Lazy Digest #6: Nvidia Bid More for Decart. It Lost Anyway.

Tue, 25 Aug 2026 09:00:00 +0100

Lazy Digest #6: Nvidia Bid More for Decart. It Lost Anyway.

Nvidia offered more money for Decart than Anthropic did. Decart’s shareholders picked Anthropic anyway.

We flagged both offers as a live rumor in issue #5 — a Monday report of $5B talks with Nvidia, a Thursday report of $6B talks with Anthropic, nobody confirming anything. This week the picture sharpened considerably, even though the deal itself is still unsigned. We now know roughly how the price settled, why the founders preferred the lower bid, and how much richer three of them are about to become.

Everywhere else, the week was close to empty. One funding round closed. A cyber-insurance unicorn sold for less than half its old valuation, and its CEO went out of his way, on the record, to explain why that’s a good outcome. And two separate Globes investigations, four days apart, made the same point from two different directions: Israel’s AI sovereignty story is not quite what the government press conferences make it sound like.

Lazy Digest #6 — Israeli tech, one week, researched with AI and filtered by a human who reads the primary sources. Estimates are labeled as estimates. Let’s go.

The Bidding War Nvidia Keeps Losing

Calcalist reported Sunday that Decart halted its Nvidia discussions the moment Anthropic’s offer arrived — despite Nvidia’s bid being the higher one, unsurprising since Nvidia was already an investor and had every reason to want the company close. Both the founders and Sequoia Capital, Decart’s largest shareholder, preferred Anthropic. Google and Elon Musk’s SpaceX are both named as parties who could still reenter the process, but the deal is now down to exchanged drafts, run through boutique bank Catalyst, with people close to it expecting a signature “as soon as next month” — timed to land before Anthropic’s own IPO, expected in September.

On the number itself, the outlets disagree by about a billion dollars, and it’s worth being precise about why. CTech’s own reporting puts the deal at roughly $7 billion. Globes, Bloomberg and the Jerusalem Post, working from sources close to the negotiation two days later, describe Anthropic’s actual cap as $6 billion — mostly in Anthropic stock, with only a few hundred million in cash — against a $7–8 billion, cash-heavier offer from Nvidia that got rejected. Both figures are circulating in the market at once. Take it as a sign the price itself is still moving, not that anyone printed a wrong number.

What’s genuinely new this week is the shape of who gets what. Founders Dean Leitersdorf, his brother Orian (chief scientist, who joined the company last year) and Moshe Shalev together retain roughly 64% of Decart — about $4 billion on paper, or somewhere around $1–1.5 billion each. That’s a little below what each Wiz founder is believed to have taken home (just under $2 billion), and the gap is the point: Wiz’s founders got cash from Google. Decart’s founders chose a smaller headline number, in Anthropic stock, betting the equity appreciates faster than the cash would have. If Anthropic prices its IPO anywhere near the $2 trillion figure now being discussed — on revenue Fortune says could reach $100–120 billion this year — that bet looks obviously correct. If the AI-IPO window closes between signing and listing, three Israeli founders are sitting on illiquid paper they may already owe Israeli capital-gains tax on, since receiving shares in lieu of cash is a taxable event even when the asset can’t be sold. One estimate, based on a roughly NIS 12 billion combined exercise value, puts the eventual state tax take at around NIS 4.2 billion.

The rest of the cap table is almost entirely foreign — Benchmark, Sequoia, Radical Ventures and Oren Zeev’s Zeev Ventures would together collect more than $2 billion, and the only Israeli fund with a real stake is Michael Eisenberg’s Aleph, and even that’s small. Same shape as Wiz: the country gets the founders’ tax bill and an R&D center; the fund returns mostly leave.

