The Denial, the Investigation, and the Cancelled Ceremony
Jim Keller Publicly Denies the Qualcomm Deal This Series Reported Last Episode. CNBC Finds Chinese Models Now Handle Up to 46% of Enterprise AI Traffic in the US. And the White House Cancels an AI Executive Order Signing Without a Word of Explanation.
Tenstorrent CEO Jim Keller publicly denied being in acquisition talks with Qualcomm, directly contradicting the reporting this series covered in Episode 23 — a correction owed to readers before anything else this week. Separately, CNBC published an investigation showing Chinese AI models now account for between 30% and 46% of enterprise API token usage flowing through US developer platforms via OpenRouter, up from an 11% average over the prior twelve months. And the White House cancelled a planned AI executive order signing ceremony with no stated reason. Ground Truth covers a correction, a shift that snuck up on the industry, and a silence worth watching.
A denial, an investigation nobody was watching for, and a ceremony that quietly never happened do not share an obvious throughline. What they share is a pattern this series keeps running into: the loudest reported story of a given week is often not the one that matters most a month later. Sometimes the correction is the real story.
Ground Truth · Episode 24 · July 12 2026Episode 23 reported that Qualcomm was in early talks to acquire Tenstorrent for $8 to $10 billion, based on reporting from The Information. On June 30th — before this series’ own episode published — Tenstorrent CEO Jim Keller publicly denied it, stating the company has not been in acquisition discussions and intends to focus on its own business. Qualcomm shares dipped roughly 1% on the original report and have since normalised. This series corrects the record here, plainly, before moving to two stories that broke fresh this week. First, CNBC’s investigation into how deep Chinese AI models have already penetrated US enterprise infrastructure, mostly unnoticed. Second, a cancelled ceremony at the White House that nobody has explained. Welcome to Episode 24 of Ground Truth.
Jim Keller Says There Was No Deal. This Series Believes Him.
Here is what this series can now say plainly: The Information reported on June 15th that Qualcomm was in early talks to acquire Tenstorrent at an $8 to $10 billion valuation. Jim Keller, on June 30th, said publicly that no such discussions were happening and that Tenstorrent’s focus remains its own roadmap. Episode 23 ran with the acquisition story on July 10th, ten days after Keller’s denial was already public — a sourcing lag this series should not have let through, and takes responsibility for.
The underlying strategic logic this series laid out last episode still holds regardless of whether this specific deal happens: Nvidia and AMD dominate AI accelerator silicon, every major buyer wants a credible third option, and RISC-V-based alternatives like Tenstorrent’s Galaxy Blackhole platform represent a real category worth watching even without a Qualcomm acquisition attached to it. Tenstorrent’s Galaxy Blackhole, launched earlier this year, packs 32 accelerators with 768 RISC-V cores each into a single 6U data centre unit — a genuinely notable product on its own merits.
What this correction is really about is sourcing discipline. Early-stage M&A reporting is inherently uncertain, denials happen, and talks fall through constantly in this industry. The failure here was not treating an early report as early; it was not checking whether that report had already been superseded by the time this series covered it. That is on us, and it will not happen again without a same-day verification pass before publication.
Thrive Holdings is reportedly raising $2 billion, with Altimeter, D1, and SoftBank participating, specifically to acquire controlling stakes in professional services firms with combined annual revenues in the tens of billions. The bet: AI-driven disruption of professional services (accounting, legal, consulting) is fundable today, not a theoretical thesis for later — institutional capital is already positioning for it.
Up to 46% of Enterprise AI Traffic Nobody Was Tracking Closely Enough
CNBC’s investigation, published July 7th, found that Chinese AI models — primarily accessed through the OpenRouter gateway that many enterprise developer platforms route through by default — now account for between 30% and 46% of all enterprise API token usage flowing through US developer platforms. That figure has been above 30% every week since February 8th, 2026. The average across the prior twelve months was just 11%, and had been as low as 4.5% before that. This is not a sudden spike; it is a steady, largely unremarked climb that crossed a genuinely significant threshold while most coverage stayed focused on frontier-lab headlines.
