual ad run rate. [15]
In other words, the December 7 analysis suggests that Scale AI is no longer just a vendor; it is now seen by some market commentators as part of Meta’s internal infrastructure and a key reason investors are bullish on Meta’s AI upside.
3. Alexandr Wang’s new public image: From startup CEO to “superintelligence” boss
On 7 December 2025, India‑based outlet The Financial Express ran a widely‑syndicated profile titled “Who is Alexandr Wang? 28‑year‑old hired for 14 billion US dollar by Mark Zuckerberg to lead Meta superintelligence labs.” [16]
The piece recaps Wang’s trajectory:
- New Mexico‑born son of Chinese‑immigrant physicists, with early talent in maths and computing.
- Dropped out of MIT in 2016 to found Scale AI, betting on surging demand for high‑quality annotated data.
- Grew Scale into a critical supplier of labeled data to Nvidia, Amazon and Meta, reaching a valuation near $14 billion by 2024. [17]
The article underscores that Meta’s $14.3 billion investment in Scale AI is what effectively “hired” Wang to lead Meta Superintelligence Labs (MSL), giving him sweeping authority over the company’s AI research, infrastructure and product direction. [18]
Financial Express quotes an internal memo in which Wang reportedly tells Meta staff that “superintelligence is coming, and in order to take it seriously, we need to organize around the key areas that will be critical to reach it.” [19]
Why this matters for Scale AI
Although Wang no longer runs day‑to‑day operations at Scale AI, this renewed media attention affects the company in several ways:
- It cements the narrative that Scale AI is deeply embedded in Meta’s most ambitious AI agenda.
- It reinforces perceptions — both positive (world‑class talent) and negative (conflicts of interest) — among other hyperscalers and foundation‑model labs choosing data partners. [20]
- For employees and potential recruits, it presents Scale AI as part of a wider “superintelligence” ecosystem, not just a data‑labeling shop.
4. Customer exodus and a $10B rival: Mercor’s rise on 7 December
If Meta’s stake elevated Scale AI, it also created new vulnerabilities.
On 7 December 2025, tech site The Tech Buzz updated its feature “Mercor Hits $10B Valuation as AI Training War Intensifies.” The article confirms that AI talent marketplace Mercor has closed a $350 million Series C at a $10 billion valuation, up 5× from its $2 billion Series B just eight months earlier. [21]
Crucially for Scale AI, the piece explicitly ties Mercor’s surge to fallout from the Meta deal:
- Mercor’s revenue spike “follows Scale AI’s dramatic exit from major clients after Meta’s $14 billion investment created conflicts of interest.” [22]
- When OpenAI and Google DeepMind reportedly needed new data‑labeling partners, Mercor was “perfectly positioned” with its network of 30,000+ specialized contractors, paying out more than $1.5 million per day — over $500 million a year. [23]
Earlier in 2025, Computerworld and independent analysts had already warned that Meta’s stake could trigger a Scale AI customer exodus, arguing that many AI labs would be uncomfortable sharing training data, model artifacts and sensitive prompts with an infrastructure partner partially owned by a direct competitor. [24]
The updated December 7 coverage does two things:
- Confirms those warnings were not hypothetical — at least some major labs have moved spend to rivals.
- Positions Mercor as the primary beneficiary of this shift, effectively splitting the data‑labeling and human‑in‑the‑loop market that Scale once dominated.
For Scale AI, this increases pressure to:
- Lean harder into public‑sector and enterprise customers where Meta’s stake is less of a direct competitive concern.
- Differentiate on security, compliance and long‑term contracts, not just speed and price.
- Push further up the stack into applications, evaluation frameworks and agents, where Meta’s involvement is an asset rather than a liability.
5. Government & enterprise footprint: Office expansion and defence presence
While analysts focus on conflicts and contracts lost, Scale AI has been quietly expanding its on‑the‑ground presence.
On 5 November 2025, Scale published a corporate blog post, “Expanding Our Presence with New Offices Around the World”, outlining four key office moves: [25]
- St. Louis, Missouri – A larger AI Center in Downtown North, designed to support geospatial and LLM data needs for the US government, with capacity to double the headcount from ~200.