And the R&D center is the actual prize here, if this closes. Anthropic currently has no development operation in Israel — its local presence today is a handful of salespeople based, oddly, in Ireland. Decart would become the company’s second R&D site outside the US, after a 15,000-square-meter, 200-person London office. Decart’s own headcount is small — 89 in Israel, 17 in the US — but the work sits at the center of Anthropic’s cost structure: making its models run faster and cheaper across Nvidia GPUs, Google’s TPUs and Amazon’s Inferentia chips, at the exact moment the entire AI industry is shifting its spending from training models to running them, which is where compute efficiency becomes the whole economic story.

It’s worth sitting with what Decart doesn’t have, because CTech’s own reporting doesn’t dress it up: after three years the company had no single defined mass-market product and no demonstrated recurring-revenue engine — it earned money project by project. The roughly $7 billion number is about 50% above the $4 billion valuation Decart carried only months ago, off a May round led by Radical Ventures with Nvidia among the participants. That’s a genuinely strong return in a short window. It is not the outcome you’d expect for a company its founder, Leitersdorf, told Calcalist he wanted to turn into “the next Google or Apple of AI.”

Put those two facts together and the deal reads less like a product acquisition and more like exactly what issue #5 guessed it was: a talent and infrastructure purchase wearing a product acquisition’s price tag. Anthropic isn’t buying Decart’s revenue. It’s buying roughly 100 engineers who are unusually good at making AI models cheaper to run — and Anthropic’s own numbers explain exactly why that’s worth paying for. Bloomberg reported this week that Anthropic’s Q2 revenue jumped 14-fold year over year to $11.5 billion, and that the company posted positive EBITDA for the first time. Annualized revenue is now estimated around $47 billion, ahead of OpenAI’s roughly $40 billion.

One more footnote for the record: Nvidia has now lost this exact contest twice inside a month — first to general uncertainty when the sale process opened, and now specifically to Anthropic, on price and buyer preference, despite bidding higher both times. Nvidia has been Israel’s most important strategic acquirer for a decade, going back to Mellanox (more on that below). Being outbid downward, on a company it already owned a piece of, by a lab that doesn’t yet have an Israeli office, is a new kind of result. Worth watching whether it happens again.

Forty-Three Cents on the Dollar

While Decart’s price argues about whether $6 billion or $7 billion is more accurate, this week’s only signed deal argued about something else entirely: whether a 57% valuation cut is actually good news. Its CEO says yes, and made the case himself, on the record, the day after signing.

Munich Re agreed Wednesday to buy At-Bay, the Israeli-founded cyber insurer, at a $575 million enterprise value, expected to close in Q1 2027. At-Bay combines cyber insurance with continuous security monitoring — it doesn’t just pay out after a breach, it watches a policyholder’s exposure for the life of the policy and prices accordingly. It’s one of the ten largest cyber insurers in the US, with $278 million in gross written premium, more than 35,000 insured businesses (mostly SMEs), and about 280 employees split between the US and Israel. Founded 2016 by Rotem Iram, Roman Itskovich, Etai Hochman and Tilly Kalisky; backers include Khosla Ventures, Lightspeed, Qumra, Microsoft’s M12, Glilot and Shlomo Kramer. Munich Re itself has been both an investor and At-Bay’s strategic partner since the early days, and now becomes the buyer, folding the company into HSB, its cyber-focused specialty arm.

The number everyone will lead with is the one CEO Rotem Iram addressed head-on: At-Bay was valued at $1.35 billion in its 2021 round. $575 million is 43 cents on that dollar.

Iram’s response is the cleanest explanation of the 2021-to-2026 reset that’s come out of any Israeli exit this year: “The disappointment is for people who didn’t understand that the world has changed. The value should be close to the cash the company creates.” His numbers: At-Bay wrote $50 million in premiums in 2021. Today it’s close to six times that, plus more than $30 million in annual cybersecurity software revenue that didn’t exist before. “In 2021 we received a high market value in a zero-interest-rate environment, with high multiples… The market today is realistic. It has returned to the fundamentals. In fact, selling for $575 million in 2026 is a very significant event and a success.”

