A Nasdaq newcomer is selling WEKA's self-measured 1000x claim, weeks after WEKA's CTO left

WEKA's 1000x and 20x figures, measured only by WEKA, are now load-bearing in a Nasdaq partnership announcement that came eight days after the company's CTO of six years walked out. Neither BIRD's stock nor WEKA's private mark has an independent benchmark behind it.

Vincent JiangVincent Jiang · 3 min read
Share
Shimon Ben-David, WEKA's chief technology officer of six years, speaking on a Six Five on the Road panel before his departure
1 / 5Slide 1 of 5
Shimon Ben-David, WEKA's chief technology officer for more than six years, whose departure was reported eight days before Smartbird's partnership announcement with WEKA.

On October 22, 2026, Smartbird, Inc 1 2. (NASDAQ: BIRD), an AI infrastructure company that sold its shoes four months earlier, put a set of performance claims in front of the stock market. Its partnership release with WEKA leads with the numbers: WEKA~$1.2B — WEKA, private, latest valuation ~$1.2B NeuralMesh extends GPU memory capacity 1000x, cuts time to first token by up to 20x, and serves 10x more concurrent users from the same GPU footprint 1 2. The source for all three is the same: WEKA, stated in production benchmarks WEKA itself measured. 1 2

The receipts are one company's word

Futurum Group, an analyst firm, published the only outside assessment of this deal so far, on October 23 2. It repeated the multiples, called them production benchmarks, and then listed its own watch item: whether independent benchmarks confirm the 20x and 1000x claims in customer production environments 2. That watch item is still open. 2 The 1000x figure matters commercially, because Smartbird's pitch to banks, hospitals and governments is that its dedicated clusters, running NeuralMesh, deliver cloud-grade performance without data ever leaving a customer-controlled environment. 1

The three multiples carrying a Nasdaq partnership are all WEKA-measured

  • Estimate
0x200x400x600x800x1,000xGPU memory capacity extension1,000xTime to first token acceleration20xConcurrent users per GPU footprint10x
Data
Value
GPU memory capacity extension (estimate)1,000x
Time to first token acceleration (estimate)20x
Concurrent users per GPU footprint (estimate)10x
Multipliers as stated by WEKA in its own release, labeled 'production benchmarks'; no independent confirmation exists yet. Sources: [1], [2].1,2

The engineer behind the numbers just left

Eight days before the announcement, on October 14, StorageNewsletter reported that Shimon Ben-David, WEKA's chief technology officer for more than six years and an employee for nearly 13, had announced his departure on LinkedIn; WEKA's leadership page already showed the change. 3

WEKA's current leadership page confirms it: nine executives listed, no CTO among them. 4 The company has not said who now owns the engineering behind NeuralMesh. WEKA's last disclosed raise remains the January 2022 round: $73 million led by Hitachi Ventures, taking total funding to $140 million, with NVIDIA−0.54% — NVIDIA, down 0.54 percent today, HPE+2.04% — HPE, up 2.04 percent today, Micron+2.35% — Micron, up 2.35 percent today and Cisco+1.63% — Cisco, up 1.63 percent today participating 5. A trade outlet resurfacing that release this week under a "WEKA Increases Funding" headline is republishing 2022, not new money. 5

What a buyer of the ticker is actually underwriting

The listed company here is not WEKA. It is Smartbird, the former Allbirds, worth nearly $4 billion at its 2021 listing and under $20 million by early 2025 before it sold the footwear business and renamed itself 6. The stock briefly rose 800% on the pivot, then gave much of it back. 6

Its whole enterprise-AI stack is new, its first ecosystem partner is WEKA, and the performance evidence for the pairing is one press release co-signed by both parties. 1 A reader deciding what to pay for BIRD is being asked to price WEKA's self-measured multiples through a June 2026 startup wearing a sneaker company's ticker, while the vendor's long-serving technologist walks out the door 1. The discipline is simple: wait for a benchmark WEKA did not run. The next reading arrives when a customer, an analyst, or an auditor publishes one.

Get Data Center by email, free

Who is building AI's data centers, and who is paying for them. In our daily email, every weekday.

Loading MU…

Reader comments

0 comments

    More in Data Center

    All in this topic →

    Data Center

    Corning bets $585 million on AI fiber, and Monday's earnings decide if it was worth it

    Corning approved a $585 million fiber expansion on AI data-center demand days before earnings that will show whether that demand survives at the price the stock already charges for it.

    Data Center

    Abbott took $9 million from data center donors since July 2025; the November 3 election sets the rules

    Data center donors gave Governor Greg Abbott his biggest months right after he moved against their economics, and every Texas office that will write the industry's costs takes the same money. The November 3 election decides who sits in those chairs when the 2027 session negotiates the terms.

    Data Center

    Modine reports November 3 with Wall Street at $305 and the tape at $133

    Modine's first results since giving its car business, and its name, to Gentherm land after the close on November 3, with the stock at $133.24 against a $305.67 average analyst target. The print tests whether 90 percent data-center growth can carry 56 times earnings on a falling gross margin, ten days before shareholders vote on the Modexus rename.

    Data Center

    Meta's new share class funds AI, and lets Zuckerberg sell 20% of it for a possible California wealth tax

    Meta's new filing pairs a share class built to raise hundreds of billions for AI with a 20% carve-out letting Zuckerberg sell into it to pay a potential California wealth tax, in exchange for eight years of residency. Shareholders carry the discount.

    Data Center

    Alphabet's $195 billion 2026 capex funds Gemini, and DeepMind gets the compute first

    Alphabet reports October 28 with a class action over Gemini 3.5 Pro's missed schedule, a discounted Argon filling clusters, and Google Cloud reserving compute for DeepMind's own models. One question decides whether the AI build is a Cloud business or a research bill.

    Data Center

    Nvidia guarantees up to $105 billion of OpenAI's Ohio campus; CoreWeave's bonds imply near 63% default odds

    Nvidia's own filings show $36 billion of neocloud backstops, a $105 billion guarantee on OpenAI's Ohio campus and up to $125 billion of residual value promises, while default insurance on its biggest neocloud prices a 63 percent chance of failure. The credit market is pricing the case where the guarantor pays.

    Get the newsletter

    The day's AI and market news, in your inbox every weekday.