The pull is straightforward economics. Models like GLM-5.2 and DeepSeek V4-Pro offer performance competitive with mid-tier Western models at a fraction of the cost, and gateway routing makes switching between providers close to frictionless for a developer who is optimising for cost per token rather than provider identity. Most engineering teams routing traffic through OpenRouter or similar gateways are not making an explicit, deliberated choice to use Chinese models — they are letting a cost-optimisation router make that choice for them, request by request.
The risk profile is different enough from routing to a Western provider that this series is devoting all of Inside The Machine’s next entry to it specifically — content restrictions, data jurisdiction, and governance questions that a pure cost comparison does not surface. The headline number here is simply this: a meaningfully large and fast-growing share of US enterprise AI traffic is already flowing somewhere most of those enterprises’ own compliance and security teams have not explicitly evaluated.
Nobody at most US enterprises decided, in a meeting, to route sensitive workloads to a Chinese model. A cost-optimising gateway decided it for them, one API call at a time, and the decision only became visible once an investigative team added up the traffic months later. That is not a security failure exactly. It is something closer to a visibility failure, and it is arguably worse.Neal Lloyd · Ground Truth, Episode 24
No Signing, No Statement, No Timeline
The White House had a planned signing ceremony for an AI executive order this week. It did not happen, and no public statement has explained why. This series has tracked the voluntary AI release standards thread since Episode 20 — the Financial Times reporting on advanced talks between the administration and OpenAI, Google, and Anthropic, with an announcement once expected as soon as the week of July 7th. A cancelled signing ceremony without explanation is consistent with several very different stories: last-minute disagreement over specific terms, a company objecting to language it had not previously seen, an unrelated scheduling conflict, or something with no connection to the substance at all.
This series is not going to guess which of those it is, and treats any confident explanation offered elsewhere this week with real scepticism until an official statement or credible reporting fills the gap. What is verifiable is only this: a ceremony that multiple outlets expected to happen this week did not happen, and as of this writing nobody involved has said why.
The reason this earns a section rather than a footnote is the pattern it fits. This series has now covered three separate instances — the GPT-5.6 delay in Episode 20, the Gemini 3.5 Pro delay across multiple episodes, and now this — where the AI industry’s biggest signals this month have been absences rather than announcements. A launch that does not happen, a signing that does not happen, a deal that turns out not to exist: reading what is missing is becoming as important as reading what gets announced.
A cancelled ceremony with no explanation is not a story. It is the absence of one, which this series has learned this month is sometimes the more important signal to sit with, rather than fill the silence with a confident guess just to have something to say.Neal Lloyd · Ground Truth, Episode 24
Ground Truth, Episode 24 · July 12 2026
Neal Lloyd covers the real-world impact of AI — money, power, geopolitics, and the stories behind the headlines. Ground Truth is his daily AI news and analysis series on emdexter.blogspot.com.
- Ep 01The Gold Rush
- Ep 02ChatGPT Knows Everything
- Ep 03Siri Is Now Google
- Ep 04America’s AI Law Is a Mess
- Ep 05Is AI Taking Your Job?
- Ep 06Microsoft vs Everyone
- Ep 07SpaceX Is Trading
- Ep 08The Government Pulled Fable 5
- Ep 09Trump and Bernie Want to Own AI
- Ep 10SpaceX Buys Cursor for $60B
- Ep 11The Fable 5 Truth
- Ep 12The Week That Changed Everything
- Ep 13Bots Now Outnumber Humans
- Ep 14Colossus: $80B Compute Landlord
- Ep 15ChatGPT Is Getting Ads
- Ep 16Alibaba Stole 28.8M Conversations
- Ep 17June 2026: The Month AI’s Rules Changed
- Ep 18The Credibility Gap
- Ep 19Sonnet 5, Fifty States, and the Return of Fable 5
- Ep 20Now Every Frontier Model Needs a Permission Slip
- Ep 21193 Countries, One Room, and a Compute Gap Nobody Can Vote Away
- Ep 22The Free Ride Ends, the Real Commission Meets, and Gemini Finally Shows Up
- Ep 23The Fable 5 Ban, Fully Told
- Ep 24The Denial, the Investigation, and the Cancelled CeremonyYou are here