He also said something companies rarely say out loud in an exit announcement: all the investors made money in the deal, and all the employees made money — no one is leaving, the founders are staying too. At-Bay has raised about $300 million total; a $575 million sale with founders and staff staying on, and the CEO calling it a win for everyone, is not the language of a distressed sale. It’s the language of a company that priced itself honestly against a market that stopped pretending 2021 multiples were real, and survived the correction that killed most of its cohort. “Such a violent change kills 99% of companies,” Iram said. “We multiplied ourselves sixfold. That’s why we’re a huge success.”

This is also Munich Re’s second Israeli insurtech purchase in eighteen months — its ERGO subsidiary bought the rest of Next Insurance for roughly $2.6 billion in March 2025. If you’re building in cyber-insurtech, there is now exactly one obvious strategic buyer in the category, and it just told you what it pays.

Two smaller signed deals rounded out the week, both in defense, both structured in a way that tells you something about who thinks their own stock is expensive.

Ondas — the Nasdaq-listed American defense-tech company that spent issue #5 winning the Israeli Defense Ministry’s “Digital Bat” drone tender — agreed to buy the manufacturing operations of Tel Aviv-listed Aran Defense for NIS 100 million ($33.9M), its first-ever purchase of a TASE-listed company’s business. Aran’s own market cap is only about NIS 200 million, so it’s selling roughly half its market value in operations that made NIS 54 million in revenue last year. The whole thing is paid in Ondas stock, priced on a trailing average, with 10% escrowed for 18 months. Ondas shares are up about 145% over the past year; buying Israeli defense capability with appreciated paper, while the seller is happy to take that paper at half its own market cap, is the kind of trade that only makes sense if both sides think Ondas stock keeps climbing. Worth noting too: Ondas’ new chairman, former Mossad chief David Barnea, was appointed about two weeks before this deal — the second ex-security-chief appointment doing commercial work in this digest alone (see SoftBank, further down).

DoiT, the Israeli-American cloud-cost unicorn, bought Attribute — a two-year-old startup that tracks AI and cloud spending down to the individual agent and customer — for an estimated $65 million. Attribute had raised about $13.5 million, so this is roughly a 5x return on seed capital in under three years, and DoiT’s fifth acquisition in two years as it builds out a FinOps stack aimed at the fact that AI inference cost is now a line item large enough to need its own accounting layer. Keep that thought — it’s the same logic behind Base44 building its own model, three sections down.

One Company Got Funded

CTech’s running 2026 funding tracker lists exactly one new-company venture round between the 16th and the 21st.

Xpander — $7.5 million seed, led by Pico Venture Partners with Emerge Ventures, Samsung Next and SeedIL. Founded 2024 by three former AWS principal engineers — David Twizer (CEO), Moriel Pahima (CTO) and Ran Sheinberg (CPO) — building infrastructure that lets companies run AI agents across different models and clouds without rebuilding their stack every time they switch providers. Alongside the round they launched Omni, an “agentic Forward Deployed Engineer” claiming a 90.9% score on the GAIA benchmark. Samsung Next’s presence is the more interesting signal than the round size: a large strategic corporate betting on multi-vendor AI infrastructure rather than locking into one lab’s stack.

That’s the entire disclosed funding week. The only other capital event was a $7 million strategic investment from Nasdaq-listed UroGen into Israeli IntraGel Therapeutics, which makes an injectable, biodegradable gel for delivering cancer treatment — not a venture round in the normal sense, but the only other money that moved.

For comparison: issue #4 counted about $315 million across a full week. Issue #5 called three rounds and $116 million the thinnest week the digest had seen. This week is thinner than that by an order of magnitude. No Israeli VC announced a new fund either — worth flagging specifically because issue #5 had Team8 closing $365 million the very same week a year ago. There’s no equivalent this time.