- London, UK – New international HQ in King’s Cross, expanding capacity from 30 to 90 seats and placing Scale in the middle of London’s AI cluster.
- Washington, D.C. → Arlington’s National Landing – A relocation to a larger office opening in December 2025, expanding capacity from 30 to 100 federal‑facing staff and improving proximity to defense and government customers.
- New York City – A move to One World Trade Center in early 2026, growing from 120 to 500 team members to serve enterprise and AI‑lab clients.
Combined with its exhibitor presence at I/ITSEC 2025 — a major military training and simulation conference held in Orlando from 1–5 December 2025 — this paints a picture of a company deeply embedded in defense and national‑security ecosystems, even as specific contracts shift to new competitors. [26]
The messaging across Scale’s own channels remains consistent: it wants to be the “reliable AI systems” provider for the world’s most important decisions, sitting at the junction of top AI labs, enterprises and government institutions. [27]
6. Research signal at NeurIPS 2025: Scale AI moves deeper into reasoning and evaluation
Beyond contracts and offices, Scale AI is also using NeurIPS 2025 — the field’s flagship research conference, held in San Diego in December — to signal how it sees the future of AI infrastructure.
Scale’s events page lists multiple workshop contributions on 6 December 2025, including: [28]
- “Rubrics as Rewards: Reinforcement Learning Beyond Verifiable Domains”
- “Adaptive Guidance Accelerates Reinforcement Learning of Reasoning Models”
- “Reasoning GYM: Reasoning Environments for Reinforcement Learning with Verifiable Rewards”
All of these sit inside the broader theme of efficient reasoning: how to train and evaluate models not just on text prediction, but on multi‑step problem solving, planning and decision‑making.
Taken together with Scale’s GenAI Platform and its evaluation‑oriented products, the NeurIPS program suggests that Scale wants to:
- Own more of the evaluation and reinforcement‑learning stack (beyond classic RLHF).
- Provide standardized reasoning benchmarks and environments that plug directly into foundation‑model training pipelines.
- Position itself as indispensable to labs that care about verifiable reasoning quality, not just raw capability.
That focus becomes especially important as foundation models transition from “chatbot demos” to mission‑critical agents in finance, defense, logistics and code generation — domains where reasoning errors can be extremely costly.
7. Valuation, private‑market trading and IPO timing
The other big December 5–7 development around Scale AI is financial, not technical.
Private‑equity marketplace Forge Global updated its Scale AI stock pages on 5 December 2025, providing a window into how secondary markets are currently pricing the company. [29]
Key details from Forge:
- The “Scale AI stock” page shows a Forge Price of $20.00 per share as of 7 December 2025, a derived indicator based on recent secondary trades and bids. [30]
- The “Scale AI IPO” overview lists a Series G‑2 valuation of $29.18 billion in June 2025, roughly in line with Meta’s reported deal pricing. [31]
- Forge emphasizes that Scale AI has not filed for an IPO, highlighting that alternatives such as SPACs, direct listings or M&A remain possible. [32]
Earlier this year, venture newsletter Launchbay Capital described Meta’s investment as “one of the largest secondary liquidity events in VC history”, estimating that the deal generated about $12.8 billion in cash payouts to Scale AI shareholders. [33]
That unusual structure means:
- Many early investors and employees have already partially cashed out, reducing IPO timing pressure.
- Meta gets significant economic exposure and access to Scale’s talent and infrastructure without a full acquisition.
- Public‑market investors are left to trade in the secondary market, where liquidity is thinner and price signals are noisier.
For now, the December 5 Forge updates indicate ongoing demand for Scale AI shares at valuations broadly consistent with the Meta deal, despite contract losses and competitive headwinds.
8. Risk and opportunity: What the 5–7 December window reveals about Scale AI’s future
Pulling these threads together, the news and analysis from 5–7 December 2025 point to a crossroads moment for Scale AI.