Against $575 million of signed exits and a $6 billion deal in its final stretch, one $7.5 million seed round isn’t a rounding error, it’s smaller than one. Mid-to-late August is thin everywhere and this shouldn’t be read as a trend on its own. But it’s now the second consecutive week this digest has had to say that, and at some point “August is slow” stops being a full explanation.

Two Reports That Undercut the Sovereignty Story

Issue #4 covered Project Nexus — the government’s tender for a domestically built quantum computer — as a real sovereignty statement, and I wrote that the instinct behind it was worth NIS 5 billion even if the machine underdelivered. Issue #5 noted the National AI Directorate had published its full plan in English, 28 pages, and called it “the follow-through.” This week, Globes’ Assaf Gilead published two separate investigations, four days apart, that complicate both of those calls considerably. Read together they’re the most important thing in this issue that isn’t a dollar figure.

The first, on the Directorate itself (18 August): behind the press conferences, the plan is thinner than advertised. There’s no itemized budget. No decision on whether Israel’s supercomputer gets built domestically or abroad — current thinking leans abroad, for speed, which is its own small irony for a sovereignty project. No decision on an Israeli language model. And nothing addressing the Electricity Authority’s freeze on new data-center grid connections, which has already stalled dozens of planned server farms and pushed at least one major prospective entrant, CoreWeave, to walk away from the Israeli market entirely.

The arithmetic is the clearest tell. The Directorate wants up to 100,000 Nvidia processors over the coming years — at roughly $135,000 per Vera Rubin chip, that’s $13.5 billion, which nobody in government is pretending is realistic. The actual working number under discussion is 3,000–6,000 additional chips for government and academic use, on top of the roughly 4,000 GPUs Israel’s sole civilian supercomputer currently leases from Nebius. The Directorate’s own 2026 budget is NIS 3 billion; it expected an increase to NIS 3.6 billion and got NIS 1 billion instead — because, and this connects straight to the defense section below, the state just found NIS 40 billion for ammunition production lines and had less appetite for AI infrastructure the same month.

There’s also a straightforward turf war: the Directorate published its own quantum tender with no stated budget, while the Innovation Authority already runs a NIS 1.1 billion national quantum program with its own 2029 target and its own funding request. One unnamed source called it a case of the right hand not knowing what the left is doing, and said it’s “embarrassing potential foreign partners.” This is also the fourth Israeli national AI plan announced in seven years — the first three, going back to 2019, were each shelved or radically cut before delivering what they promised.

The second investigation, two days later (20 August): Israel’s own Central Bureau of Statistics now publishes a separate GDP figure that excludes what it calls “exports that did not cross the country’s borders” — economic activity credited to Israeli-registered companies but actually manufactured overseas. The category is dominated by one company: Nvidia’s Mellanox division, acquired in 2019, whose Israeli-attributed quarterly revenue went from around $3 billion through 2024 to nearly $15 billion in the first quarter of this year, even though the chips are actually fabricated at TSMC in Taiwan.

Strip that out and the headline 15.4% Q2 GDP growth figure becomes 14.4% — and on an annual comparison basis, one economist’s calculation puts it at 7.2% with Nvidia against 5.5% without. The sharper number: without Nvidia and similar arrangements, Israel’s economy would have contracted 5.8% in the first quarter of this year. Full-year 2025 growth of 3.5% becomes 2.1% on the same adjustment. One analyst calculates that Israel’s real debt-to-GDP ratio, officially around 70%, is closer to 74% once the same accounting quirk is stripped out of the denominator.

None of this is a criticism of Nvidia, which pays real Israeli tax — $1.28 billion in 2025, by its own SEC filing — and employs 6,000 people here. It’s a statement about what the headline number actually measures. One economist’s framing: this is “production falling from the sky” — real money, real tax revenue, no relationship to the productivity of the Israeli workforce. The comparison Globes reaches for is Intel, which built Israel’s semiconductor industry, then cut a quarter of its 12,000 local jobs and froze two new plants when its own numbers stopped working. Mellanox is a smaller, newer version of the same dependency, and it is currently doing more to move Israel’s GDP than the entire rest of the tech sector combined.

Put both investigations next to Decart, above, and you get a coherent picture rather than three unrelated stories. The government’s own sovereign-AI plan is underfunded and uncoordinated. The country’s headline growth number is substantially one foreign company’s accounting choice. And the most credible independent Israeli AI company just agreed to be absorbed into an American lab. None of that means Israeli AI talent isn’t real — it obviously is, and the Decart price is itself evidence of that. It means the state-level story — sovereignty, independence, a domestic model — is currently more aspiration than fact, and this week two of Israel’s own institutions said so on the record.

Defense: Buying More, Selling for Less

Two defense stories this week point in opposite directions, and the gap between them is the story.

UVision, the loitering-munitions maker owned by Aaron Frankel, lost its CFO mid-roadshow. Ronny Barak departed under circumstances the company hasn’t explained; Eyal Rubin, CFO of Nasdaq-listed Nasus Pharma, is expected to replace him starting in September. UVision is trying to IPO on Nasdaq at a $3.5 billion pre-money valuation. Israeli institutional investors offered $2.5–2.9 billion for a pre-IPO stake and got turned down. Since then, the broader defense-stock selloff has made even that lower range look optimistic — Next Vision, a comparable Israeli defense-tech name, has fallen about 40% on the Tel Aviv exchange from its March peak. UVision did also land a genuinely good piece of news the same week: a $50 million-plus follow-on US Army order for its HERO 120 munition, alongside American partner Mistral. Strong demand, weak public-market appetite for the stock that demand is supposed to justify — that’s the whole tension of Israeli defense-tech right now, in one company.

The reason for the second half of that tension surfaced a few days later, and it’s worth including even though the report landed right at the edge of this issue’s window, because it describes decisions agreed within it: Israel is adding roughly NIS 40 billion ($13.3 billion) to its 2026 defense budget, pushing total defense spending to NIS 184 billion ($61.3 billion) — the second-highest ever. The trigger is blunt: several missile and artillery production lines were within weeks of shutting down for lack of financing, and the Defense Ministry owes its three largest contractors a combined NIS 15.5 billion ($5.2 billion) — NIS 5.5bn to IAI, NIS 7bn to Rafael, NIS 3bn to Elbit. IAI’s own Q2 cash flow went negative by roughly $795 million even as revenue rose 35% and its order backlog hit a record $35 billion. One senior defense official’s quote to Calcalist is the line of the week on the subject: “This is not how Sparta operates. This is how Bardak operates” — Israeli slang for chaos.

There’s a direct line from this back to the last section. The Directorate’s AI budget got cut from a promised NIS 3.6 billion to NIS 1 billion the same month the state found NIS 40 billion for artillery shells. That’s not really a mystery, and it’s not really a scandal either — it’s what happens to civilian technology budgets in a country that’s still, structurally, at war. It does mean anyone modeling government AI spending as a durable tailwind should model it against defense’s claim on the same shekel, not in isolation.

Ondas, covered above for buying Aran Defense in stock, is the private-market mirror of all this: still buying Israeli defense assets aggressively, still paying in appreciated paper, unbothered by the public-market pricing currently squeezing UVision. Two very different capital structures reading the same sector two very different ways.

The Bright Spot

Two genuinely good operating numbers this week, both about the same underlying move: owning your own AI model instead of renting one.

Base44, the “vibe coding” tool Maor Shlomo sold to Wix within months of building it, hit $200 million in annual recurring revenue — doubling from $100 million in just five months, and adding roughly $50 million since May alone. The company just shipped Base1, its own dedicated large language model, specifically to cut the inference and compute costs that come with running someone else’s. Wix says Base44’s gross margin should improve to around 60% in the second half of the year, up from near break-even, and it’s planning to reinvest most of those savings into marketing rather than take them as profit. Wix also disclosed it will pay Shlomo another $41 million on milestone achievement, pushing his total proceeds from the sale above $150 million. It’s the same logic behind DoiT’s purchase of Attribute, above: AI inference is now expensive enough, and unpredictable enough, that owning the model and metering your own costs is worth the engineering effort.