Short‑term pressure points
- Defense competition is real
Losing a $708 million NGA contract to Enabled Intelligence confirms that defense agencies are willing to back specialized, mission‑aligned rivals even when Scale AI has a strong track record. [34] - Conflicts of interest are reshaping the market
The newly updated Mercor story shows that some of Scale’s largest commercial customers have already diversified away, helping a competitor reach a $10 billion valuation in under three years. [35] - Regulatory & antitrust scrutiny isn’t going away
Commentary on Meta’s 49% stake has repeatedly raised questions about neutrality, data governance and competition, and December coverage continues to frame the deal as both a strategic masterstroke and a potential regulatory red flag. [36]
Long‑term strengths
- Deep entanglement with Meta’s superintelligence strategy
AInvest’s December 7 forecast and the Financial Express profile both underscore that Meta views Scale AI as core infrastructure for its multi‑trillion‑dollar AI ambitions, rather than a bolt‑on supplier. That alignment could translate into massive, long‑duration revenue if Meta executes on its plan. [37] - Global footprint and government credibility
Expanded offices in St. Louis, London, National Landing and New York, plus visibility at I/ITSEC and previous DoD agreements, give Scale AI a strong platform to keep winning public‑sector and regulated‑industry work, even as particular contracts go to rivals. [38] - Technical depth in reasoning and evaluation
Its NeurIPS 2025 program shows Scale investing in reinforcement‑learning environments and reasoning benchmarks — a layer of the stack that will likely become more important as AI systems move from “cute chatbot” to mission‑critical decision‑maker. [39]
Strategic outlook
From an industry‑watcher standpoint (not investment advice):
- Base case: Scale AI leans into its role as Meta’s data and evaluation backbone, while selectively retaining high‑margin government and enterprise work where conflicts are manageable.
- Upside case: It successfully convinces other big labs that contractual firewalls, on‑prem deployments and independent governance make it a safe “trusted third party” even with Meta’s stake, allowing it to regain some lost ground. [40]
- Downside case: Customer exodus accelerates, regulators apply tighter scrutiny to Meta‑Scale data flows, and rivals like Mercor and Enabled Intelligence carve up the most lucrative segments of the market before Scale can reposition.
What the 5–7 December 2025 news window makes clear is that Scale AI is no longer just a quiet infrastructure company. It sits at the centre of geopolitics (Pentagon contracts), big‑tech power plays (Meta’s superintelligence labs) and financial speculation (pre‑IPO secondary markets).
For enterprises and governments deciding whether to build on Scale, the message is nuanced:
- The company offers maturity, scale and deep integration with one of the world’s most aggressive AI strategies.
- But it also comes with non‑trivial strategic and competitive baggage, especially for organisations that compete with Meta or have stringent neutrality requirements.
Those trade‑offs — laid bare by the latest December coverage — will define how big a role Scale AI plays in the next chapter of the AI era.
References
1. www.ainvest.com, 2. www.ainvest.com, 3. www.ainvest.com, 4. aibusiness.com, 5. scale.com, 6. aibusiness.com, 7. www.reuters.com, 8. scale.com, 9. www.reuters.com, 10. techcrunch.com, 11. scale.com, 12. www.ainvest.com, 13. www.ainvest.com, 14. www.ainvest.com, 15. www.ainvest.com, 16. www.financialexpress.com, 17. www.financialexpress.com, 18. www.financialexpress.com, 19. www.financialexpress.com, 20. www.computerworld.com, 21. www.techbuzz.ai, 22. www.techbuzz.ai, 23. www.techbuzz.ai, 24. www.computerworld.com, 25. scale.com, 26. iitsec2025.smallworldlabs.com, 27. iitsec2025.smallworldlabs.com, 28. scale.com, 29. forgeglobal.com, 30. forgeglobal.com, 31. forgeglobal.com, 32. forgeglobal.com, 33. launchbaycapital.com, 34. www.ainvest.com, 35. www.techbuzz.ai, 36. medium.com, 37. www.ainvest.com, 38. scale.com, 39. scale.com, 40. www.computerworld.com