LTX, the video-AI company spinning out of Lightricks as part of its planned split into two businesses, launched an open, hardware-agnostic “world model” — pitched not as a filmmaking tool but as training infrastructure for robots, real-time avatars and physical-AI simulation. CEO Zeev Farbman told Globes the company spends about $50 million a year just training the model, which is the part he says most startups claiming to “build a world model” simply can’t afford; LTX’s plan is to give that expensive part away free below $10 million in customer revenue and license it above that threshold, priced differently by industry. It’s explicitly positioned against Nvidia’s Cosmos platform, on the argument that hardware neutrality — running on Apple, AMD and Nvidia silicon alike — is worth more to robotics builders than being tied to one chipmaker. LTX has about 250 people, mostly R&D, and is still hiring even as the Lightricks split cuts 75 roles on the mobile-app side.

Two Israeli companies, two different bets that the same problem — AI compute is expensive and someone else’s pricing is a liability — is worth solving yourself rather than renting the solution. Both are right often enough to notice.

Everything Else Worth Your Time

Pentera cut another 60 people, 30 of them in Israel, four months after cutting about 40 in April — roughly 100 layoffs total, or 20% of its 470-person workforce, even as the company keeps hiring for AI-specific roles. Same pattern this digest has flagged before: legacy headcount shrinking, AI headcount growing, inside the same company, in the same quarter. It’s the only standalone Israeli tech layoff of the week.

Yossi Cohen, the former Mossad chief, got promoted at SoftBank — from head of Israeli operations to strategic advisor to Masayoshi Son himself, a meaningfully more senior seat. His remit is also shifting, from the cybersecurity deals that made his name at SoftBank (Wiz, Claroty, Cato Networks) toward data centers, AI infrastructure and chip investing — which happens to be exactly the category Israel’s electricity freeze is currently strangling. Worth filing next to David Barnea’s new chairmanship at Ondas: two former heads of Israeli intelligence agencies now doing dealmaking for foreign capital, in the same week.

Police arrested a man in Ashkelon on suspicion of planting malware across dozens of Israeli companies and quietly harvesting sensitive files, apparently acting alone with no prior record and no evident extortion motive. It became the most-read story on CTech all week, and the uncomfortable subtext is obvious for a country that exports cyber defense as a national industry: dozens of companies were compromised by one unaffiliated individual for an extended period before anyone connected the dots.

Israel’s crypto industry restarted its push for regulatory reform. Nir Hirschmann-Rub, head of the Crypto Companies Forum, told CTech that despite recent moves by the Bank of Israel and the Capital Markets Authority, “Israel is significantly behind schedule” against the US, which has connected licensed exchanges to the federal clearing system and let crypto firms hold government bonds. His pitch is framed explicitly against bank profits — Israeli banks made NIS 32 billion in 2025, about NIS 10 billion of it from fees — rather than as a speculative-asset argument. It’s a lobbying position, not a neutral fact, but it’s a sharper pitch than “crypto is the future.”

One counterpoint to the sovereignty pessimism above, for balance. Hetz Ventures’ Chen Zvi published an opinion piece the same week as the Directorate investigation, arguing the real Israeli AI story isn’t sovereignty at the state level, it’s density at the company level: the number of Israeli AI companies has nearly doubled since 2021 to roughly 2,350, they’ve raised more than $30 billion between them, and Israel has the world’s highest share of LinkedIn users listing AI expertise, at about 2.1%, per the Stanford AI Index. His framing — Israel isn’t missing the AI wave, it just hasn’t cashed in yet on the infrastructure layer — and the state-level reporting above are both true at once, and the gap between them is basically the whole argument of this issue.

Not new this week, still open: Wonderful’s reported $500 million raise, first flagged in issue #5, remains unconfirmed. No update.

What I’d Do With This Week

If you’re negotiating an acquisition: Decart’s founders chose a lower headline number, in stock, over a higher one in cash, because they believed the buyer’s equity would outperform. At-Bay’s founders took a 57% cut on their old mark and its CEO is out there calling it a success on the record. Both are the same lesson from opposite directions — the number that matters is what you actually collect, not the number in the press release, and both sets of founders understood that before signing.

If you’re building anything that touches inference cost: Base44 built its own model. LTX gives its model away below $10M revenue specifically to build a moat before charging for it. DoiT bought a company that just measures AI spending. Three different companies solved variations of the same problem this week — someone else’s token pricing is now a liability large enough to engineer around — and none of them are foundation-model labs.

If you’re modeling Israeli government AI spending as a tailwind: be precise about what it’s competing against. The Directorate’s budget increase got cut by NIS 2.6 billion the same month the state found NIS 40 billion for artillery production lines. That’s not a one-off. As long as Israel is funding an active war posture, civilian technology budgets are the thing that gets trimmed first, and this week is the clearest evidence of that trade yet.

If you’re reading Israeli GDP or employment data: use the ex-Nvidia numbers, not the headline ones. A 15.4% quarter is 14.4% once Mellanox’s Taiwan-manufactured chips are stripped out, and Q1 was a contraction, not growth, on the same adjustment. This is now a standing caveat for this digest, not a one-time correction.

If you’re in defense-tech and drafting a listing document: UVision’s experience says the private-market comps you’re using are running ahead of what public investors will actually pay right now. Build your range around what Israeli institutions offered UVision, not what the company originally asked for.

If you’re watching the ecosystem number: one funded company, $7.5 million, against $575 million of signed exits and a $6 billion deal in its final stretch. That’s the sharpest funding-to-exit gap this digest has recorded, worse than issue #5’s three-rounds-against-a-rumor. Two thin weeks back to back is no longer purely an August story.

See you next week. Same lazy hands, same working head.

With 💜, Alex Sherman

Research assisted by AI, every claim read and filtered by a human. Primary sources: CTech / Calcalist — including its running funding, M&A and layoffs trackers — Globes, the Jerusalem Post, Bloomberg, and company disclosures. Not confirmed: the Decart–Anthropic deal, on which no agreement has been signed and no party has commented on the record; the price itself is reported at both ~$6B (Globes, Bloomberg, JPost) and ~$7B (CTech), and is treated here as unresolved. The Decart tax estimate (~NIS 4.2B) is a third-party calculation, not a disclosed figure. At-Bay’s founder quotes are drawn from Calcalist’s interview with CEO Rotem Iram. Aran Defense’s transaction terms are from the company’s own TASE filing; DoiT–Attribute’s price is a Calcalist estimate. Base44 and LTX’s revenue and cost figures are company-provided and unaudited. The NIS 40 billion defense budget increase was reported just after this issue’s window closed but describes decisions agreed within it, and is included on that basis. Wonderful’s $500M raise, first reported in issue #5, remains unconfirmed and was not re-reported here. Disclosure: this digest is researched using AI tools including Anthropic’s, and Anthropic is a reported party to the week’s largest story, for the second issue running. Previous issues: #1, #2, #3, #4, #5.

National AI Directorate on Sherman - Israeli Startups

Lazy Digest #6: Nvidia Bid More for Decart. It Lost Anyway.

Lazy Digest #6: Nvidia Bid More for Decart. It Lost Anyway.

The Bidding War Nvidia Keeps Losing

Forty-Three Cents on the Dollar

One Company Got Funded

Two Reports That Undercut the Sovereignty Story

Defense: Buying More, Selling for Less

The Bright Spot

Everything Else Worth Your Time

What I’d Do With This Week